From ca002318ae07e67ab3a9cd055f1b35672f3efb8e Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:00:05 -0400 Subject: [PATCH 01/42] Add one shared chronicle-first env helper Chronicle's operational stores migrate by dual-run (chronicle#143, mechanism 3). Every env read was ledger-first with no chronicle-named alternative, and each of the three modules that read configuration had grown its own helper. Add chronicle/env.py: env_value/env_flag expand a name into CHRONICLE_, LEDGER_, POLICYENGINE_LEDGER_ and return the first set value, warning once per process with ChronicleEnvDeprecationWarning when a ledger-era name supplied it. Names outside those three prefixes, such as POLICYENGINE_SUPABASE_URL, are read literally so the helper renames the ledger-era surface only. The warning subclasses FutureWarning, not DeprecationWarning, so operators running the CLI actually see it. The Supabase schema name stays "ledger"; only the env var that overrides it moves to CHRONICLE_SCHEMA. Renaming the schema is a later slice. Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 74 ++++++++++------------ chronicle/__init__.py | 1 + chronicle/env.py | 120 ++++++++++++++++++++++++++++++++++++ chronicle/source_package.py | 25 ++------ db/pe_source_inventory.py | 21 +++---- db/supabase_client.py | 23 +++---- 6 files changed, 174 insertions(+), 90 deletions(-) create mode 100644 chronicle/env.py diff --git a/PROGRESS.md b/PROGRESS.md index 24773009..e46b096c 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -1,51 +1,41 @@ -# Lane C5 progress +# Operational rename, slice 1 (chronicle#143, mechanism 3) + +Lane C5's handoff notes previously lived here; its durable record is +`LANE_C5_REPORT.md`. This file now tracks the active lane on this branch. ## State -- Branch: `be-2025-vintages` from `origin/main` at `5c15bfd`. -- Worktree inputs are staged under `.lane-raw/` and must remain uncommitted. -- Lane C5 is complete, validated, independently reviewed, and ready for handoff. -- The requested staged C2 report is absent, but root `LANE_C2_REPORT.md` is byte-identical - to the sibling lane's staged copy (SHA-256 `4590e0dc...50f06e7`) and is the pattern used. +- Branch: `ops-rename-slice1`, cut from `origin/main` at `ff3efd3`. +- Scope: env names, R2 bucket configurability, `ledger.db` -> `chronicle.db`, + and the docs for all three. Code and docs only; no infrastructure changes. +- Out of scope and deliberately untouched: the `ledger` console-script alias, + the Supabase `"ledger"` schema and mirror table names, governance role ids and + concept authorities, hash domains and schema ids, anything under `releases/`. ## Done -- Read the repository Chronicle boundary rules in `AGENTS.md`. -- Read `.lane-raw/SOURCES.md` and confirmed all five named publisher artifacts are present. -- Confirmed the worktree is otherwise clean apart from `.lane-raw/` and the shared `.venv` link. -- Verified all five staged artifact SHA-256 pins exactly. -- Mapped FPB workbook cells: 990 facts across T01/T06/T07/T11/T17/T24, with - 2022–2025 observations and 2026–2031 `source_projection` facts. -- Confirmed PDF boundary evidence: printed page 19 calls 2026 the first projection year; - annex table units appear on printed pages 45, 48, 49, 53, 58, and 65. -- Chosen Eurostat layout: two vintage-specific source-package aliases share new manifest - entries, preserving the prior package YAMLs, raw bytes, and fact outputs unchanged. -- Reproduced the Statbel curator logic: 18 NUTS1 × sex × age-band cells totaling 11,825,551. -- Added the hash-pinned FPB workbook and publication PDF plus the - `fpb-economic-outlook-2026-2031-june-2026` package alias. -- Built 990 line-specific publisher facts (99 per year): 396 observations for - 2022–2025 and 594 `source_projection` facts for 2026–2031. -- Passed FPB `validate-package` and `build-suite`: 990 facts, full cell lineage, - zero acceptance errors, and pinned 2025 cells 320578 / 77771 / 5602 million euro. -- Re-ran the Statbel 2026 curator logic on the 2025 ZIP and added the hash-pinned - raw capture plus its deterministic 18-row curated CSV. -- Passed Statbel 2025 `validate-package` and `build-suite`: 18 facts totaling - 11,825,551, 66 constraints, full lineage, and zero acceptance errors. -- Added the Eurostat `gov_10a_taxag` 2025 and `spr_exp_func` 2024 manifest - entries plus vintage-specific package aliases, without modifying either - prior artifact or prior package specification. -- Passed both new Eurostat package validations and suite builds: 12 tax facts - and 9 ESSPROS facts, full lineage, and zero acceptance errors. -- Extended Belgium and Eurostat regressions for FPB table counts/cells and - assertion boundary, vintage non-overlap, prior-output digests, Statbel pins, - and the declared 0.25% Statbel/FPB population comparison tolerance. -- Passed 43 focused tests and the full merged-bundle regression: 157,177 facts, - 148 packages, zero aggregate-key duplicates, and expected goldens throughout. -- Recorded pins, counts, boundary evidence, curator commands, validation tails, - and consumer fact families in `LANE_C5_REPORT.md`. -- Passed independent `ledger-source-fidelity` and `ledger-boundary` reviews with - no required corrections. +- Read `AGENTS.md`, `docs/storage-architecture.md`, + `docs/agent-source-package-harness.md`, and the mechanism-3 migration spec in + the first comment of PolicyEngine/chronicle#143. +- Enumerated every ledger-named env read in tracked Python: the four real + variables (`LEDGER_SOURCE_ARTIFACT_CACHE_DIR`, `LEDGER_SOURCE_ARTIFACT_FETCH`, + `LEDGER_PE_US_DATA_ROOT`, `LEDGER_PE_UK_DATA_ROOT`) plus + `POLICYENGINE_LEDGER_SCHEMA`. `LEDGER_MIRROR_TABLES`, + `LEDGER_MIRROR_PRIMARY_KEYS`, and `LEDGER_DB_SCHEMA_VERSION` are module + constants, not env reads, and name out-of-scope surfaces. +- Added `chronicle/env.py`: one shared `env_value`/`env_flag`/`env_names` + helper reading `CHRONICLE_` first, then `LEDGER_` and + `POLICYENGINE_LEDGER_` with a once-per-process + `ChronicleEnvDeprecationWarning` naming the preferred variable. +- Replaced all three ad-hoc helpers (`db/supabase_client._env`, + `chronicle/source_package._env_value`/`_truthy_env`, + `db/pe_source_inventory._env_value`) with the shared helper. ## Next -- None; ready for handoff. No push was performed. +- Make R2 bucket names configurable with unchanged `ledger-*` defaults. +- Emit `chronicle.db` for new suite outputs; keep reading `ledger.db`. +- Fix the backwards fallback statement in the docs and sweep every env-name, + bucket-name, and db-filename mention in README/AGENTS/docs. +- Add the hermetic dual-read tests; run pytest, ruff check, ruff format --check. +- Push and open the PR. Do not merge. diff --git a/chronicle/__init__.py b/chronicle/__init__.py index 04d50943..e3ddfb06 100644 --- a/chronicle/__init__.py +++ b/chronicle/__init__.py @@ -13,6 +13,7 @@ "consumer_contract", "core", "database", + "env", "facts", "harness", "jurisdictions", diff --git a/chronicle/env.py b/chronicle/env.py new file mode 100644 index 00000000..80667f4d --- /dev/null +++ b/chronicle/env.py @@ -0,0 +1,120 @@ +"""Environment configuration for the Chronicle rename window. + +Chronicle's operational stores migrate by dual-run (PolicyEngine/chronicle#143, +mechanism 3): every configuration variable gets a ``CHRONICLE_``-prefixed name +that is read first, while the ledger-era ``LEDGER_`` and +``POLICYENGINE_LEDGER_`` names keep working behind a deprecation warning. That +window lets downstream publish flows migrate on their own schedule instead of +breaking the moment Chronicle ships a rename. + +Names that carry none of those three prefixes are read literally: this helper +renames the ledger-era surface, not every PolicyEngine variable. +""" + +from __future__ import annotations + +import os +from typing import TypeVar +import warnings + +__all__ = [ + "CHRONICLE_ENV_PREFIX", + "ChronicleEnvDeprecationWarning", + "LEGACY_ENV_PREFIXES", + "env_flag", + "env_names", + "env_value", + "reset_env_deprecation_state", +] + +CHRONICLE_ENV_PREFIX = "CHRONICLE_" + +# Ordered most specific first so prefix stripping is unambiguous. +LEGACY_ENV_PREFIXES = ("POLICYENGINE_LEDGER_", "LEDGER_") + +TRUTHY_ENV_VALUES = frozenset({"1", "true", "yes", "on"}) + + +class ChronicleEnvDeprecationWarning(FutureWarning): + """A ledger-era environment variable supplied a Chronicle setting. + + Subclasses :class:`FutureWarning` rather than :class:`DeprecationWarning` + so the notice reaches operators running the CLI, who are the people who + have to move the variable. ``DeprecationWarning`` is silenced by default + outside ``__main__``. + """ + + +_Default = TypeVar("_Default") + +_WARNED_LEGACY_NAMES: set[str] = set() + + +def _env_suffix(name: str) -> str | None: + """Return the rename-window suffix of ``name``, or None if it has none.""" + for prefix in (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES): + if name.startswith(prefix) and len(name) > len(prefix): + return name[len(prefix) :] + return None + + +def env_names(name: str) -> tuple[str, ...]: + """Return the lookup order for ``name``. + + The chronicle-preferred name comes first, then the ledger-era names that + remain accepted during the migration window. A name outside the rename + window is returned unchanged, as its own single-element lookup order. + """ + suffix = _env_suffix(name) + if suffix is None: + return (name,) + return ( + f"{CHRONICLE_ENV_PREFIX}{suffix}", + *(f"{prefix}{suffix}" for prefix in LEGACY_ENV_PREFIXES), + ) + + +def _warn_legacy(found: str, preferred: str) -> None: + """Warn once per process that a ledger-era variable supplied a value.""" + if found in _WARNED_LEGACY_NAMES: + return + _WARNED_LEGACY_NAMES.add(found) + warnings.warn( + f"{found} is a ledger-era Chronicle environment variable; " + f"set {preferred} instead. The old name is still honored during the " + "Chronicle rename window and will be removed once consumers migrate.", + ChronicleEnvDeprecationWarning, + stacklevel=3, + ) + + +def reset_env_deprecation_state() -> None: + """Forget which legacy names have already warned. Test-support hook.""" + _WARNED_LEGACY_NAMES.clear() + + +def env_value(*names: str, default: _Default = None) -> str | _Default: + """Read the first set value across ``names``, chronicle-preferred first. + + Each name is expanded through :func:`env_names`, so a caller can pass the + chronicle name and still pick up a value set under a ledger-era name. + Empty values are treated as unset, matching the helpers this replaces. + """ + for name in names: + candidates = env_names(name) + preferred = candidates[0] + for candidate in candidates: + value = os.environ.get(candidate) + if value: + if candidate != preferred: + _warn_legacy(candidate, preferred) + return value + return default + + +def env_flag(*names: str) -> bool: + """Return whether the first set value across ``names`` reads as true.""" + value = env_value(*names) + if value is None: + return False + return value.strip().lower() in TRUTHY_ENV_VALUES diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 2a6320ff..4ddc1807 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -3,7 +3,6 @@ from __future__ import annotations import hashlib -import os from dataclasses import dataclass, replace from importlib.resources import files from io import BytesIO @@ -32,6 +31,7 @@ AggregateFact, build_label, ) +from chronicle.env import env_flag, env_value from chronicle.epoch import SCHEMA_IDS, schema_id from chronicle.sources.cells import ( SourceArtifactMetadata, @@ -373,8 +373,8 @@ "usda_snap/fy2025_monthly_state_caseloads" ), } -SOURCE_ARTIFACT_CACHE_ENV = "LEDGER_SOURCE_ARTIFACT_CACHE_DIR" -SOURCE_ARTIFACT_FETCH_ENV = "LEDGER_SOURCE_ARTIFACT_FETCH" +SOURCE_ARTIFACT_CACHE_ENV = "CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR" +SOURCE_ARTIFACT_FETCH_ENV = "CHRONICLE_SOURCE_ARTIFACT_FETCH" DEFAULT_SOURCE_ARTIFACT_CACHE_DIR = ( Path.home() / ".cache" / "policyengine-chronicle" / "source-artifacts" ) @@ -2238,7 +2238,7 @@ def _read_source_artifact_content( if cache_path.exists(): return cache_path.read_bytes() - if not _truthy_env(SOURCE_ARTIFACT_FETCH_ENV): + if not env_flag(SOURCE_ARTIFACT_FETCH_ENV): raise FileNotFoundError( f"Source artifact {spec['filename']} is not packaged and was not " f"found in {cache_path}. Set {SOURCE_ARTIFACT_FETCH_ENV}=1 to fetch " @@ -2260,7 +2260,7 @@ def _read_source_artifact_content( def _source_artifact_cache_path(spec: dict[str, Any]) -> Path: cache_root = Path( - _env_value( + env_value( SOURCE_ARTIFACT_CACHE_ENV, default=DEFAULT_SOURCE_ARTIFACT_CACHE_DIR, ) @@ -2297,21 +2297,6 @@ def _validate_source_artifact_sha( ) -def _env_value(*names: str, default: str | Path) -> str | Path: - for name in names: - value = os.environ.get(name) - if value: - return value - return default - - -def _truthy_env(*names: str) -> bool: - return any( - os.environ.get(name, "").strip().lower() in {"1", "true", "yes", "on"} - for name in names - ) - - def _single_archive_member(archive: ZipFile, *, suffixes: tuple[str, ...]) -> str: members = [ name diff --git a/db/pe_source_inventory.py b/db/pe_source_inventory.py index a52fff23..f64ed870 100644 --- a/db/pe_source_inventory.py +++ b/db/pe_source_inventory.py @@ -2,14 +2,15 @@ from __future__ import annotations -import os from pathlib import Path +from chronicle.env import env_value + from .schema import Jurisdiction from .source_files import SourceArtifactSpec, make_slug, make_url_slug -PE_US_DATA_ROOT_ENV = "LEDGER_PE_US_DATA_ROOT" -PE_UK_DATA_ROOT_ENV = "LEDGER_PE_UK_DATA_ROOT" +PE_US_DATA_ROOT_ENV = "CHRONICLE_PE_US_DATA_ROOT" +PE_UK_DATA_ROOT_ENV = "CHRONICLE_PE_UK_DATA_ROOT" SOURCE_SUFFIXES = { ".csv", @@ -272,21 +273,13 @@ ] -def _env_value(*names: str) -> str | None: - for name in names: - value = os.environ.get(name) - if value: - return value - return None - - def _resolve_required_root( root: Path | None, *, flag: str, env_var: str, ) -> Path: - value = root if root is not None else _env_value(env_var) + value = root if root is not None else env_value(env_var) if value is None: raise ValueError(f"{flag} or {env_var} is required.") path = Path(value).expanduser() @@ -612,8 +605,8 @@ def pe_source_specs( ) -> list[SourceArtifactSpec]: """Return source files used by the PE-US and PE-UK calibration pipelines.""" specs: list[SourceArtifactSpec] = [] - us_configured = pe_us_root is not None or _env_value(PE_US_DATA_ROOT_ENV) - uk_configured = pe_uk_root is not None or _env_value(PE_UK_DATA_ROOT_ENV) + us_configured = pe_us_root is not None or env_value(PE_US_DATA_ROOT_ENV) + uk_configured = pe_uk_root is not None or env_value(PE_UK_DATA_ROOT_ENV) if include_us and (us_configured or not include_uk or not uk_configured): specs.extend(pe_us_source_specs(pe_us_root)) if include_uk and (uk_configured or not include_us or not us_configured): diff --git a/db/supabase_client.py b/db/supabase_client.py index e14bb5bb..50ce4f30 100644 --- a/db/supabase_client.py +++ b/db/supabase_client.py @@ -8,7 +8,6 @@ from __future__ import annotations -import os from dataclasses import dataclass from functools import lru_cache from typing import Any, Dict, List, Optional @@ -16,18 +15,14 @@ from supabase import create_client, Client +from chronicle.env import env_value -def _env(*names: str) -> str | None: - """Read PolicyEngine-owned storage config.""" - for name in names: - value = os.environ.get(name) - if value: - return value - return None - - -LEDGER_SCHEMA = _env("POLICYENGINE_LEDGER_SCHEMA") or "ledger" -TARGETS_SCHEMA = _env("POLICYENGINE_TARGETS_SCHEMA") or "targets" +# The hosted Postgres schema is still named "ledger"; only the environment +# variable that overrides it has moved to the chronicle prefix. Renaming the +# schema itself is a later slice of PolicyEngine/chronicle#143, coordinated +# with the CI writers that already target the ledger schema. +LEDGER_SCHEMA = env_value("CHRONICLE_SCHEMA") or "ledger" +TARGETS_SCHEMA = env_value("POLICYENGINE_TARGETS_SCHEMA") or "targets" @dataclass @@ -49,14 +44,14 @@ def from_env(cls) -> "SupabaseConfig": Raises: ValueError: If required environment variables are missing """ - url = _env("POLICYENGINE_SUPABASE_URL") + url = env_value("POLICYENGINE_SUPABASE_URL") if not url: raise ValueError( "POLICYENGINE_SUPABASE_URL not set. " "Set this to your Supabase project URL." ) - secret_key = _env( + secret_key = env_value( "POLICYENGINE_SUPABASE_SERVICE_KEY", "POLICYENGINE_SUPABASE_SECRET_KEY", ) From 03c5b5d86f5fe657b64cfced57b861b0a40acdc6 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:06:52 -0400 Subject: [PATCH 02/42] Make R2 buckets configurable and emit chronicle.db Bucket names: add CHRONICLE_R2_RAW_BUCKET / CHRONICLE_R2_DERIVED_BUCKET, resolved in the function body rather than bound as keyword defaults so the setting reaches long-lived processes. Defaults are unchanged at ledger-raw and ledger-derived; only the follow-up cutover PR flips them. Plumbed through fetch-artifact, publish-raw, publish-derived and bootstrap-r2, whose --r2-bucket/--raw-bucket/--derived-bucket now default to the resolver. Because the bucket can now vary, two manifest write paths could restate recorded storage.r2 blocks. Archived witness records pin raw R2 URLs by hash, so both now preserve history: publish-raw reports recorded_r2_bucket_is_preserved_history instead of uploading, and fetch-artifact keeps an already-recorded r2 block rather than overwriting it with a different bucket. The consumer-fact boundary guard matched the literal strings ledger-derived: and r2://ledger-derived/, so a renamed bucket would have made it silently stop firing. It now matches on shape (bucket ends in -derived, or key starts with derived/) and parses the r2:// URI, which also closes the uri-only hole where a fact carrying no bucket or key slipped past. Database artifact: new suite outputs write chronicle.db. infer_build_id reads chronicle.db then ledger.db (both branches were byte-identical, so the existing fallback was a no-op), and _derived_artifact_kind classifies both names (its set literal held one element twice). The sidecar resource list moves with the write site because _resource_descriptor stats every listed path. Fix db/cli.py, which imported the private env reader that the shared helper replaced. That ImportError broke every db CLI subcommand, including the three the CI job runs, while pytest stayed green. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 104 ++++++++++++++++++++++++++++----- chronicle/consumer_contract.py | 32 ++++++---- chronicle/database.py | 9 +++ chronicle/env.py | 38 ++++++++---- chronicle/harness.py | 53 +++++++++++------ chronicle/suite.py | 10 +++- db/cli.py | 5 +- tests/test_chronicle_mirror.py | 8 +-- tests/test_chronicle_suite.py | 10 ++-- 9 files changed, 204 insertions(+), 65 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index fa56d620..6d1e9781 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -18,14 +18,41 @@ import httpx import yaml +from chronicle.database import ( + CHRONICLE_DB_FILENAME, + CHRONICLE_DB_FILENAMES, + LEGACY_CHRONICLE_DB_FILENAME, +) +from chronicle.env import env_value from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain + +R2_RAW_BUCKET_ENV = "CHRONICLE_R2_RAW_BUCKET" +R2_DERIVED_BUCKET_ENV = "CHRONICLE_R2_DERIVED_BUCKET" + +# The bucket defaults stay at their ledger-era names. Archived witness records +# pin raw R2 URLs by hash, so ledger-raw and ledger-derived are preserved +# read-only forever and no recorded manifest URI is ever rewritten. The env +# vars exist so the cutover in docs/storage-architecture.md can be rehearsed, +# and so flipping to chronicle-raw/chronicle-derived is a default change rather +# than a code change (PolicyEngine/chronicle#143, mechanism 3). DEFAULT_R2_RAW_BUCKET = "ledger-raw" DEFAULT_R2_DERIVED_BUCKET = "ledger-derived" DEFAULT_R2_PREFIX = "raw" DEFAULT_R2_DERIVED_PREFIX = "derived" + +def default_r2_raw_bucket() -> str: + """Resolve the raw bucket: ``$CHRONICLE_R2_RAW_BUCKET`` or the default.""" + return env_value(R2_RAW_BUCKET_ENV, default=DEFAULT_R2_RAW_BUCKET) + + +def default_r2_derived_bucket() -> str: + """Resolve the derived bucket: ``$CHRONICLE_R2_DERIVED_BUCKET`` or default.""" + return env_value(R2_DERIVED_BUCKET_ENV, default=DEFAULT_R2_DERIVED_BUCKET) + + # New UK and New Zealand uploads are namespaced by country. US objects predate # the country segment and deliberately keep their legacy ``raw/{source_id}`` # and ``derived/{source_id}`` shapes. Publisher directories are the stable @@ -392,11 +419,12 @@ def fetch_source_artifact( table: str | None = None, filename: str | None = None, upload_r2: bool = False, - r2_bucket: str = DEFAULT_R2_RAW_BUCKET, + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> ArtifactFetchReport: """Fetch/register a source artifact and optionally upload it to R2.""" + r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, @@ -482,12 +510,13 @@ def publish_derived_artifacts( package_id: str, year: int, build_id: str | None = None, - r2_bucket: str = DEFAULT_R2_DERIVED_BUCKET, + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", build_artifacts_output: str | Path | None = None, ) -> DerivedArtifactPublishReport: """Upload a deterministic build output directory to the derived R2 bucket.""" + r2_bucket = r2_bucket or default_r2_derived_bucket() input_path = Path(input_dir) if not input_path.exists(): return DerivedArtifactPublishReport( @@ -617,11 +646,12 @@ def publish_source_artifacts( manifest_filename: str = "manifest.yaml", source_id: str | None = None, package_id: str | None = None, - r2_bucket: str = DEFAULT_R2_RAW_BUCKET, + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> RawArtifactPublishReport: """Upload manifest-declared raw source artifacts and record R2 locations.""" + r2_bucket = r2_bucket or default_r2_raw_bucket() root_path = Path(root) if not root_path.exists(): return RawArtifactPublishReport( @@ -791,12 +821,15 @@ def inventory_source_artifacts( def bootstrap_r2_buckets( *, - raw_bucket: str = DEFAULT_R2_RAW_BUCKET, - derived_bucket: str = DEFAULT_R2_DERIVED_BUCKET, + raw_bucket: str | None = None, + derived_bucket: str | None = None, wrangler_command: str = "npx wrangler", ) -> R2BootstrapReport: """Create the R2 buckets Chronicle expects, if Wrangler is authenticated.""" - buckets = (raw_bucket, derived_bucket) + buckets = ( + raw_bucket or default_r2_raw_bucket(), + derived_bucket or default_r2_derived_bucket(), + ) commands: list[ArtifactCommandResult] = [] errors: list[str] = [] @@ -1010,9 +1043,9 @@ def infer_build_id(input_dir: str | Path) -> str | None: if build_id: return str(build_id) - db_path = input_path / "ledger.db" + db_path = input_path / CHRONICLE_DB_FILENAME if not db_path.exists(): - db_path = input_path / "ledger.db" + db_path = input_path / LEGACY_CHRONICLE_DB_FILENAME if db_path.exists(): with sqlite3.connect(db_path) as connection: row = connection.execute( @@ -1043,6 +1076,17 @@ def _filename_from_url(source_url: str) -> str: return Path(unquote(parsed.path)).name +def _recorded_r2(spec: Any) -> dict[str, Any]: + """Return a manifest file spec's recorded ``storage.r2`` block, if any.""" + if not isinstance(spec, dict): + return {} + storage = spec.get("storage") + if not isinstance(storage, dict): + return {} + recorded = storage.get("r2") + return recorded if isinstance(recorded, dict) else {} + + def _upsert_manifest( manifest_path: Path, *, @@ -1076,8 +1120,18 @@ def _upsert_manifest( "size_bytes": size_bytes, "fetched_at": fetched_at, } - if r2_location is not None: - file_entry["storage"] = {"r2": r2_location.to_dict()} + recorded_r2 = _recorded_r2(payload["files"].get(year)) + new_r2 = r2_location.to_dict() if r2_location is not None else None + if recorded_r2 and ( + new_r2 is None or recorded_r2.get("bucket") != new_r2.get("bucket") + ): + # A recorded storage.r2 block is historical truth: archived witness + # records pin raw R2 URLs by hash. Re-fetching under a renamed bucket + # copies bytes; it does not restate where the bytes were first + # published (PolicyEngine/chronicle#143, mechanism 3). + file_entry["storage"] = {"r2": recorded_r2} + elif new_r2 is not None: + file_entry["storage"] = {"r2": new_r2} payload["files"][year] = file_entry manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False), @@ -1169,8 +1223,32 @@ def _publish_raw_manifest_entry( package_path=manifest_path, ), ) - storage = spec.get("storage") if isinstance(spec.get("storage"), dict) else {} - recorded_r2 = storage.get("r2") if isinstance(storage.get("r2"), dict) else {} + recorded_r2 = _recorded_r2(spec) + recorded_bucket = recorded_r2.get("bucket") + if recorded_bucket and recorded_bucket != location.bucket: + # The recorded bucket is preserved history. Publishing the same bytes + # into a renamed bucket is a backfill copy, not a restatement, so the + # manifest must not be rewritten to point at the new bucket. + errors.append( + "recorded_r2_bucket_is_preserved_history:" + f"recorded={recorded_bucket}:requested={location.bucket}" + ) + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(artifact_path), + sha256=sha256_actual, + size_bytes=size_bytes, + r2_location=None, + upload=None, + errors=tuple(errors), + ), + None, + ) recorded_key = recorded_r2.get("key") if recorded_key and recorded_key != location.key: errors.append( @@ -1275,7 +1353,7 @@ def _inventory_entry( def _derived_artifact_kind(artifact_name: str) -> str: - if artifact_name in {"ledger.db", "ledger.db"}: + if artifact_name in CHRONICLE_DB_FILENAMES: return "sqlite_database" if artifact_name.endswith(".jsonl"): return "jsonl" diff --git a/chronicle/consumer_contract.py b/chronicle/consumer_contract.py index a1c78325..f9c93f3a 100644 --- a/chronicle/consumer_contract.py +++ b/chronicle/consumer_contract.py @@ -468,6 +468,24 @@ def validate_consumer_fact_contract( ) +def _r2_uri_parts(uri: str) -> tuple[str, str]: + """Split an ``r2://bucket/key`` URI into its bucket and key.""" + if not uri.startswith("r2://"): + return "", "" + bucket, _, key = uri[len("r2://") :].partition("/") + return bucket, key + + +def _points_at_derived(bucket: str, key: str) -> bool: + """Whether an R2 bucket/key pair addresses derived build output. + + Matched on shape rather than on the ledger-era bucket names, so the guard + keeps firing once the buckets are renamed (PolicyEngine/chronicle#143, + mechanism 3). + """ + return bucket.endswith("-derived") or key.startswith("derived/") + + def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: """Return a boundary error if a fact is a downstream target derivation.""" source = fact.source @@ -489,20 +507,14 @@ def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: "itself. Target construction, aging, and reconciliation belong in " "Microcosm." ) - if source_file.startswith("ledger-derived:"): + source_file_bucket, bucket_separator, _ = source_file.partition(":") + if bucket_separator and source_file_bucket.endswith("-derived"): return ( "Chronicle consumer facts must cite raw publisher artifacts. Derived " "target-construction artifacts belong in Microcosm." ) - if ( - raw_r2_bucket.endswith("-derived") - or raw_r2_key.startswith("derived/") - or raw_r2_uri.startswith( - ( - "r2://ledger-derived/", - "r2://ledger-raw/derived/", - ) - ) + if _points_at_derived(raw_r2_bucket, raw_r2_key) or _points_at_derived( + *_r2_uri_parts(raw_r2_uri) ): return ( "Chronicle consumer facts must point at raw source artifacts, not " diff --git a/chronicle/database.py b/chronicle/database.py index 3f18ef78..5297db16 100644 --- a/chronicle/database.py +++ b/chronicle/database.py @@ -42,6 +42,15 @@ LEDGER_DB_SCHEMA_VERSION = schema_id("relational", Epoch.LEDGER) +# New suite outputs write chronicle.db. Existing builds wrote ledger.db and are +# still read and published unchanged, so the legacy name stays accepted for +# inference and artifact classification (PolicyEngine/chronicle#143, +# mechanism 3). The relational schema id above is a frozen machine surface that +# migrates with the epoch lane, not with this rename. +CHRONICLE_DB_FILENAME = "chronicle.db" +LEGACY_CHRONICLE_DB_FILENAME = "ledger.db" +CHRONICLE_DB_FILENAMES = (CHRONICLE_DB_FILENAME, LEGACY_CHRONICLE_DB_FILENAME) + @dataclass(frozen=True) class ChronicleDbBuildReport: diff --git a/chronicle/env.py b/chronicle/env.py index 80667f4d..3a2f7af6 100644 --- a/chronicle/env.py +++ b/chronicle/env.py @@ -75,7 +75,11 @@ def env_names(name: str) -> tuple[str, ...]: def _warn_legacy(found: str, preferred: str) -> None: - """Warn once per process that a ledger-era variable supplied a value.""" + """Warn once per process that a ledger-era variable supplied a value. + + ``stacklevel=4`` walks out through :func:`_first_set` and its public + wrapper so the notice points at the code that asked for the setting. + """ if found in _WARNED_LEGACY_NAMES: return _WARNED_LEGACY_NAMES.add(found) @@ -84,7 +88,7 @@ def _warn_legacy(found: str, preferred: str) -> None: f"set {preferred} instead. The old name is still honored during the " "Chronicle rename window and will be removed once consumers migrate.", ChronicleEnvDeprecationWarning, - stacklevel=3, + stacklevel=4, ) @@ -93,12 +97,11 @@ def reset_env_deprecation_state() -> None: _WARNED_LEGACY_NAMES.clear() -def env_value(*names: str, default: _Default = None) -> str | _Default: - """Read the first set value across ``names``, chronicle-preferred first. +def _first_set(names: tuple[str, ...]) -> str | None: + """Return the first set value across ``names``, warning on a legacy hit. - Each name is expanded through :func:`env_names`, so a caller can pass the - chronicle name and still pick up a value set under a ledger-era name. - Empty values are treated as unset, matching the helpers this replaces. + Both public readers call this at the same stack depth so the deprecation + warning is always attributed to their caller, not to this module. """ for name in names: candidates = env_names(name) @@ -109,12 +112,27 @@ def env_value(*names: str, default: _Default = None) -> str | _Default: if candidate != preferred: _warn_legacy(candidate, preferred) return value - return default + return None + + +def env_value(*names: str, default: _Default = None) -> str | _Default: + """Read the first set value across ``names``, chronicle-preferred first. + + Each name is expanded through :func:`env_names`, so a caller can pass the + chronicle name and still pick up a value set under a ledger-era name. + Empty values are treated as unset, matching the helpers this replaces. + """ + value = _first_set(names) + return default if value is None else value def env_flag(*names: str) -> bool: - """Return whether the first set value across ``names`` reads as true.""" - value = env_value(*names) + """Return whether the first set value across ``names`` reads as true. + + The chronicle-preferred name wins even when it reads false, so an operator + who has migrated can turn a flag off without unsetting the legacy name. + """ + value = _first_set(names) if value is None: return False return value.strip().lower() in TRUTHY_ENV_VALUES diff --git a/chronicle/harness.py b/chronicle/harness.py index 298a1380..54c14139 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -8,6 +8,8 @@ from pathlib import Path from chronicle.artifacts import ( + DEFAULT_R2_DERIVED_BUCKET, + DEFAULT_R2_RAW_BUCKET, ArtifactFetchReport, ArtifactInventoryReport, DerivedArtifactPublishReport, @@ -335,7 +337,7 @@ def fetch_artifact_file( table: str | None = None, filename: str | None = None, upload_r2: bool = False, - r2_bucket: str = "ledger-raw", + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> ArtifactFetchReport: @@ -372,7 +374,7 @@ def publish_raw_artifact_files( manifest_filename: str = "manifest.yaml", source_id: str | None = None, package_id: str | None = None, - r2_bucket: str = "ledger-raw", + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> RawArtifactPublishReport: @@ -390,8 +392,8 @@ def publish_raw_artifact_files( def bootstrap_r2_storage( *, - raw_bucket: str = "ledger-raw", - derived_bucket: str = "ledger-derived", + raw_bucket: str | None = None, + derived_bucket: str | None = None, wrangler_command: str = "npx wrangler", ) -> R2BootstrapReport: """Create Chronicle R2 buckets when Wrangler is authenticated.""" @@ -409,7 +411,7 @@ def publish_derived_artifact_files( package_id: str, year: int, build_id: str | None = None, - r2_bucket: str = "ledger-derived", + r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", build_artifacts_output: str | Path | None = None, @@ -888,8 +890,11 @@ def main(argv: list[str] | None = None) -> int: ) artifact_parser.add_argument( "--r2-bucket", - default="ledger-raw", - help="R2 bucket for raw artifacts when --upload-r2 is set.", + default=None, + help=( + "R2 bucket for raw artifacts when --upload-r2 is set. Defaults to " + f"$CHRONICLE_R2_RAW_BUCKET, else {DEFAULT_R2_RAW_BUCKET}." + ), ) artifact_parser.add_argument( "--r2-prefix", @@ -923,7 +928,7 @@ def main(argv: list[str] | None = None) -> int: raw_publish_parser = subparsers.add_parser( "publish-raw", - help="Upload manifest-declared raw source artifacts to ledger-raw R2", + help="Upload manifest-declared raw source artifacts to the raw R2 bucket", ) raw_publish_parser.add_argument( "--root", @@ -946,8 +951,11 @@ def main(argv: list[str] | None = None) -> int: ) raw_publish_parser.add_argument( "--r2-bucket", - default="ledger-raw", - help="R2 bucket for immutable raw artifacts.", + default=None, + help=( + "R2 bucket for immutable raw artifacts. Defaults to " + f"$CHRONICLE_R2_RAW_BUCKET, else {DEFAULT_R2_RAW_BUCKET}." + ), ) raw_publish_parser.add_argument( "--r2-prefix", @@ -969,13 +977,19 @@ def main(argv: list[str] | None = None) -> int: ) r2_parser.add_argument( "--raw-bucket", - default="ledger-raw", - help="R2 bucket name for immutable raw source artifacts.", + default=None, + help=( + "R2 bucket name for immutable raw source artifacts. Defaults to " + f"$CHRONICLE_R2_RAW_BUCKET, else {DEFAULT_R2_RAW_BUCKET}." + ), ) r2_parser.add_argument( "--derived-bucket", - default="ledger-derived", - help="R2 bucket name for derived Chronicle build artifacts.", + default=None, + help=( + "R2 bucket name for derived Chronicle build artifacts. Defaults to " + f"$CHRONICLE_R2_DERIVED_BUCKET, else {DEFAULT_R2_DERIVED_BUCKET}." + ), ) r2_parser.add_argument( "--wrangler-command", @@ -985,7 +999,7 @@ def main(argv: list[str] | None = None) -> int: derived_publish_parser = subparsers.add_parser( "publish-derived", - help="Upload deterministic Chronicle build outputs to ledger-derived R2", + help="Upload deterministic Chronicle build outputs to the derived R2 bucket", ) derived_publish_parser.add_argument( "--dir", @@ -1014,13 +1028,16 @@ def main(argv: list[str] | None = None) -> int: help=( "Build ID under an accepted epoch prefix, ledger.build.v1: or " "chronicle.build.v2:; any other form is refused. Defaults to " - "the ID inferred from reports or ledger.db." + "the ID inferred from reports, chronicle.db, or a legacy ledger.db." ), ) derived_publish_parser.add_argument( "--r2-bucket", - default="ledger-derived", - help="R2 bucket for derived build artifacts.", + default=None, + help=( + "R2 bucket for derived build artifacts. Defaults to " + f"$CHRONICLE_R2_DERIVED_BUCKET, else {DEFAULT_R2_DERIVED_BUCKET}." + ), ) derived_publish_parser.add_argument( "--r2-prefix", diff --git a/chronicle/suite.py b/chronicle/suite.py index 157b7420..0705c862 100644 --- a/chronicle/suite.py +++ b/chronicle/suite.py @@ -22,7 +22,11 @@ build_fact_key, validate_facts, ) -from chronicle.database import ChronicleDbBuildReport, build_chronicle_db +from chronicle.database import ( + CHRONICLE_DB_FILENAME, + ChronicleDbBuildReport, + build_chronicle_db, +) from chronicle.epoch import canonicalize_key from chronicle.sources.cells import ( SourceCell, @@ -372,7 +376,7 @@ def build_source_suite( concept_report.to_dict(), ) - db_path = output_path / "ledger.db" + db_path = output_path / CHRONICLE_DB_FILENAME db_report = build_chronicle_db( facts, db_path, @@ -1675,7 +1679,7 @@ def _write_package_sidecars(output_path: Path, *, source: str, year: int) -> Non output_path / "source_regions.jsonl", output_path / "facts.jsonl", output_path / "consumer_facts.jsonl", - output_path / "ledger.db", + output_path / CHRONICLE_DB_FILENAME, output_path / "reports" / "source_rows.json", output_path / "reports" / "source_cells.json", output_path / "reports" / "source_regions.json", diff --git a/db/cli.py b/db/cli.py index 52257240..66538149 100644 --- a/db/cli.py +++ b/db/cli.py @@ -276,14 +276,15 @@ def cmd_query(args): def _pe_source_root_env_default(jurisdiction: str) -> str | None: + from chronicle.env import env_value + from .pe_source_inventory import ( PE_UK_DATA_ROOT_ENV, PE_US_DATA_ROOT_ENV, - _env_value, ) env_var = PE_US_DATA_ROOT_ENV if jurisdiction == "us" else PE_UK_DATA_ROOT_ENV - return _env_value(env_var) + return env_value(env_var) def main(): diff --git a/tests/test_chronicle_mirror.py b/tests/test_chronicle_mirror.py index 787429c1..c06a33bd 100644 --- a/tests/test_chronicle_mirror.py +++ b/tests/test_chronicle_mirror.py @@ -18,7 +18,7 @@ def test_export_chronicle_db_tables_writes_jsonl_and_manifest(tmp_path): - db_path = tmp_path / "ledger.db" + db_path = tmp_path / "chronicle.db" output_dir = tmp_path / "mirror" build_chronicle_db( build_soi_table_1_1_facts(2023), @@ -53,7 +53,7 @@ def test_export_chronicle_db_tables_writes_jsonl_and_manifest(tmp_path): def test_export_chronicle_db_tables_orders_rows_deterministically(tmp_path): - db_path = tmp_path / "ledger.db" + db_path = tmp_path / "chronicle.db" first_output_dir = tmp_path / "mirror-first" second_output_dir = tmp_path / "mirror-second" build_chronicle_db( @@ -72,7 +72,7 @@ def test_export_chronicle_db_tables_orders_rows_deterministically(tmp_path): def test_export_db_tables_cli_emits_manifest_summary(tmp_path, capsys): - db_path = tmp_path / "ledger.db" + db_path = tmp_path / "chronicle.db" output_dir = tmp_path / "mirror" build_chronicle_db( build_soi_table_1_1_facts(2023), @@ -99,7 +99,7 @@ def test_export_db_tables_cli_emits_manifest_summary(tmp_path, capsys): def test_load_supabase_mirror_dry_run_counts_exported_rows(tmp_path): - db_path = tmp_path / "ledger.db" + db_path = tmp_path / "chronicle.db" output_dir = tmp_path / "mirror" build_chronicle_db( build_soi_table_1_1_facts(2023), diff --git a/tests/test_chronicle_suite.py b/tests/test_chronicle_suite.py index 6911d6e2..c08f8abe 100644 --- a/tests/test_chronicle_suite.py +++ b/tests/test_chronicle_suite.py @@ -96,7 +96,7 @@ def test_build_source_suite_writes_artifacts_and_reports(tmp_path): assert (output_dir / "source_regions.jsonl").exists() assert (output_dir / "facts.jsonl").exists() assert (output_dir / "consumer_facts.jsonl").exists() - assert (output_dir / "ledger.db").exists() + assert (output_dir / "chronicle.db").exists() assert (output_dir / "datapackage.json").exists() assert (output_dir / "ro-crate-metadata.json").exists() assert (output_dir / "reports" / "source_regions.json").exists() @@ -124,7 +124,7 @@ def test_build_source_suite_writes_artifacts_and_reports(tmp_path): "source_regions.jsonl", "facts.jsonl", "consumer_facts.jsonl", - "ledger.db", + "chronicle.db", "reports/build_summary.json", "reports/source_regions.json", "reports/selectors.json", @@ -150,7 +150,7 @@ def test_build_source_suite_writes_artifacts_and_reports(tmp_path): "concept_alignment_validation_skipped" ] - with sqlite3.connect(output_dir / "ledger.db") as connection: + with sqlite3.connect(output_dir / "chronicle.db") as connection: facts_count = connection.execute( "SELECT COUNT(*) FROM aggregate_facts" ).fetchone()[0] @@ -184,7 +184,7 @@ def test_build_source_suite_supports_soi_table_1_4(tmp_path): "concept_alignment_validation_skipped" ) assert (output_dir / "source_regions.jsonl").exists() - assert (output_dir / "ledger.db").exists() + assert (output_dir / "chronicle.db").exists() def test_agent_acceptance_accepts_aggregate_income_range_source_rows(): @@ -739,7 +739,7 @@ def test_build_suite_cli_emits_json_summary(tmp_path, capsys): assert payload["outputs"]["source_regions"] == str( output_dir / "source_regions.jsonl" ) - assert payload["outputs"]["database"] == str(output_dir / "ledger.db") + assert payload["outputs"]["database"] == str(output_dir / "chronicle.db") assert payload["outputs"]["consumer_facts"] == str( output_dir / "consumer_facts.jsonl" ) From c38659467949f6eeb0d0a565af49ad20f929f6cd Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:25:23 -0400 Subject: [PATCH 03/42] Add hermetic tests for the rename window tests/test_chronicle_env.py covers the shared helper end to end: the lookup ladder, CHRONICLE_* winning over both ledger-era spellings without a warning, each legacy spelling still working with a once-per-process warning attributed to the caller, empty values counting as unset, and a migrated operator being able to turn a flag off without unsetting the stale legacy name. It then exercises the real call sites -- source-package artifact cache and fetch flag, the db CLI's --pe-us-root default, the Supabase schema override, and R2 bucket resolution -- so a helper regression cannot pass by only testing the helper. An autouse fixture strips every rename-window variable from the ambient environment, so the file is hermetic under any shell. The artifacts tests cover the chronicle.db write and ledger.db read fallback, both database names classifying as sqlite_database, publish-derived following the configured bucket, and the two manifest-preservation paths: publish-raw refusing to restate a recorded bucket and fetch-artifact keeping the recorded r2 block while still uploading the backfill copy. Co-Authored-By: Claude Fable 5.1 --- tests/test_chronicle_artifacts.py | 169 +++++++++++++++ tests/test_chronicle_env.py | 334 ++++++++++++++++++++++++++++++ 2 files changed, 503 insertions(+) create mode 100644 tests/test_chronicle_env.py diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 7104ec03..a96d6c25 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -4,6 +4,7 @@ import hashlib import json +import sqlite3 import pytest import yaml @@ -658,3 +659,171 @@ def test_top_level_cli_dispatches_publish_derived(tmp_path, capsys, monkeypatch) assert exc.value.code == 0 assert payload["valid"] + + +def _sqlite_build(path, build_id): + """Write a minimal build database carrying one ledger_builds row.""" + with sqlite3.connect(path) as connection: + connection.execute("CREATE TABLE ledger_builds (build_id TEXT PRIMARY KEY)") + connection.execute("INSERT INTO ledger_builds VALUES (?)", (build_id,)) + + +@pytest.mark.parametrize("db_name", ["chronicle.db", "ledger.db"]) +def test_infer_build_id_reads_new_and_legacy_database_names(tmp_path, db_name): + suite = tmp_path / "suite" + suite.mkdir() + _sqlite_build(suite / db_name, "ledger.build.v1:from-db") + + assert infer_build_id(suite) == "ledger.build.v1:from-db" + + +def test_infer_build_id_prefers_the_chronicle_database(tmp_path): + suite = tmp_path / "suite" + suite.mkdir() + _sqlite_build(suite / "chronicle.db", "ledger.build.v1:chronicle") + _sqlite_build(suite / "ledger.db", "ledger.build.v1:legacy") + + assert infer_build_id(suite) == "ledger.build.v1:chronicle" + + +@pytest.mark.parametrize("db_name", ["chronicle.db", "ledger.db"]) +def test_publish_derived_classifies_both_database_names(tmp_path, db_name): + suite = tmp_path / "suite" + reports = suite / "reports" + reports.mkdir(parents=True) + build_id = "ledger.build.v1:kind" + (reports / "database.json").write_text(json.dumps({"build_id": build_id})) + (suite / db_name).write_bytes(b"db") + wrangler = tmp_path / "wrangler" + wrangler.write_text("#!/bin/sh\necho ok\n") + wrangler.chmod(0o755) + + report = publish_derived_artifacts( + suite, + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + wrangler_command=str(wrangler), + ) + rows = {row["artifact_name"]: row for row in build_artifact_rows(report)} + + assert rows[db_name]["artifact_kind"] == "sqlite_database" + + +def test_publish_derived_uses_the_configured_bucket(tmp_path, monkeypatch): + monkeypatch.setenv("CHRONICLE_R2_DERIVED_BUCKET", "chronicle-derived") + suite = tmp_path / "suite" + reports = suite / "reports" + reports.mkdir(parents=True) + (reports / "database.json").write_text( + json.dumps({"build_id": "ledger.build.v1:bucket"}) + ) + (suite / "facts.jsonl").write_text("{}\n") + log = tmp_path / "wrangler.log" + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + + report = publish_derived_artifacts( + suite, + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + wrangler_command=str(wrangler), + ) + + assert report.valid + assert report.entries[0].r2_location.bucket == "chronicle-derived" + assert "chronicle-derived/derived/irs_soi/" in log.read_text() + + +def test_publish_raw_refuses_to_restate_a_recorded_bucket(tmp_path, monkeypatch): + """A recorded storage.r2 bucket is preserved history, not a publish target. + + Archived witness records pin raw R2 URLs by hash, so backfilling the same + bytes into a renamed bucket must not rewrite the manifest. + """ + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-1-1" + source = tmp_path / "soi.xlsx" + source.write_bytes(b"official SOI workbook") + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + output_dir=output_dir, + ) + manifest_path = output_dir / "manifest.yaml" + manifest = yaml.safe_load(manifest_path.read_text()) + artifact = manifest["files"][2023] + recorded_key = ( + f"raw/irs_soi/soi-table-1-1/2023/{artifact['sha256']}/{artifact['filename']}" + ) + artifact["storage"] = { + "r2": { + "provider": "r2", + "bucket": "ledger-raw", + "key": recorded_key, + "uri": f"r2://ledger-raw/{recorded_key}", + } + } + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + log = tmp_path / "wrangler.log" + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + + report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) + unchanged = yaml.safe_load(manifest_path.read_text()) + + assert not report.valid + assert ( + report.entries[0] + .errors[0] + .startswith("recorded_r2_bucket_is_preserved_history:") + ) + assert not log.exists() + assert unchanged["files"][2023]["storage"]["r2"]["bucket"] == "ledger-raw" + + +def test_fetch_artifact_keeps_an_already_recorded_bucket(tmp_path, monkeypatch): + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-1-1" + source = tmp_path / "soi.xlsx" + source.write_bytes(b"official SOI workbook") + log = tmp_path / "wrangler.log" + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + output_dir=output_dir, + upload_r2=True, + wrangler_command=str(wrangler), + ) + manifest_path = output_dir / "manifest.yaml" + first = yaml.safe_load(manifest_path.read_text()) + assert first["files"][2023]["storage"]["r2"]["bucket"] == "ledger-raw" + + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + report = fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-1", + year=2023, + output_dir=output_dir, + upload_r2=True, + wrangler_command=str(wrangler), + ) + second = yaml.safe_load(manifest_path.read_text()) + + # The backfill copy really is uploaded to the new bucket, but the manifest + # keeps recording where the bytes were first published. + assert report.r2_location.bucket == "chronicle-raw" + assert "chronicle-raw" in log.read_text() + assert ( + second["files"][2023]["storage"]["r2"] == first["files"][2023]["storage"]["r2"] + ) diff --git a/tests/test_chronicle_env.py b/tests/test_chronicle_env.py new file mode 100644 index 00000000..4ffe0206 --- /dev/null +++ b/tests/test_chronicle_env.py @@ -0,0 +1,334 @@ +"""Tests for the chronicle-first environment read window. + +Chronicle's operational stores migrate by dual-run (PolicyEngine/chronicle#143, +mechanism 3): ``CHRONICLE_*`` names win, ledger-era names keep working behind a +deprecation warning. Every test here is hermetic — the fixture strips every +variable in the rename window from the ambient environment first. +""" + +from __future__ import annotations + +import importlib +import os +from pathlib import Path + +import pytest + +from chronicle.artifacts import ( + DEFAULT_R2_DERIVED_BUCKET, + DEFAULT_R2_RAW_BUCKET, + default_r2_derived_bucket, + default_r2_raw_bucket, +) +from chronicle.env import ( + CHRONICLE_ENV_PREFIX, + ChronicleEnvDeprecationWarning, + LEGACY_ENV_PREFIXES, + env_flag, + env_names, + env_value, + reset_env_deprecation_state, +) +from chronicle.harness import main as harness_main +from chronicle.source_package import ( + SOURCE_ARTIFACT_CACHE_ENV, + SOURCE_ARTIFACT_FETCH_ENV, +) + +RENAME_WINDOW_PREFIXES = (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES) + + +@pytest.fixture(autouse=True) +def isolated_rename_window_env(monkeypatch): + """Run each test with no rename-window variable inherited from the shell.""" + for name in list(os.environ): + if name.startswith(RENAME_WINDOW_PREFIXES): + monkeypatch.delenv(name, raising=False) + reset_env_deprecation_state() + yield + reset_env_deprecation_state() + + +def _fake_wrangler(tmp_path, log): + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + return wrangler + + +# --------------------------------------------------------------------------- +# Lookup order +# --------------------------------------------------------------------------- + + +def test_env_names_puts_chronicle_first_then_ledger_era_names(): + assert env_names("CHRONICLE_SOURCE_ARTIFACT_FETCH") == ( + "CHRONICLE_SOURCE_ARTIFACT_FETCH", + "POLICYENGINE_LEDGER_SOURCE_ARTIFACT_FETCH", + "LEDGER_SOURCE_ARTIFACT_FETCH", + ) + + +def test_env_names_expands_a_ledger_era_name_to_the_same_ladder(): + assert env_names("LEDGER_PE_US_DATA_ROOT") == env_names("CHRONICLE_PE_US_DATA_ROOT") + + +def test_env_names_leaves_variables_outside_the_rename_window_alone(): + assert env_names("POLICYENGINE_SUPABASE_URL") == ("POLICYENGINE_SUPABASE_URL",) + assert env_names("POLICYENGINE_TARGETS_SCHEMA") == ("POLICYENGINE_TARGETS_SCHEMA",) + + +def test_bare_prefix_is_not_treated_as_a_renamed_variable(): + assert env_names("LEDGER_") == ("LEDGER_",) + + +# --------------------------------------------------------------------------- +# Precedence and the deprecation warning +# --------------------------------------------------------------------------- + + +def test_chronicle_name_wins_over_both_ledger_era_names(monkeypatch, recwarn): + monkeypatch.setenv("CHRONICLE_PE_US_DATA_ROOT", "/chronicle") + monkeypatch.setenv("LEDGER_PE_US_DATA_ROOT", "/ledger") + monkeypatch.setenv("POLICYENGINE_LEDGER_PE_US_DATA_ROOT", "/policyengine-ledger") + + assert env_value("CHRONICLE_PE_US_DATA_ROOT") == "/chronicle" + assert not [ + warning + for warning in recwarn.list + if issubclass(warning.category, ChronicleEnvDeprecationWarning) + ] + + +def test_ledger_name_alone_still_works_and_warns(monkeypatch): + monkeypatch.setenv("LEDGER_PE_US_DATA_ROOT", "/ledger") + + with pytest.warns(ChronicleEnvDeprecationWarning) as warnings_raised: + assert env_value("CHRONICLE_PE_US_DATA_ROOT") == "/ledger" + + message = str(warnings_raised[0].message) + assert "LEDGER_PE_US_DATA_ROOT" in message + assert "CHRONICLE_PE_US_DATA_ROOT" in message + + +def test_policyengine_ledger_name_alone_still_works_and_warns(monkeypatch): + monkeypatch.setenv("POLICYENGINE_LEDGER_SCHEMA", "ledger") + + with pytest.warns(ChronicleEnvDeprecationWarning) as warnings_raised: + assert env_value("CHRONICLE_SCHEMA") == "ledger" + + message = str(warnings_raised[0].message) + assert "POLICYENGINE_LEDGER_SCHEMA" in message + assert "CHRONICLE_SCHEMA" in message + + +def test_deprecation_warning_is_raised_once_per_process(monkeypatch, recwarn): + monkeypatch.setenv("LEDGER_PE_UK_DATA_ROOT", "/ledger") + + for _ in range(3): + assert env_value("CHRONICLE_PE_UK_DATA_ROOT") == "/ledger" + + deprecations = [ + warning + for warning in recwarn.list + if issubclass(warning.category, ChronicleEnvDeprecationWarning) + ] + assert len(deprecations) == 1 + + +def test_deprecation_warning_is_attributed_to_the_calling_module(monkeypatch): + monkeypatch.setenv("LEDGER_SOURCE_ARTIFACT_FETCH", "1") + + with pytest.warns(ChronicleEnvDeprecationWarning) as warnings_raised: + assert env_flag(SOURCE_ARTIFACT_FETCH_ENV) + + # env_flag and env_value must report at the same depth, or operators get a + # notice pointing at Chronicle's own source instead of their call site. + assert Path(warnings_raised[0].filename).name == "test_chronicle_env.py" + + +def test_unset_variables_fall_back_to_the_default(): + assert env_value("CHRONICLE_PE_US_DATA_ROOT") is None + assert env_value("CHRONICLE_PE_US_DATA_ROOT", default="/fallback") == "/fallback" + + +def test_empty_values_count_as_unset(monkeypatch): + monkeypatch.setenv("CHRONICLE_PE_US_DATA_ROOT", "") + monkeypatch.setenv("LEDGER_PE_US_DATA_ROOT", "/ledger") + + with pytest.warns(ChronicleEnvDeprecationWarning): + assert env_value("CHRONICLE_PE_US_DATA_ROOT") == "/ledger" + + +# --------------------------------------------------------------------------- +# Flags +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize("value", ["1", "true", "TRUE", "yes", "on", " on "]) +def test_env_flag_accepts_truthy_spellings(monkeypatch, value): + monkeypatch.setenv("CHRONICLE_SOURCE_ARTIFACT_FETCH", value) + assert env_flag(SOURCE_ARTIFACT_FETCH_ENV) + + +@pytest.mark.parametrize("value", ["0", "false", "no", "off", "maybe"]) +def test_env_flag_rejects_other_values(monkeypatch, value): + monkeypatch.setenv("CHRONICLE_SOURCE_ARTIFACT_FETCH", value) + assert not env_flag(SOURCE_ARTIFACT_FETCH_ENV) + + +def test_env_flag_lets_the_chronicle_name_turn_a_legacy_flag_off(monkeypatch): + monkeypatch.setenv("CHRONICLE_SOURCE_ARTIFACT_FETCH", "0") + monkeypatch.setenv("LEDGER_SOURCE_ARTIFACT_FETCH", "1") + + # An operator who has migrated must be able to turn the flag off without + # first hunting down the stale ledger-era variable. + assert not env_flag(SOURCE_ARTIFACT_FETCH_ENV) + + +# --------------------------------------------------------------------------- +# Real call sites +# --------------------------------------------------------------------------- + + +def test_source_artifact_env_constants_are_chronicle_named(): + assert SOURCE_ARTIFACT_CACHE_ENV == "CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR" + assert SOURCE_ARTIFACT_FETCH_ENV == "CHRONICLE_SOURCE_ARTIFACT_FETCH" + + +@pytest.mark.parametrize( + "name", + ["CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR", "LEDGER_SOURCE_ARTIFACT_CACHE_DIR"], +) +def test_source_artifact_cache_dir_honors_both_names(monkeypatch, tmp_path, name): + from chronicle.source_package import _source_artifact_cache_path + + monkeypatch.setenv(name, str(tmp_path)) + + cache_path = _source_artifact_cache_path( + {"filename": "table.xlsx", "sha256": "abc123"} + ) + + assert cache_path == tmp_path / "abc123" / "table.xlsx" + + +@pytest.mark.parametrize( + "name", ["CHRONICLE_PE_US_DATA_ROOT", "LEDGER_PE_US_DATA_ROOT"] +) +def test_pe_source_root_cli_default_honors_both_names(monkeypatch, name): + from db.cli import _pe_source_root_env_default + + monkeypatch.setenv(name, "/pe-us") + + assert _pe_source_root_env_default("us") == "/pe-us" + + +def test_pe_source_inventory_env_constants_are_chronicle_named(): + from db.pe_source_inventory import PE_UK_DATA_ROOT_ENV, PE_US_DATA_ROOT_ENV + + assert PE_US_DATA_ROOT_ENV == "CHRONICLE_PE_US_DATA_ROOT" + assert PE_UK_DATA_ROOT_ENV == "CHRONICLE_PE_UK_DATA_ROOT" + + +def test_db_cli_parser_builds_with_the_env_backed_defaults(monkeypatch, capsys): + """The db CLI builds its parser before dispatching any subcommand. + + Its --pe-us-root/--pe-uk-root defaults call into the env helper, so an + import error there breaks `chronicle init`, `load` and `stats` alike while + the rest of the test suite stays green. + """ + import db.cli + + monkeypatch.setenv("CHRONICLE_PE_US_DATA_ROOT", "/pe-us") + monkeypatch.setattr("sys.argv", ["chronicle", "--help"]) + + with pytest.raises(SystemExit) as exit_info: + db.cli.main() + + assert exit_info.value.code == 0 + assert "Manage Chronicle target input data" in capsys.readouterr().out + + +@pytest.mark.parametrize("name", ["CHRONICLE_SCHEMA", "POLICYENGINE_LEDGER_SCHEMA"]) +def test_supabase_schema_honors_both_names(monkeypatch, name): + import db.supabase_client + + monkeypatch.setenv(name, "chronicle_probe") + try: + reloaded = importlib.reload(db.supabase_client) + assert reloaded.LEDGER_SCHEMA == "chronicle_probe" + finally: + monkeypatch.delenv(name, raising=False) + importlib.reload(db.supabase_client) + + +def test_supabase_schema_default_is_unchanged(): + import db.supabase_client + + # The hosted schema name itself is out of this slice; only the variable + # that overrides it moved. + assert db.supabase_client.LEDGER_SCHEMA == "ledger" + + +# --------------------------------------------------------------------------- +# R2 bucket configuration +# --------------------------------------------------------------------------- + + +def test_r2_bucket_defaults_are_still_the_ledger_era_names(): + assert DEFAULT_R2_RAW_BUCKET == "ledger-raw" + assert DEFAULT_R2_DERIVED_BUCKET == "ledger-derived" + assert default_r2_raw_bucket() == "ledger-raw" + assert default_r2_derived_bucket() == "ledger-derived" + + +def test_r2_buckets_follow_the_chronicle_env_vars(monkeypatch): + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + monkeypatch.setenv("CHRONICLE_R2_DERIVED_BUCKET", "chronicle-derived") + + assert default_r2_raw_bucket() == "chronicle-raw" + assert default_r2_derived_bucket() == "chronicle-derived" + + +def test_r2_buckets_honor_ledger_era_names_with_a_warning(monkeypatch): + monkeypatch.setenv("LEDGER_R2_RAW_BUCKET", "legacy-raw") + + with pytest.warns(ChronicleEnvDeprecationWarning): + assert default_r2_raw_bucket() == "legacy-raw" + + +def test_bootstrap_r2_cli_creates_the_configured_buckets(monkeypatch, tmp_path): + log = tmp_path / "wrangler.log" + wrangler = _fake_wrangler(tmp_path, log) + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + monkeypatch.setenv("CHRONICLE_R2_DERIVED_BUCKET", "chronicle-derived") + + exit_code = harness_main(["bootstrap-r2", "--wrangler-command", str(wrangler)]) + + commands = log.read_text() + assert exit_code == 0 + assert "r2 bucket create chronicle-raw" in commands + assert "r2 bucket create chronicle-derived" in commands + + +def test_bootstrap_r2_cli_flags_still_override_the_environment(monkeypatch, tmp_path): + log = tmp_path / "wrangler.log" + wrangler = _fake_wrangler(tmp_path, log) + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + + harness_main( + [ + "bootstrap-r2", + "--raw-bucket", + "explicit-raw", + "--derived-bucket", + "explicit-derived", + "--wrangler-command", + str(wrangler), + ] + ) + + commands = log.read_text() + assert "r2 bucket create explicit-raw" in commands + assert "r2 bucket create explicit-derived" in commands + assert "chronicle-raw" not in commands From 69c7c4fc3ef6dec7eba85e87fb7bc7563b63e9eb Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:31:20 -0400 Subject: [PATCH 04/42] Document the rename window and the bucket cutover docs/storage-architecture.md stated the fallback direction backwards: it claimed the CHRONICLE_-prefixed variables were the old names kept as migration fallbacks, when they are the names to migrate to. Replaced with an "Environment Variable Rename Window" section that gives the actual lookup order, the once-per-process warning, and a table of all seven variables. Two behaviors get stated outright because both invert a naive fallback: the CHRONICLE_ name wins even when it reads false, so a migrated operator can turn a flag off without hunting down a stale legacy export; and an empty value counts as unset. The section also records what is deliberately NOT aliased -- POLICYENGINE_SUPABASE_* and POLICYENGINE_TARGETS_SCHEMA carry no ledger prefix and are read literally -- and that CHRONICLE_SCHEMA renames the override variable, not the schema value, which moves in a later slice. Added a "Bucket Cutover" section covering the six steps: create the buckets on the account wrangler.toml pins, enumerate, backfill-copy, verify, flip the defaults in a follow-up, and leave the ledger-era buckets read-only forever. The enumeration and verification are given as commands rather than prose. Every raw key ends {sha256}/{filename}, so the key is its own checksum witness and verification needs no manifest lookup. Counted rather than assumed: 186 distinct ledger-raw objects across 154 tracked manifest files, and the doc says to recount rather than trust that number, since source packages land continuously. The derived bucket needs no backfill -- derived artifacts are reproducible and already keyed by {build_id}, so a rebuild republishes them wherever configured. README and the harness doc follow: bucket references become archive roles with the default named, ledger.db becomes chronicle.db with the legacy name noted as still readable, and bootstrap-r2's example drops the hardcoded bucket flags now that they default to the resolver. Left alone: LEDGER_EXPLORER_DATA_DIRS in the README. It configures the explorer app, which is not in this repository, so renaming it here would document a name no shipped code honors -- the same defect this commit fixes in the other direction. Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 42 ++++-- README.md | 31 +++-- docs/agent-source-package-harness.md | 27 ++-- docs/storage-architecture.md | 191 +++++++++++++++++++++++++-- 4 files changed, 249 insertions(+), 42 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index e46b096c..da5d13e5 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -11,6 +11,8 @@ Lane C5's handoff notes previously lived here; its durable record is - Out of scope and deliberately untouched: the `ledger` console-script alias, the Supabase `"ledger"` schema and mirror table names, governance role ids and concept authorities, hash domains and schema ids, anything under `releases/`. +- This PR does not touch the source-data boundary. No package spec, parser, + selector, manifest, or fact value changes. ## Done @@ -24,18 +26,40 @@ Lane C5's handoff notes previously lived here; its durable record is `LEDGER_MIRROR_PRIMARY_KEYS`, and `LEDGER_DB_SCHEMA_VERSION` are module constants, not env reads, and name out-of-scope surfaces. - Added `chronicle/env.py`: one shared `env_value`/`env_flag`/`env_names` - helper reading `CHRONICLE_` first, then `LEDGER_` and - `POLICYENGINE_LEDGER_` with a once-per-process - `ChronicleEnvDeprecationWarning` naming the preferred variable. + helper reading `CHRONICLE_` first, then `POLICYENGINE_LEDGER_` and + `LEDGER_` with a once-per-process `ChronicleEnvDeprecationWarning` naming + the preferred variable. - Replaced all three ad-hoc helpers (`db/supabase_client._env`, `chronicle/source_package._env_value`/`_truthy_env`, `db/pe_source_inventory._env_value`) with the shared helper. +- Made the R2 bucket names configurable via `CHRONICLE_R2_RAW_BUCKET` and + `CHRONICLE_R2_DERIVED_BUCKET`, plumbed through fetch-artifact, publish-raw, + publish-derived and bootstrap-r2. Defaults unchanged at `ledger-raw` and + `ledger-derived`. Both manifest write paths now preserve a recorded + `storage.r2` block instead of restating it under a renamed bucket. +- Emitted `chronicle.db` for new suite outputs, with `ledger.db` still accepted + on read and on derived-artifact kind inference. +- Added `tests/test_chronicle_env.py` plus artifact tests: 75 hermetic tests + covering the lookup ladder, precedence, the once-per-process warning, and every + real call site. +- Swept the docs. `docs/storage-architecture.md` gained an "Environment Variable + Rename Window" section (the old text stated the fallback direction backwards) + and a "Bucket Cutover" section; `docs/agent-source-package-harness.md` and + `README.md` follow. Verified 186 distinct `ledger-raw` objects across 154 + tracked manifest files, every key content-addressed by sha256. + +## Verification + +- `uv run pytest -q`: green. +- `uv run ruff check .`: clean. +- `uv run ruff format --check .`: clean for every file this branch touches. 13 + files are unformatted on `main` already and are byte-identical here; CI runs + `ruff check` only, so they are pre-existing and out of scope. +- CI's db CLI gate (`chronicle init` / `load all` / `stats`): passes. ## Next -- Make R2 bucket names configurable with unchanged `ledger-*` defaults. -- Emit `chronicle.db` for new suite outputs; keep reading `ledger.db`. -- Fix the backwards fallback statement in the docs and sweep every env-name, - bucket-name, and db-filename mention in README/AGENTS/docs. -- Add the hermetic dual-read tests; run pytest, ruff check, ruff format --check. -- Push and open the PR. Do not merge. +- Push and open the PR against `main`. Do not merge. +- Follow-up PR, after Max creates and backfills the new buckets: flip + `DEFAULT_R2_RAW_BUCKET` / `DEFAULT_R2_DERIVED_BUCKET` to `chronicle-raw` / + `chronicle-derived`. diff --git a/README.md b/README.md index f75fe697..e0ade235 100644 --- a/README.md +++ b/README.md @@ -65,10 +65,12 @@ contract that aligns it to another period (see | Microcosm Target Contracts | Selection, measurement bindings, and active subset | Period alignment, support-aware activation, solver inputs, diagnostics | The storage split is documented in -[`docs/storage-architecture.md`](docs/storage-architecture.md): `ledger-raw` -stores immutable source bytes, `ledger-derived` stores reproducible build +[`docs/storage-architecture.md`](docs/storage-architecture.md): a raw R2 archive +stores immutable source bytes, a derived R2 archive stores reproducible build artifacts, and Supabase/Postgres hosts the queryable relational Chronicle registry -mirrored from accepted builds. +mirrored from accepted builds. The bucket names are configuration +(`$CHRONICLE_R2_RAW_BUCKET` and `$CHRONICLE_R2_DERIVED_BUCKET`), still defaulting +to the ledger-era `ledger-raw` and `ledger-derived`. ## Repository Model @@ -247,7 +249,7 @@ This writes: source_regions.jsonl facts.jsonl consumer_facts.jsonl - ledger.db + chronicle.db reports/ source_rows.json source_cells.json @@ -349,8 +351,9 @@ needed, even when your Cloudflare user belongs to several accounts: # One-time per machine (opens a browser consent page): bunx wrangler login -# One-time per account (already done for the PolicyEngine account): -uv run chronicle bootstrap-r2 --raw-bucket ledger-raw --derived-bucket ledger-derived +# One-time per account (already done for the PolicyEngine account). The bucket +# flags default to $CHRONICLE_R2_RAW_BUCKET / $CHRONICLE_R2_DERIVED_BUCKET: +uv run chronicle bootstrap-r2 # Fetch/register a source artifact, write db/data/.../manifest.yaml, and upload # the exact bytes to R2 when Wrangler is authenticated: @@ -367,8 +370,8 @@ uv run chronicle fetch-artifact \ # Audit local manifests and checksums: uv run chronicle inventory-artifacts --root db/data -# Upload all existing manifest-declared local artifacts to ledger-raw and write -# storage.r2 metadata back into the manifests: +# Upload all existing manifest-declared local artifacts to the raw archive and +# write storage.r2 metadata back into the manifests: uv run chronicle publish-raw --root db/data ``` @@ -406,10 +409,10 @@ To prepare the deterministic SQLite artifact for a hosted Supabase/Postgres mirror, export each relational table to JSONL plus a manifest: ```bash -uv run chronicle export-db-tables --db /tmp/chronicle-suite/ledger.db --out /tmp/chronicle-mirror --replace +uv run chronicle export-db-tables --db /tmp/chronicle-suite/chronicle.db --out /tmp/chronicle-mirror --replace ``` -To publish the deterministic build outputs to the `ledger-derived` R2 bucket: +To publish the deterministic build outputs to the derived R2 archive: ```bash uv run chronicle publish-derived \ @@ -437,6 +440,14 @@ uv run chronicle load-supabase-mirror \ Use `--dry-run` first to validate JSONL row counts and file coverage without writing to Supabase. +Chronicle settings are read chronicle-first: `CHRONICLE_X` wins, and the +ledger-era `POLICYENGINE_LEDGER_X` and `LEDGER_X` spellings still work behind a +one-time deprecation warning naming the variable to move to. +[`docs/storage-architecture.md`](docs/storage-architecture.md#environment-variable-rename-window) +lists every variable in that window, and +[Bucket Cutover](docs/storage-architecture.md#bucket-cutover) covers the R2 +bucket rename. + Chronicle facts keep source concepts and canonical concepts separately. For example, the SOI Table 1.1 adjusted gross income column is preserved as `irs_soi.adjusted_gross_income`, while the canonical concept is diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 5816ec47..afeb98f4 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -14,8 +14,9 @@ lineage, provenance, constraints, and a passing `build-suite` report. The first gate for a new package is source-artifact acquisition. Agents should register raw source files with `uv run chronicle fetch-artifact` before authoring selectors. This writes the local artifact, captures checksum and retrieval -metadata in `manifest.yaml`, and can upload the exact bytes to the private -`ledger-raw` R2 bucket when Wrangler is authenticated. Agents can audit the local +metadata in `manifest.yaml`, and can upload the exact bytes to the private raw +R2 bucket (`ledger-raw` today; overridable with `CHRONICLE_R2_RAW_BUCKET`) when +Wrangler is authenticated. Agents can audit the local artifact registry with `uv run chronicle inventory-artifacts --root db/data`. For already-downloaded manifest artifacts, agents should run `uv run chronicle publish-raw --root db/data` to upload checksum-verified bytes to @@ -23,12 +24,15 @@ R2 and write `storage.r2` metadata back into each manifest entry. Builds do not require production raw bytes to be committed to Git. Source packages first read packaged fixture bytes, then -`LEDGER_SOURCE_ARTIFACT_CACHE_DIR` (defaulting to +`CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR` (defaulting to `~/.cache/policyengine-chronicle/source-artifacts`). If a manifest artifact is -missing locally, set `LEDGER_SOURCE_ARTIFACT_FETCH=1` to fetch it from the +missing locally, set `CHRONICLE_SOURCE_ARTIFACT_FETCH=1` to fetch it from the manifest `source_url`, verify the declared SHA-256, and write it to that cache. -The old `CHRONICLE_`-prefixed environment variables remain accepted only as -migration fallbacks. +The ledger-era spellings `LEDGER_SOURCE_ARTIFACT_CACHE_DIR` and +`LEDGER_SOURCE_ARTIFACT_FETCH` are still honored during the rename window and +emit a one-time deprecation warning naming the `CHRONICLE_` variable to set +instead; see "Environment Variable Rename Window" in +[`docs/storage-architecture.md`](storage-architecture.md#environment-variable-rename-window). For broad PE source migration, generate the agent queue from the manifest before assigning work: @@ -603,16 +607,19 @@ uv run chronicle build-suite packages/irs_soi/table_1_1 \ --require-axiom-validation ``` -The SQLite `ledger.db` is the source of hosted mirrors. To prepare tables for +The SQLite `chronicle.db` is the source of hosted mirrors. To prepare tables for Supabase/Postgres bulk loading, export the DB artifact rather than inserting cells through the Supabase client: ```bash -uv run chronicle export-db-tables --db /tmp/chronicle-suite/ledger.db --out /tmp/chronicle-mirror --replace +uv run chronicle export-db-tables --db /tmp/chronicle-suite/chronicle.db --out /tmp/chronicle-mirror --replace ``` -Accepted build-suite outputs can be published to the private `ledger-derived` R2 -bucket after validation: +Builds produced before the rename wrote `ledger.db`. That name is still read and +published unchanged, so point `--db` at whichever file the build emitted. + +Accepted build-suite outputs can be published to the private derived R2 bucket +after validation: ```bash uv run chronicle publish-derived \ diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index cf898933..e381353f 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -9,17 +9,24 @@ class of Chronicle data belongs. Chronicle uses three storage layers with different jobs. -`ledger-raw` is the immutable source-byte archive. It stores exact publisher +The raw archive is the immutable source-byte store. It holds exact publisher artifacts as fetched: workbooks, CSVs, PDFs, ZIPs, HTML snapshots, and similar government-statistics release files. Raw objects are content-addressed by checksum and should never be overwritten in place. -`ledger-derived` is the reproducible artifact archive. It stores build outputs +The derived archive is the reproducible artifact store. It holds build outputs that Chronicle can regenerate from raw bytes, package specs, parser code, and build configuration. Examples include parsed-cell or parsed-row Parquet/JSONL files, -source record outputs, `ledger.db`, mirror JSONL exports, QA reports, Data +source record outputs, `chronicle.db`, mirror JSONL exports, QA reports, Data Package metadata, and RO-Crate metadata. +Both bucket names are configuration, not constants. The raw archive is +`$CHRONICLE_R2_RAW_BUCKET` and the derived archive is +`$CHRONICLE_R2_DERIVED_BUCKET`; the shipped defaults are still the ledger-era +`ledger-raw` and `ledger-derived`. [Bucket Cutover](#bucket-cutover) records how +those defaults move to `chronicle-raw` and `chronicle-derived` and why the +ledger-era buckets are preserved read-only rather than retired. + Supabase/Postgres is the queryable relational registry for accepted Chronicle builds. It stores rows that applications, agents, and downstream systems need to search and join: source artifacts, source rows/cells, source records, @@ -32,8 +39,8 @@ Hosted tables mirror accepted build outputs and provide a shared query surface. ## Ownership Matrix -| Data class | Git/local package | `ledger-raw` R2 | `ledger-derived` R2 | SQLite `ledger.db` | Supabase/Postgres | -|------------|-------------------|---------------|-------------------|------------------|-------------------| +| Data class | Git/local package | Raw R2 | Derived R2 | SQLite `chronicle.db` | Supabase/Postgres | +|------------|-------------------|--------|------------|-----------------------|-------------------| | Source package specs | Authoritative YAML and parser code | No | Optional packaged snapshot | No | Metadata only | | Raw publisher files | Tiny fixtures only | Authoritative bytes | No | Metadata only | Metadata plus R2 pointer | | Source manifests | Authoritative checked metadata | No | Optional snapshot | Metadata loaded into tables | Queryable artifact registry | @@ -76,7 +83,7 @@ Examples: ```text derived/uk/ons/ons-mye-2024-uk/2024/{build_id}/source_cells.jsonl -derived/nz/ird/ird-working-for-families-statistics-sept-2025/2024/{build_id}/ledger.db +derived/nz/ird/ird-working-for-families-statistics-sept-2025/2024/{build_id}/chronicle.db ``` Legacy US derived keys likewise remain `derived/{source_id}/...`. @@ -114,20 +121,21 @@ The intended flow is: 1. Register raw source artifacts with `uv run chronicle fetch-artifact`, which writes local bytes, records checksums in `manifest.yaml`, and can upload the - exact bytes to `ledger-raw`. Existing manifest-declared artifacts can be + exact bytes to the raw archive. Existing manifest-declared artifacts can be checksum-validated, uploaded, and linked with `uv run chronicle publish-raw`. Production package specs may omit raw bytes from Git as long as the manifest keeps `source_url` and SHA-256 metadata; builds can fill - `LEDGER_SOURCE_ARTIFACT_CACHE_DIR` by setting - `LEDGER_SOURCE_ARTIFACT_FETCH=1`. The old `CHRONICLE_`-prefixed environment - variables remain accepted only as migration fallbacks. + `CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR` by setting + `CHRONICLE_SOURCE_ARTIFACT_FETCH=1`. Ledger-era spellings of both still work; + see [Environment Variable Rename Window](#environment-variable-rename-window). 2. Validate and build a source package with `uv run chronicle validate-package` and `uv run chronicle build-suite`. 3. Produce local deterministic outputs: parsed rows/cells, source records, - aggregate facts, `ledger.db`, QA reports, Data Package metadata, and RO-Crate - metadata. + aggregate facts, `chronicle.db`, QA reports, Data Package metadata, and + RO-Crate metadata. Builds before this rename wrote `ledger.db`; every reader + still accepts that name. 4. Export relational mirror files with `uv run chronicle export-db-tables`. -5. Publish derived build outputs to `ledger-derived`: +5. Publish derived build outputs to the derived archive: ```bash uv run chronicle publish-derived \ @@ -150,6 +158,163 @@ The Supabase project must have the checked migration applied and the `chronicle` schema exposed in PostgREST/Data API settings before the REST loader can write to it. Use `--dry-run` to verify local JSONL files without writing. +## Environment Variable Rename Window + +Every Chronicle setting is read chronicle-first by one shared helper, +`chronicle/env.py`. For a setting `X`, the lookup order is: + +1. `CHRONICLE_X` +2. `POLICYENGINE_LEDGER_X` +3. `LEDGER_X` + +The first name that holds a non-empty value wins. When that name is a ledger-era +one, the process emits a single `ChronicleEnvDeprecationWarning` naming the +`CHRONICLE_`-prefixed variable to set instead. The warning fires once per legacy +name per process, and it subclasses `FutureWarning` rather than +`DeprecationWarning` so it actually reaches operators running the CLI. + +Two consequences are worth stating outright, because both are the reverse of +what a naive fallback would do: + +- The `CHRONICLE_` name wins even when its value reads false. An operator who + has migrated can set `CHRONICLE_SOURCE_ARTIFACT_FETCH=0` and have the flag + turn off, without first hunting down a stale `LEDGER_SOURCE_ARTIFACT_FETCH=1` + somewhere in their profile. +- An empty value counts as unset, so exporting an empty `CHRONICLE_` name does + not mask a set legacy name. + +| Chronicle name | Ledger-era names still accepted | Meaning | +|----------------|--------------------------------|---------| +| `CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR` | `LEDGER_SOURCE_ARTIFACT_CACHE_DIR` | Where fetched raw bytes are cached; defaults to `~/.cache/policyengine-chronicle/source-artifacts` | +| `CHRONICLE_SOURCE_ARTIFACT_FETCH` | `LEDGER_SOURCE_ARTIFACT_FETCH` | Fetch a missing manifest artifact from its `source_url` during a build | +| `CHRONICLE_PE_US_DATA_ROOT` | `LEDGER_PE_US_DATA_ROOT` | Local checkout root for PE US source inventory | +| `CHRONICLE_PE_UK_DATA_ROOT` | `LEDGER_PE_UK_DATA_ROOT` | Local checkout root for PE UK source inventory | +| `CHRONICLE_SCHEMA` | `POLICYENGINE_LEDGER_SCHEMA` | Postgres schema the Supabase client reads and writes | +| `CHRONICLE_R2_RAW_BUCKET` | `LEDGER_R2_RAW_BUCKET` | Raw R2 archive bucket; defaults to `ledger-raw` | +| `CHRONICLE_R2_DERIVED_BUCKET` | `LEDGER_R2_DERIVED_BUCKET` | Derived R2 archive bucket; defaults to `ledger-derived` | + +The two R2 rows are new in this window rather than renamed: those buckets were +hardcoded before, so the ledger-era spellings are accepted for consistency, not +because anything ever set them. + +Variables carrying none of the three prefixes are read literally. This helper +renames the ledger-era surface, not every PolicyEngine variable, so +`POLICYENGINE_SUPABASE_URL`, `POLICYENGINE_SUPABASE_SERVICE_KEY` and +`POLICYENGINE_TARGETS_SCHEMA` keep their names and gain no aliases. + +The hosted schema *value* is a separate migration. `CHRONICLE_SCHEMA` renames +the variable that overrides the schema; the schema still defaults to `ledger`, +and the mirror table names are unchanged. Those move in a later slice +coordinated with the CI writers. + +## Bucket Cutover + +Chronicle's operational stores migrate by dual-run +(PolicyEngine/chronicle#143, mechanism 3): stand up the chronicle-named home, +backfill it, repoint writers, retire the old home. The R2 buckets take one +exception to the last step. Archived witness records pin raw R2 URLs by hash, so +`ledger-raw` and `ledger-derived` are preserved read-only forever rather than +deleted, and manifests keep the `storage.r2` URIs they already recorded as +historical truth. A backfill copies bytes into the new bucket; it never rewrites +where those bytes were first published. `publish-raw` and `fetch-artifact` +enforce that: both refuse to restate a recorded `storage.r2` block under a +different bucket. + +The cutover therefore has one irreversible-looking step that is in fact additive +(creating and filling the new buckets), one cheap reversible step (flipping the +defaults, which is a one-line change in `chronicle/artifacts.py`), and no +deletion step at all. + +### 1. Create the new buckets + +Bucket creation needs a Cloudflare login carrying R2 permissions, so it is an +operator step rather than something CI can do. `wrangler.toml` already pins the +PolicyEngine account (`account_id = "20d90f557651969925eece96e58e24dc"`), so no +`CLOUDFLARE_ACCOUNT_ID` is needed even for a user who belongs to several +accounts: + +```bash +bunx wrangler login +uv run chronicle bootstrap-r2 --raw-bucket chronicle-raw --derived-bucket chronicle-derived +``` + +`bootstrap-r2` verifies authentication with `wrangler whoami` before creating +anything, and creating a bucket that already exists is not an error. + +### 2. Enumerate what has to be copied + +Tracked manifests are the authoritative registry of raw objects. Every one of +them points at `ledger-raw` today: + +```bash +git ls-files '*manifest*.yaml' '*manifest*.yml' \ + | xargs grep -ho 'r2://ledger-raw/[^"'"'"' ]*' | sort -u > /tmp/chronicle-raw-objects.txt +wc -l < /tmp/chronicle-raw-objects.txt +``` + +That is 186 distinct objects at `ff3efd3`, spread over 154 manifest files. Recount +rather than trusting the number: source packages land continuously, and each new +package adds objects. + +The derived bucket needs no backfill. Derived artifacts are reproducible by +definition and are already keyed by `{build_id}`, so a rebuild republishes them +into whichever bucket is configured. + +### 3. Backfill-copy the raw objects + +Keys are content-addressed and identical across buckets, so the copy is a +straight get/put per object: + +```bash +mkdir -p /tmp/chronicle-r2-backfill +while read -r uri; do + key=${uri#r2://ledger-raw/} + dest=/tmp/chronicle-r2-backfill/$key + mkdir -p "$(dirname "$dest")" + bunx wrangler r2 object get "ledger-raw/$key" --file "$dest" --remote + bunx wrangler r2 object put "chronicle-raw/$key" --file "$dest" --remote +done < /tmp/chronicle-raw-objects.txt +``` + +### 4. Verify the copy against the keys themselves + +Every raw key ends `.../{sha256}/{filename}`, so the key is its own checksum +witness and verification needs no manifest lookup: + +```bash +while read -r uri; do + key=${uri#r2://ledger-raw/} + expected=$(printf '%s\n' "$key" | awk -F/ '{print $(NF-1)}') + actual=$(shasum -a 256 "/tmp/chronicle-r2-backfill/$key" | cut -d' ' -f1) + [ "$expected" = "$actual" ] || echo "MISMATCH $key" +done < /tmp/chronicle-raw-objects.txt +``` + +Silence means every downloaded object hashes to the checksum its key claims. +That covers the read from `ledger-raw`; to cover the write to `chronicle-raw`, +re-download each key from the new bucket into a second directory and rerun the +same loop against it. + +### 5. Flip the defaults, in a follow-up PR + +Once the new buckets are filled and verified, change `DEFAULT_R2_RAW_BUCKET` and +`DEFAULT_R2_DERIVED_BUCKET` in `chronicle/artifacts.py` to `chronicle-raw` and +`chronicle-derived`. Until then, operators can opt in per-shell: + +```bash +export CHRONICLE_R2_RAW_BUCKET=chronicle-raw +export CHRONICLE_R2_DERIVED_BUCKET=chronicle-derived +``` + +New raw publications land in the new bucket from that point. Manifests written +before the flip keep pointing at `ledger-raw`, which is why the old bucket stays +readable. + +### 6. Set the ledger-era buckets read-only + +`ledger-raw` and `ledger-derived` keep serving archived witness records after the +flip. They should accept no further writes and should never be deleted. + ## Non-Goals Supabase should not store large raw binary artifacts. It should point to R2. From b3c80ac1704676e74fffab0110238f275f69197d Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:34:42 -0400 Subject: [PATCH 05/42] Pin the boundary guard against the bucket rename The consumer-fact boundary guard now matches derived artifacts on shape, but nothing tested the cases that motivated the change. Both of these pass every check the old literal matching applied, so they would have slipped through silently once the buckets are renamed: - a fact carrying no bucket and no key, only a `r2://chronicle-derived/derived/source/fact.json` URI. The old code compared the URI against two hardcoded prefixes, `r2://ledger-derived/` and `r2://ledger-raw/derived/`, and matched neither. - a `source_file` prefixed `chronicle-derived:`. The old code matched the literal prefix `ledger-derived:`. Co-Authored-By: Claude Fable 5.1 --- tests/test_chronicle_consumer_contract.py | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+) diff --git a/tests/test_chronicle_consumer_contract.py b/tests/test_chronicle_consumer_contract.py index cda97b66..30ac88d5 100644 --- a/tests/test_chronicle_consumer_contract.py +++ b/tests/test_chronicle_consumer_contract.py @@ -886,6 +886,29 @@ def test_export_consumer_facts_cli_rejects_contract_invalid_facts(tmp_path, caps }, "irs_soi.ty2024.table.us.taxable_interest_amount.ledger_derived", ), + # The guard used to match two hardcoded URI prefixes, so a URI naming + # any derived bucket other than `ledger-derived` did not match. Once the + # buckets are renamed (PolicyEngine/chronicle#143, mechanism 3) that is + # every derived URI, so the guard has to match on shape. + ( + { + "source_name": "irs_soi", + "source_file": "publisher.xlsx", + "raw_r2_bucket": None, + "raw_r2_key": None, + "raw_r2_uri": "r2://chronicle-derived/derived/source/fact.json", + }, + "publisher.raw.fact", + ), + ( + { + "source_name": "irs_soi", + "source_file": "chronicle-derived:taxable_interest.json", + "raw_r2_bucket": "ledger-raw", + "raw_r2_uri": "r2://ledger-raw/raw/source/publisher.xlsx", + }, + "publisher.raw.fact", + ), ], ) def test_consumer_contract_rejects_downstream_derived_target_facts( From 5fdea3941021ec79efcd97e1b6a368de5e325c5c Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:29:41 -0400 Subject: [PATCH 06/42] Isolate the rename window for every test, not one module MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The env-isolation fixture lived in tests/test_chronicle_env.py, so only that module ran with CHRONICLE_/POLICYENGINE_LEDGER_/LEDGER_ variables stripped. Tests elsewhere assert the defaults those variables override — the raw and derived bucket names, the Supabase schema — and failed when an operator's shell had them set: CHRONICLE_R2_RAW_BUCKET=zzz CHRONICLE_SCHEMA=zzz uv run pytest -q FAILED tests/test_chronicle_artifacts.py::test_publish_source_artifacts_uploads_manifest_entries The fixture moves to tests/conftest.py as a suite-wide autouse fixture, and tests/test_chronicle_env.py gains an assertion that no rename-window variable reaches a test. db.supabase_client resolves LEDGER_SCHEMA and TARGETS_SCHEMA at import, which happens during collection — before any fixture runs — so the namespace test re-imports the module under the cleared environment instead of asserting on the constant it bound at collection time. Co-Authored-By: Claude Fable 5.1 --- tests/conftest.py | 33 +++++++++++++++++++++++++++++++ tests/test_chronicle_env.py | 33 +++++++++++++++++-------------- tests/test_chronicle_namespace.py | 24 +++++++++++++++------- 3 files changed, 68 insertions(+), 22 deletions(-) create mode 100644 tests/conftest.py diff --git a/tests/conftest.py b/tests/conftest.py new file mode 100644 index 00000000..378fa0b3 --- /dev/null +++ b/tests/conftest.py @@ -0,0 +1,33 @@ +"""Shared fixtures for the Chronicle test suite. + +Chronicle is mid-rename (PolicyEngine/chronicle#143, mechanism 3), so its +settings answer to three prefixes at once: ``CHRONICLE_``, and the ledger-era +``POLICYENGINE_LEDGER_`` and ``LEDGER_``. Any of them can be set in an +operator's shell, and many tests assert the defaults those variables override. +Isolation therefore belongs to the whole suite, not to one module. +""" + +from __future__ import annotations + +import os + +import pytest + +from chronicle.env import ( + CHRONICLE_ENV_PREFIX, + LEGACY_ENV_PREFIXES, + reset_env_deprecation_state, +) + +RENAME_WINDOW_PREFIXES = (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES) + + +@pytest.fixture(autouse=True) +def isolated_rename_window_env(monkeypatch): + """Run every test with no rename-window variable inherited from the shell.""" + for name in list(os.environ): + if name.startswith(RENAME_WINDOW_PREFIXES): + monkeypatch.delenv(name, raising=False) + reset_env_deprecation_state() + yield + reset_env_deprecation_state() diff --git a/tests/test_chronicle_env.py b/tests/test_chronicle_env.py index 4ffe0206..65029d14 100644 --- a/tests/test_chronicle_env.py +++ b/tests/test_chronicle_env.py @@ -2,7 +2,8 @@ Chronicle's operational stores migrate by dual-run (PolicyEngine/chronicle#143, mechanism 3): ``CHRONICLE_*`` names win, ledger-era names keep working behind a -deprecation warning. Every test here is hermetic — the fixture strips every +deprecation warning. Every test here is hermetic — the suite-wide +``isolated_rename_window_env`` fixture in ``tests/conftest.py`` strips every variable in the rename window from the ambient environment first. """ @@ -27,7 +28,6 @@ env_flag, env_names, env_value, - reset_env_deprecation_state, ) from chronicle.harness import main as harness_main from chronicle.source_package import ( @@ -35,19 +35,6 @@ SOURCE_ARTIFACT_FETCH_ENV, ) -RENAME_WINDOW_PREFIXES = (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES) - - -@pytest.fixture(autouse=True) -def isolated_rename_window_env(monkeypatch): - """Run each test with no rename-window variable inherited from the shell.""" - for name in list(os.environ): - if name.startswith(RENAME_WINDOW_PREFIXES): - monkeypatch.delenv(name, raising=False) - reset_env_deprecation_state() - yield - reset_env_deprecation_state() - def _fake_wrangler(tmp_path, log): wrangler = tmp_path / "wrangler" @@ -61,6 +48,22 @@ def _fake_wrangler(tmp_path, log): # --------------------------------------------------------------------------- +def test_every_test_runs_with_the_rename_window_cleared(): + """Isolation is suite-wide (tests/conftest.py), not module-scoped. + + Modules well outside this one assert the defaults these variables override + — the raw and derived bucket names, the Supabase schema — so an operator's + shell must not reach any test. + """ + leaked = sorted( + name + for name in os.environ + if name.startswith((CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES)) + ) + + assert leaked == [] + + def test_env_names_puts_chronicle_first_then_ledger_era_names(): assert env_names("CHRONICLE_SOURCE_ARTIFACT_FETCH") == ( "CHRONICLE_SOURCE_ARTIFACT_FETCH", diff --git a/tests/test_chronicle_namespace.py b/tests/test_chronicle_namespace.py index 9b0ffd41..e8890948 100644 --- a/tests/test_chronicle_namespace.py +++ b/tests/test_chronicle_namespace.py @@ -1,5 +1,7 @@ """Tests for the Chronicle namespace.""" +import importlib + from chronicle.client import get_supabase_client from chronicle.normalization import convert_units from chronicle.targets import ( @@ -8,11 +10,7 @@ query_targets, ) from db.schema import Target as DbTarget -from db.supabase_client import ( - LEDGER_SCHEMA, - TARGETS_SCHEMA, - query_targets as db_query_targets, -) +from db.supabase_client import query_targets as db_query_targets def test_chronicle_targets_reexport_schema_objects(): @@ -29,8 +27,20 @@ def test_chronicle_client_reexports_supabase_client(): def test_chronicle_supabase_schema_boundaries_are_defaulted(): - assert LEDGER_SCHEMA == "ledger" - assert TARGETS_SCHEMA == "targets" + """The schema names are import-time constants, so re-read them here. + + ``db.supabase_client`` resolves them from the environment when it is first + imported, which happens at collection — before the suite-wide + ``isolated_rename_window_env`` fixture clears an operator's + ``CHRONICLE_SCHEMA``. Reloading under the cleared environment is what makes + this a test of the defaults rather than of the shell. + """ + import db.supabase_client + + supabase_client = importlib.reload(db.supabase_client) + + assert supabase_client.LEDGER_SCHEMA == "ledger" + assert supabase_client.TARGETS_SCHEMA == "targets" def test_chronicle_normalization_exports_helpers(): From aad6d69f54ec83bd83bcf69154cb4b601ac4b4de Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:29:53 -0400 Subject: [PATCH 07/42] Refuse to attach a recorded R2 URI to different bytes MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit A raw R2 key is content-addressed, so a recorded storage.r2 block is a claim about specific bytes. fetch-artifact preserved that block whenever one existed and the fetch did not re-upload into the same bucket, then rewrote the entry's sha256 and size_bytes from the newly fetched bytes. When a publisher re-publishes under the same URL and vintage — the IRS did exactly this to the 2022 IRA tables today (PolicyEngine/chronicle#225) — the manifest ended up describing the new bytes while pointing at the old object's key and URI. Verified against this branch's parent: a second fetch of different bytes leaves sha256=109dcf49… in an entry whose storage.r2 key is addressed by c63744a4…, both when the fetch only registers the bytes and when the bucket default has moved. Identity now decides, not the bucket: - the recorded key's last two segments (sha256, filename) are compared with the fetched bytes. Identical: the recorded block is preserved exactly, whichever bucket is configured now. Different: SourceArtifactRevisionError, raised before the cached artifact or its manifest entry is touched, naming recorded and fetched sha256/size_bytes and the ADR rule that the same vintage with new bytes is a new release revision. - --record-revision opts in: the fetched bytes get their own content-addressed key under the configured bucket, never the old key, and the superseded block moves to storage.previous_r2 with its sha256, size_bytes and fetched_at so the earlier bytes stay addressable. - publish-raw applies the same check before treating a recorded block as history, so a local file the recorded object does not hold is refused rather than uploaded. storage.previous_r2 is a sibling of storage.r2: every reader (inventory-artifacts, publish-raw, source_package._artifact_content and the suite's raw-R2-link acceptance check) reads storage.r2 alone, and publish-raw already spreads the existing storage block when it rewrites, so a revision survives publication untouched. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 330 ++++++++++++++++++++++++------ chronicle/harness.py | 57 ++++-- tests/test_chronicle_artifacts.py | 287 ++++++++++++++++++++++++++ 3 files changed, 599 insertions(+), 75 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 6d1e9781..8bf9c7cd 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -53,6 +53,18 @@ def default_r2_derived_bucket() -> str: return env_value(R2_DERIVED_BUCKET_ENV, default=DEFAULT_R2_DERIVED_BUCKET) +class SourceArtifactRevisionError(RuntimeError): + """Fetched bytes are not the bytes the recorded R2 object holds. + + Raw R2 keys are content-addressed, so a recorded ``storage.r2`` block is a + claim about specific bytes. When a publisher re-publishes under the same + URL and vintage, keeping that block would attach its provenance to bytes it + never described. Chronicle refuses instead: same vintage plus new bytes is a + new release revision (docs/adr-chronicle-fact-identity-v2.md), registered + with ``fetch-artifact --record-revision``. + """ + + # New UK and New Zealand uploads are namespaced by country. US objects predate # the country segment and deliberately keep their legacy ``raw/{source_id}`` # and ``derived/{source_id}`` shapes. Publisher directories are the stable @@ -419,11 +431,19 @@ def fetch_source_artifact( table: str | None = None, filename: str | None = None, upload_r2: bool = False, + record_revision: bool = False, r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> ArtifactFetchReport: - """Fetch/register a source artifact and optionally upload it to R2.""" + """Fetch/register a source artifact and optionally upload it to R2. + + ``record_revision`` opts into registering a publisher revision: the fetched + bytes get their own content-addressed key under the configured bucket and + the superseded object moves to ``storage.previous_r2``. Without it, bytes + that disagree with the recorded object raise + :class:`SourceArtifactRevisionError` before anything is overwritten. + """ r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) resolved_r2_prefix = resolve_r2_prefix( @@ -438,14 +458,26 @@ def fetch_source_artifact( if not artifact_filename: raise ValueError("Could not infer artifact filename; pass --filename.") - output.mkdir(parents=True, exist_ok=True) - local_path = output / artifact_filename - local_path.write_bytes(content) - sha256 = hashlib.sha256(content).hexdigest() size_bytes = len(content) manifest_path = output / "manifest.yaml" + # Guard before the cached artifact is touched. A rejected fetch must leave + # the recorded bytes and their manifest entry exactly as they were. + _assert_recorded_object_holds_these_bytes( + manifest_path, + year=year, + filename=artifact_filename, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=r2_bucket, + record_revision=record_revision, + ) + + output.mkdir(parents=True, exist_ok=True) + local_path = output / artifact_filename + local_path.write_bytes(content) + r2_location = ArtifactStorageLocation( provider="r2", bucket=r2_bucket, @@ -484,6 +516,7 @@ def fetch_source_artifact( size_bytes=size_bytes, fetched_at=fetched_at, r2_location=(r2_location if upload_r2 and r2_upload and r2_upload.ok else None), + record_revision=record_revision, ) return ArtifactFetchReport( @@ -1076,17 +1109,181 @@ def _filename_from_url(source_url: str) -> str: return Path(unquote(parsed.path)).name -def _recorded_r2(spec: Any) -> dict[str, Any]: - """Return a manifest file spec's recorded ``storage.r2`` block, if any.""" +def _read_manifest(manifest_path: Path) -> dict[str, Any]: + """Return a manifest's parsed payload, or an empty mapping.""" + if not manifest_path.exists(): + return {} + payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} + return payload if isinstance(payload, dict) else {} + + +def _manifest_file_spec(payload: dict[str, Any], year: Any) -> dict[str, Any]: + """Return one manifest ``files`` entry, or an empty mapping.""" + files = payload.get("files") + if not isinstance(files, dict): + return {} + spec = files.get(year) + return spec if isinstance(spec, dict) else {} + + +def _recorded_storage(spec: Any) -> dict[str, Any]: + """Return a manifest file spec's recorded ``storage`` block, if any.""" if not isinstance(spec, dict): return {} storage = spec.get("storage") - if not isinstance(storage, dict): - return {} - recorded = storage.get("r2") + return storage if isinstance(storage, dict) else {} + + +def _recorded_r2(spec: Any) -> dict[str, Any]: + """Return a manifest file spec's recorded ``storage.r2`` block, if any.""" + recorded = _recorded_storage(spec).get("r2") return recorded if isinstance(recorded, dict) else {} +def _r2_key_identity(key: Any) -> tuple[str, str]: + """Return the ``(sha256, filename)`` a raw R2 key is addressed by. + + Raw keys are ``{prefix}/{source_id}/{package_id}/{year}/{sha256}/{filename}`` + (see :func:`build_r2_key`), so the last two segments say which bytes the + object holds. A key in any other shape yields empty strings and therefore + never matches fetched bytes. + """ + if not isinstance(key, str): + return ("", "") + parts = [part for part in key.split("/") if part] + if len(parts) < 2: + return ("", "") + return (parts[-2], parts[-1]) + + +def _r2_holds_these_bytes( + recorded_r2: dict[str, Any], + *, + sha256: str, + filename: str, +) -> bool: + """Whether a recorded ``storage.r2`` block addresses exactly these bytes.""" + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + return bool(recorded_sha256) and (recorded_sha256, recorded_filename) == ( + sha256, + Path(filename).name, + ) + + +def _revision_error_message( + *, + manifest_path: Path, + year: Any, + filename: str, + recorded_spec: dict[str, Any], + recorded_r2: dict[str, Any], + sha256: str, + size_bytes: int, + r2_bucket: str, +) -> str: + """Explain a refused fetch: recorded identity, fetched identity, next step.""" + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + declared_sha256 = recorded_spec.get("sha256") + recorded_size = ( + recorded_spec.get("size_bytes") if declared_sha256 == recorded_sha256 else None + ) + message = ( + f"{manifest_path} entry {year!r} already records the R2 object " + f"{recorded_r2.get('uri') or recorded_r2.get('key')}, which holds " + f"sha256={recorded_sha256 or 'unknown'} " + f"filename={recorded_filename or 'unknown'} " + f"size_bytes={recorded_size if recorded_size is not None else 'unknown'}. " + f"The fetched bytes are sha256={sha256} filename={Path(filename).name} " + f"size_bytes={size_bytes}. Chronicle will not attach a recorded, " + "content-addressed R2 URI to bytes it does not describe." + ) + if declared_sha256 and declared_sha256 != recorded_sha256: + message += ( + f" (The entry also declares sha256={declared_sha256}, which its own " + "R2 key contradicts: an earlier fetch rewrote the hash without " + "moving the object.)" + ) + return message + ( + " The same vintage with new bytes is a new release revision " + "(docs/adr-chronicle-fact-identity-v2.md). Re-run with " + "--record-revision to store the fetched bytes under their own " + f"content-addressed key in {r2_bucket} and keep the superseded object " + "in storage.previous_r2." + ) + + +def _assert_recorded_object_holds_these_bytes( + manifest_path: Path, + *, + year: Any, + filename: str, + sha256: str, + size_bytes: int, + r2_bucket: str, + record_revision: bool, +) -> None: + """Refuse a publisher revision that has not been opted into.""" + if record_revision: + return + recorded_spec = _manifest_file_spec(_read_manifest(manifest_path), year) + recorded_r2 = _recorded_r2(recorded_spec) + if not recorded_r2: + return + if _r2_holds_these_bytes(recorded_r2, sha256=sha256, filename=filename): + return + raise SourceArtifactRevisionError( + _revision_error_message( + manifest_path=manifest_path, + year=year, + filename=filename, + recorded_spec=recorded_spec, + recorded_r2=recorded_r2, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=r2_bucket, + ) + ) + + +def _superseding_storage( + recorded_spec: dict[str, Any], + *, + new_r2: dict[str, Any] | None, + superseded_at: str, +) -> dict[str, Any]: + """Return a storage block in which the recorded object becomes history. + + ``storage.r2`` only ever names the object that holds the entry's current + bytes. The superseded block is appended, oldest first, to + ``storage.previous_r2`` so the earlier bytes stay addressable by the URI + archived witness records already pin. + """ + storage = dict(_recorded_storage(recorded_spec)) + previous = storage.get("previous_r2") + entries = list(previous) if isinstance(previous, list) else [] + recorded_r2 = _recorded_r2(recorded_spec) + if recorded_r2: + entry = dict(recorded_r2) + recorded_sha256, _recorded_filename = _r2_key_identity(recorded_r2.get("key")) + if recorded_sha256: + entry["sha256"] = recorded_sha256 + if recorded_spec.get("sha256") == recorded_sha256: + # Only carry metadata the superseded key agrees with: a manifest + # can arrive here already describing the new bytes. + for field in ("size_bytes", "fetched_at", "source_url"): + value = recorded_spec.get(field) + if value is not None: + entry[field] = value + entry["superseded_at"] = superseded_at + entries.append(entry) + storage["previous_r2"] = entries + if new_r2 is None: + storage.pop("r2", None) + else: + storage["r2"] = new_r2 + return storage + + def _upsert_manifest( manifest_path: Path, *, @@ -1102,11 +1299,9 @@ def _upsert_manifest( size_bytes: int, fetched_at: str, r2_location: ArtifactStorageLocation | None, + record_revision: bool = False, ) -> None: - if manifest_path.exists(): - payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - else: - payload = {} + payload = _read_manifest(manifest_path) payload.setdefault("source_id", source_id) payload.setdefault("package_id", package_id) payload.setdefault("dataset", dataset) @@ -1120,18 +1315,44 @@ def _upsert_manifest( "size_bytes": size_bytes, "fetched_at": fetched_at, } - recorded_r2 = _recorded_r2(payload["files"].get(year)) + recorded_spec = _manifest_file_spec(payload, year) + recorded_storage = _recorded_storage(recorded_spec) + recorded_r2 = _recorded_r2(recorded_spec) new_r2 = r2_location.to_dict() if r2_location is not None else None - if recorded_r2 and ( - new_r2 is None or recorded_r2.get("bucket") != new_r2.get("bucket") + if recorded_r2 and _r2_holds_these_bytes( + recorded_r2, sha256=sha256, filename=filename ): - # A recorded storage.r2 block is historical truth: archived witness - # records pin raw R2 URLs by hash. Re-fetching under a renamed bucket - # copies bytes; it does not restate where the bytes were first - # published (PolicyEngine/chronicle#143, mechanism 3). - file_entry["storage"] = {"r2": recorded_r2} + # A recorded storage.r2 block for these exact bytes is historical + # truth: archived witness records pin raw R2 URLs by hash. Re-fetching + # under a renamed bucket copies bytes; it does not restate where the + # bytes were first published (PolicyEngine/chronicle#143, mechanism 3). + file_entry["storage"] = {**recorded_storage, "r2": recorded_r2} + elif recorded_r2: + # Different bytes under the same vintage. The guard in + # fetch_source_artifact refuses this without --record-revision; repeat + # the check here so no caller can reach a false-provenance write. + if not record_revision: + raise SourceArtifactRevisionError( + _revision_error_message( + manifest_path=manifest_path, + year=year, + filename=filename, + recorded_spec=recorded_spec, + recorded_r2=recorded_r2, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), + ) + ) + file_entry["storage"] = _superseding_storage( + recorded_spec, + new_r2=new_r2, + superseded_at=fetched_at, + ) elif new_r2 is not None: - file_entry["storage"] = {"r2": new_r2} + file_entry["storage"] = {**recorded_storage, "r2": new_r2} + elif recorded_storage: + file_entry["storage"] = dict(recorded_storage) payload["files"][year] = file_entry manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False), @@ -1192,7 +1413,10 @@ def _publish_raw_manifest_entry( if sha256_expected and sha256_actual != sha256_expected: errors.append("checksum_mismatch") - if errors: + def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: + """Report the entry unpublished, with nothing uploaded or rewritten.""" + if reason is not None: + errors.append(reason) return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), @@ -1210,6 +1434,27 @@ def _publish_raw_manifest_entry( None, ) + if errors: + return refuse() + + recorded_r2 = _recorded_r2(spec) + if recorded_r2 and not _r2_holds_these_bytes( + recorded_r2, sha256=sha256_actual or "", filename=filename + ): + # The recorded object is addressed by different bytes, so it is not + # this file's history. Uploading anyway would either publish under a + # key that misdescribes its content or restate a URI that belongs to + # the superseded bytes. Registering a publisher revision is + # `fetch-artifact --record-revision`, not a publish-time rewrite. + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + return refuse( + "recorded_r2_identity_mismatch:" + f"recorded_sha256={recorded_sha256 or 'unknown'}:" + f"recorded_filename={recorded_filename or 'unknown'}:" + f"local_sha256={sha256_actual}:" + f"local_filename={Path(filename).name}" + ) + location = ArtifactStorageLocation( provider="r2", bucket=r2_bucket, @@ -1223,54 +1468,21 @@ def _publish_raw_manifest_entry( package_path=manifest_path, ), ) - recorded_r2 = _recorded_r2(spec) recorded_bucket = recorded_r2.get("bucket") if recorded_bucket and recorded_bucket != location.bucket: # The recorded bucket is preserved history. Publishing the same bytes # into a renamed bucket is a backfill copy, not a restatement, so the # manifest must not be rewritten to point at the new bucket. - errors.append( + return refuse( "recorded_r2_bucket_is_preserved_history:" f"recorded={recorded_bucket}:requested={location.bucket}" ) - return ( - RawArtifactPublishEntry( - manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, - year=str(year), - filename=filename, - local_path=str(artifact_path), - sha256=sha256_actual, - size_bytes=size_bytes, - r2_location=None, - upload=None, - errors=tuple(errors), - ), - None, - ) recorded_key = recorded_r2.get("key") if recorded_key and recorded_key != location.key: - errors.append( + return refuse( "recorded_r2_key_disagrees_with_country_prefix:" f"recorded={recorded_key}:expected={location.key}" ) - return ( - RawArtifactPublishEntry( - manifest_path=str(manifest_path), - source_id=source_id, - package_id=package_id, - year=str(year), - filename=filename, - local_path=str(artifact_path), - sha256=sha256_actual, - size_bytes=size_bytes, - r2_location=None, - upload=None, - errors=tuple(errors), - ), - None, - ) upload = _upload_r2_object( location, artifact_path, diff --git a/chronicle/harness.py b/chronicle/harness.py index 54c14139..dc9df8ba 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -5,6 +5,7 @@ import argparse import json import shlex +import sys from pathlib import Path from chronicle.artifacts import ( @@ -15,6 +16,7 @@ DerivedArtifactPublishReport, R2BootstrapReport, RawArtifactPublishReport, + SourceArtifactRevisionError, bootstrap_r2_buckets, fetch_source_artifact, inventory_source_artifacts, @@ -337,11 +339,17 @@ def fetch_artifact_file( table: str | None = None, filename: str | None = None, upload_r2: bool = False, + record_revision: bool = False, r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", ) -> ArtifactFetchReport: - """Fetch/register a raw source artifact and optionally upload it to R2.""" + """Fetch/register a raw source artifact and optionally upload it to R2. + + Raises :class:`SourceArtifactRevisionError` when the fetched bytes are not + the bytes the manifest's recorded R2 object holds, unless + ``record_revision`` opts into registering the publisher revision. + """ return fetch_source_artifact( source_url, source_id=source_id, @@ -353,6 +361,7 @@ def fetch_artifact_file( table=table, filename=filename, upload_r2=upload_r2, + record_revision=record_revision, r2_bucket=r2_bucket, r2_prefix=r2_prefix, wrangler_command=wrangler_command, @@ -888,6 +897,17 @@ def main(argv: list[str] | None = None) -> int: action="store_true", help="Upload the artifact to R2 after local checksum capture.", ) + artifact_parser.add_argument( + "--record-revision", + action="store_true", + help=( + "Register a publisher revision: the fetched bytes get their own " + "content-addressed key under the configured bucket and the " + "superseded object moves to storage.previous_r2. Without this " + "flag, bytes that disagree with the recorded R2 object are " + "refused." + ), + ) artifact_parser.add_argument( "--r2-bucket", default=None, @@ -1322,21 +1342,26 @@ def main(argv: list[str] | None = None) -> int: print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if args.command == "fetch-artifact": - report = fetch_artifact_file( - args.url, - source_id=args.source_id, - package_id=args.package_id, - year=args.year, - output_dir=args.out_dir, - dataset=args.dataset, - source_page=args.source_page, - table=args.table, - filename=args.filename, - upload_r2=args.upload_r2, - r2_bucket=args.r2_bucket, - r2_prefix=args.r2_prefix, - wrangler_command=args.wrangler_command, - ) + try: + report = fetch_artifact_file( + args.url, + source_id=args.source_id, + package_id=args.package_id, + year=args.year, + output_dir=args.out_dir, + dataset=args.dataset, + source_page=args.source_page, + table=args.table, + filename=args.filename, + upload_r2=args.upload_r2, + record_revision=args.record_revision, + r2_bucket=args.r2_bucket, + r2_prefix=args.r2_prefix, + wrangler_command=args.wrangler_command, + ) + except SourceArtifactRevisionError as error: + print(f"error: {error}", file=sys.stderr) + return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 if args.command == "inventory-artifacts": diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index a96d6c25..7346584c 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -11,6 +11,7 @@ from chronicle.cli import main as cli_main from chronicle.artifacts import ( + SourceArtifactRevisionError, build_artifact_key, build_artifact_rows, build_derived_r2_key, @@ -827,3 +828,289 @@ def test_fetch_artifact_keeps_an_already_recorded_bucket(tmp_path, monkeypatch): assert ( second["files"][2023]["storage"]["r2"] == first["files"][2023]["storage"]["r2"] ) + + +# --------------------------------------------------------------------------- +# Publisher revisions +# +# A raw R2 key is content-addressed, so a recorded storage.r2 block is a claim +# about specific bytes. On 2026-09-02 the IRS re-published 22in05ira.xlsx and +# 22in06ira.xlsx under their existing URLs (PolicyEngine/chronicle#225): a +# repeated fetch must never pair those new bytes with the old object's URI. +# --------------------------------------------------------------------------- + +REPUBLISHED_URL = "https://www.irs.gov/pub/irs-soi/22in05ira.xlsx" +REPUBLISHED_FILENAME = "22in05ira.xlsx" +FIRST_PUBLICATION = b"IRA table 5, first publication" +SECOND_PUBLICATION = b"IRA table 5, silently re-published with revised rows" + + +def _wrangler_stub(tmp_path, log): + wrangler = tmp_path / "wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") + wrangler.chmod(0o755) + return wrangler + + +def _serve(monkeypatch, content): + """Serve ``content`` from the publisher URL, without touching the network.""" + + def _fake_read_artifact(source_url): + assert source_url == REPUBLISHED_URL + return content, REPUBLISHED_FILENAME + + monkeypatch.setattr("chronicle.artifacts._read_artifact", _fake_read_artifact) + + +def _fetch_republished(output_dir, wrangler, *, upload_r2=True, **kwargs): + return fetch_source_artifact( + REPUBLISHED_URL, + source_id="irs_soi", + package_id="soi-table-5", + year=2022, + output_dir=output_dir, + upload_r2=upload_r2, + wrangler_command=str(wrangler), + **kwargs, + ) + + +def test_repeated_fetch_of_identical_bytes_preserves_the_recorded_block( + tmp_path, monkeypatch +): + """Same bytes: the recorded block survives whatever bucket is configured.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + _fetch_republished(output_dir, wrangler) + first = yaml.safe_load(manifest_path.read_text())["files"][2022] + + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + report = _fetch_republished(output_dir, wrangler) + second = yaml.safe_load(manifest_path.read_text())["files"][2022] + + assert report.valid + # The backfill copy really goes to the renamed bucket, but the manifest + # keeps recording where these bytes were first published. + assert report.r2_location.bucket == "chronicle-raw" + assert "chronicle-raw" in log.read_text() + assert second["storage"] == first["storage"] + assert second["storage"]["r2"]["bucket"] == "ledger-raw" + assert "previous_r2" not in second["storage"] + assert second["sha256"] == first["sha256"] + + +@pytest.mark.parametrize( + ("upload_r2", "configured_bucket"), + [ + pytest.param(True, None, id="reuploaded"), + # The two routes that reached a manifest in the wild: a fetch that only + # registers the bytes, and a fetch once the bucket default has moved. + # Both preserved the recorded block while rewriting sha256/size_bytes. + pytest.param(False, None, id="registered-without-upload"), + pytest.param(True, "chronicle-raw", id="after-the-bucket-rename"), + ], +) +def test_repeated_fetch_of_different_bytes_is_refused( + tmp_path, monkeypatch, upload_r2, configured_bucket +): + """A publisher revision must not inherit the recorded object's provenance.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + artifact_path = output_dir / REPUBLISHED_FILENAME + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + manifest_before = manifest_path.read_bytes() + uploads_before = log.read_text() + + if configured_bucket: + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", configured_bucket) + _serve(monkeypatch, SECOND_PUBLICATION) + with pytest.raises(SourceArtifactRevisionError) as raised: + _fetch_republished(output_dir, wrangler, upload_r2=upload_r2) + + message = str(raised.value) + assert first_report.sha256 in message + assert hashlib.sha256(SECOND_PUBLICATION).hexdigest() in message + assert f"size_bytes={len(FIRST_PUBLICATION)}" in message + assert f"size_bytes={len(SECOND_PUBLICATION)}" in message + assert "release revision" in message + assert "--record-revision" in message + # Nothing was overwritten, copied or uploaded on the way to the refusal. + assert manifest_path.read_bytes() == manifest_before + assert artifact_path.read_bytes() == FIRST_PUBLICATION + assert log.read_text() == uploads_before + + +def test_record_revision_writes_a_new_key_and_keeps_the_previous_object( + tmp_path, monkeypatch +): + """The opt-in records the new bytes' own key under the configured bucket.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + superseded = yaml.safe_load(manifest_path.read_text())["files"][2022] + + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + _serve(monkeypatch, SECOND_PUBLICATION) + report = _fetch_republished(output_dir, wrangler, record_revision=True) + revised = yaml.safe_load(manifest_path.read_text())["files"][2022] + revised_sha256 = hashlib.sha256(SECOND_PUBLICATION).hexdigest() + + assert report.valid + # storage.r2 names the object that holds the entry's current bytes... + assert revised["sha256"] == revised_sha256 + assert revised["size_bytes"] == len(SECOND_PUBLICATION) + assert revised["storage"]["r2"]["bucket"] == "chronicle-raw" + assert revised["storage"]["r2"]["key"] == ( + f"raw/irs_soi/soi-table-5/2022/{revised_sha256}/{REPUBLISHED_FILENAME}" + ) + assert revised["storage"]["r2"]["uri"] == ( + f"r2://chronicle-raw/{revised['storage']['r2']['key']}" + ) + # ...and never the superseded key, which stays addressable as history. + previous = revised["storage"]["previous_r2"] + assert [entry["uri"] for entry in previous] == [superseded["storage"]["r2"]["uri"]] + assert previous[0]["bucket"] == "ledger-raw" + assert previous[0]["sha256"] == first_report.sha256 + assert previous[0]["size_bytes"] == len(FIRST_PUBLICATION) + assert previous[0]["fetched_at"] == superseded["fetched_at"] + assert previous[0]["superseded_at"] == revised["fetched_at"] + assert (output_dir / REPUBLISHED_FILENAME).read_bytes() == SECOND_PUBLICATION + assert f"chronicle-raw/{revised['storage']['r2']['key']}" in log.read_text() + + +def test_a_revised_manifest_still_reads_as_one_r2_linked_artifact( + tmp_path, monkeypatch +): + """storage.previous_r2 is a sibling key, so every storage.r2 reader is intact.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + wrangler = _wrangler_stub(tmp_path, tmp_path / "wrangler.log") + _serve(monkeypatch, FIRST_PUBLICATION) + _fetch_republished(output_dir, wrangler) + _serve(monkeypatch, SECOND_PUBLICATION) + _fetch_republished(output_dir, wrangler, record_revision=True) + + inventory = inventory_source_artifacts(output_dir) + + assert inventory.valid + assert inventory.counts["r2_link_count"] == 1 + assert inventory.counts["checksum_mismatch_count"] == 0 + assert inventory.entries[0].r2["bucket"] == "ledger-raw" + assert inventory.entries[0].sha256_actual == ( + hashlib.sha256(SECOND_PUBLICATION).hexdigest() + ) + + +def test_publish_raw_refuses_a_file_the_recorded_object_does_not_hold( + tmp_path, monkeypatch +): + """Recorded sha256 != local sha256 is a revision, not a backfill.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + + # Reproduce the state a pre-fix fetch left behind: new bytes on disk, the + # entry's own hash rewritten, the recorded key still addressing the old + # bytes. + revised_sha256 = hashlib.sha256(SECOND_PUBLICATION).hexdigest() + (output_dir / REPUBLISHED_FILENAME).write_bytes(SECOND_PUBLICATION) + manifest = yaml.safe_load(manifest_path.read_text()) + manifest["files"][2022]["sha256"] = revised_sha256 + manifest["files"][2022]["size_bytes"] = len(SECOND_PUBLICATION) + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + manifest_before = manifest_path.read_bytes() + uploads_before = log.read_text() + + report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) + + assert not report.valid + assert report.entries[0].upload is None + assert report.entries[0].r2_location is None + assert report.entries[0].errors == ( + "recorded_r2_identity_mismatch:" + f"recorded_sha256={first_report.sha256}:" + f"recorded_filename={REPUBLISHED_FILENAME}:" + f"local_sha256={revised_sha256}:" + f"local_filename={REPUBLISHED_FILENAME}", + ) + assert log.read_text() == uploads_before + assert manifest_path.read_bytes() == manifest_before + + +def test_publish_raw_uploads_a_registered_revision_and_keeps_its_history( + tmp_path, monkeypatch +): + """Once the revision is registered, publishing it is ordinary work.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + _serve(monkeypatch, SECOND_PUBLICATION) + _fetch_republished(output_dir, wrangler, record_revision=True) + revised_sha256 = hashlib.sha256(SECOND_PUBLICATION).hexdigest() + + report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) + published = yaml.safe_load(manifest_path.read_text())["files"][2022] + + assert report.valid + assert published["storage"]["r2"]["key"].endswith( + f"/{revised_sha256}/{REPUBLISHED_FILENAME}" + ) + assert [entry["sha256"] for entry in published["storage"]["previous_r2"]] == [ + first_report.sha256 + ] + + +def test_fetch_artifact_cli_refuses_a_revision_then_records_it_on_request( + tmp_path, monkeypatch, capsys +): + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + wrangler = _wrangler_stub(tmp_path, tmp_path / "wrangler.log") + argv = [ + "fetch-artifact", + "--url", + REPUBLISHED_URL, + "--source-id", + "irs_soi", + "--package-id", + "soi-table-5", + "--year", + "2022", + "--out-dir", + str(output_dir), + "--upload-r2", + "--wrangler-command", + str(wrangler), + ] + _serve(monkeypatch, FIRST_PUBLICATION) + assert harness_main(argv) == 0 + capsys.readouterr() + + _serve(monkeypatch, SECOND_PUBLICATION) + refused = harness_main(argv) + refusal = capsys.readouterr() + + assert refused == 1 + assert "--record-revision" in refusal.err + assert refusal.out == "" + + assert harness_main([*argv, "--record-revision"]) == 0 + recorded = json.loads(capsys.readouterr().out) + + assert recorded["sha256"] == hashlib.sha256(SECOND_PUBLICATION).hexdigest() + assert recorded["r2_location"]["key"].endswith( + f"/{recorded['sha256']}/{REPUBLISHED_FILENAME}" + ) From f0265b33d75eb11215f48a95ea98d94694a2d04f Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:30:52 -0400 Subject: [PATCH 08/42] Document publisher revisions alongside the bucket rename The rename-window text said a recorded storage.r2 block is preserved because the bucket may have moved. That is the weaker half of the rule: the block is preserved because it addresses the bytes in hand. Adds a Publisher Revisions section covering the identity check, the refusal, --record-revision, the storage.previous_r2 shape and why every existing storage.r2 reader is unaffected, and points the bucket-cutover, publish-flow and harness passages at it. Co-Authored-By: Claude Fable 5.1 --- README.md | 7 +++ docs/agent-source-package-harness.md | 9 ++++ docs/storage-architecture.md | 73 ++++++++++++++++++++++++++-- 3 files changed, 86 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index e0ade235..6b6e170c 100644 --- a/README.md +++ b/README.md @@ -367,6 +367,13 @@ uv run chronicle fetch-artifact \ --table "Publication 1304 Table 1.2" \ --upload-r2 +# Re-fetching is safe: identical bytes keep the recorded storage.r2 block, and +# bytes that disagree with it are refused. When a publisher has re-published +# under the same URL and vintage, register the revision explicitly — the new +# bytes get their own content-addressed key and the superseded object is kept +# in storage.previous_r2: +uv run chronicle fetch-artifact ... --record-revision + # Audit local manifests and checksums: uv run chronicle inventory-artifacts --root db/data diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index afeb98f4..11200ebb 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -22,6 +22,15 @@ For already-downloaded manifest artifacts, agents should run `uv run chronicle publish-raw --root db/data` to upload checksum-verified bytes to R2 and write `storage.r2` metadata back into each manifest entry. +Both commands treat a recorded `storage.r2` block as a claim about specific +bytes, because raw keys are content-addressed. Re-fetching or publishing bytes +the recorded object does not hold is refused; when a publisher has re-published +under the same URL and vintage, register the revision with +`uv run chronicle fetch-artifact ... --record-revision`, which stores the new +bytes under their own key and keeps the superseded object in +`storage.previous_r2`. See +[Publisher Revisions](storage-architecture.md#publisher-revisions). + Builds do not require production raw bytes to be committed to Git. Source packages first read packaged fixture bytes, then `CHRONICLE_SOURCE_ARTIFACT_CACHE_DIR` (defaulting to diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index e381353f..6f4cfb00 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -91,6 +91,68 @@ Legacy US derived keys likewise remain `derived/{source_id}/...`. Derived artifacts are reproducible and may be replaced by a new build, but a specific `{build_id}` path should be immutable once published. +## Publisher Revisions + +A raw key embeds the sha256 of the bytes it holds, so a manifest's recorded +`storage.r2` block is a claim about specific bytes, not a pointer to a file +name. Publishers do not always honor that: on 2026-09-02 the IRS re-published +`22in05ira.xlsx` and `22in06ira.xlsx` under their existing URLs +(PolicyEngine/chronicle#225). + +`fetch-artifact` therefore compares the recorded key's `{sha256}/{filename}` +tail with the bytes it just fetched, before it writes anything: + +- **Identical** — the recorded block is preserved exactly, whichever bucket is + configured now. Re-fetching after the bucket rename copies bytes; it does not + restate where they were first published. +- **Different** — the fetch is refused. Nothing is overwritten: not the cached + artifact, not the manifest entry, and no object is uploaded. The error names + the recorded and the fetched `sha256`/`size_bytes`. Per + `docs/adr-chronicle-fact-identity-v2.md`, the same vintage with new bytes is a + new release revision, so registering it is a decision an operator makes, not a + silent rewrite. + +`--record-revision` makes that decision explicit. The fetched bytes get their +own content-addressed key under the configured bucket — never the recorded key — +and the superseded block moves to `storage.previous_r2`: + +```yaml +files: + 2022: + filename: 22in05ira.xlsx + sha256: + size_bytes: + fetched_at: "2026-09-02T17:04:11+00:00" + storage: + r2: + provider: r2 + bucket: chronicle-raw + key: raw/irs_soi/soi-table-5/2022//22in05ira.xlsx + uri: r2://chronicle-raw/raw/irs_soi/soi-table-5/2022//22in05ira.xlsx + previous_r2: + - provider: r2 + bucket: ledger-raw + key: raw/irs_soi/soi-table-5/2022//22in05ira.xlsx + uri: r2://ledger-raw/raw/irs_soi/soi-table-5/2022//22in05ira.xlsx + sha256: + size_bytes: + fetched_at: "2026-06-11T14:22:05+00:00" + superseded_at: "2026-09-02T17:04:11+00:00" +``` + +`storage.r2` only ever names the object that holds the entry's current bytes, +and `previous_r2` lists superseded objects oldest first, so the bytes an +archived witness record pinned stay addressable at the URI it pinned. Every +reader — `inventory-artifacts`, `publish-raw`, source-package artifact loading, +and the suite's raw-R2-link acceptance check — reads `storage.r2` alone, so a +revised entry reads exactly like an unrevised one; `publish-raw` preserves the +rest of the `storage` block when it writes back. + +`publish-raw` applies the same identity check before treating a recorded block +as history. A local file the recorded object does not hold is reported as +`recorded_r2_identity_mismatch` and nothing is uploaded: registering a revision +is a fetch-time decision, not a publish-time rewrite. + ## Relational Registry Contract The hosted `chronicle` schema should be the lookup surface for Chronicle, not the place @@ -121,7 +183,10 @@ The intended flow is: 1. Register raw source artifacts with `uv run chronicle fetch-artifact`, which writes local bytes, records checksums in `manifest.yaml`, and can upload the - exact bytes to the raw archive. Existing manifest-declared artifacts can be + exact bytes to the raw archive. Re-fetching an entry whose bytes the + publisher has changed is refused unless the revision is registered with + `--record-revision`; see [Publisher Revisions](#publisher-revisions). + Existing manifest-declared artifacts can be checksum-validated, uploaded, and linked with `uv run chronicle publish-raw`. Production package specs may omit raw bytes from Git as long as the manifest keeps `source_url` and SHA-256 metadata; builds can fill @@ -217,8 +282,10 @@ exception to the last step. Archived witness records pin raw R2 URLs by hash, so deleted, and manifests keep the `storage.r2` URIs they already recorded as historical truth. A backfill copies bytes into the new bucket; it never rewrites where those bytes were first published. `publish-raw` and `fetch-artifact` -enforce that: both refuse to restate a recorded `storage.r2` block under a -different bucket. +enforce that: a recorded block that addresses the bytes in hand is preserved +whichever bucket is configured, and `publish-raw` refuses to restate it under a +different one. Bytes that the recorded object does not hold are not that +object's history at all; see [Publisher Revisions](#publisher-revisions). The cutover therefore has one irreversible-looking step that is in fact additive (creating and filling the new buckets), one cheap reversible step (flipping the From a4bfec6b81ade17b333311c164d432c325becd06 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:32:09 -0400 Subject: [PATCH 09/42] Record the gate round-1 fixes in PROGRESS.md Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 40 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 40 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index da5d13e5..27b1babe 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -48,6 +48,42 @@ Lane C5's handoff notes previously lived here; its durable record is `README.md` follow. Verified 186 distinct `ledger-raw` objects across 154 tracked manifest files, every key content-addressed by sha256. +## Review fixes (gate round 1) + +The Fable+Sol gate requested changes; both findings are applied on this branch. + +- **[high] `fetch-artifact` could attach a recorded R2 URI to new bytes.** The + preserve rule keyed on the bucket, so a repeated fetch that did not re-upload + into the same bucket kept the recorded `storage.r2` block while rewriting the + entry's `sha256`/`size_bytes`. Reproduced against this branch's parent: the + entry ends up declaring `109dcf49…` with a key addressed by `c63744a4…`, both + when the fetch only registers the bytes and when the bucket default has moved. + The rule now keys on identity — the recorded key's `{sha256}/{filename}` tail + against the fetched bytes. Identical preserves the block exactly; different + raises `SourceArtifactRevisionError` before the cached artifact or its + manifest entry is touched, naming recorded and fetched `sha256`/`size_bytes` + and the ADR rule that same vintage plus new bytes is a new release revision. + `--record-revision` opts in: the new bytes get their own content-addressed key + under the configured bucket, never the old key, and the superseded block moves + to `storage.previous_r2`. `publish-raw` applies the same check before treating + a recorded block as history (`recorded_r2_identity_mismatch`, nothing + uploaded). +- **[low] Env isolation was scoped to one module.** The autouse fixture moved to + `tests/conftest.py` and now clears all three prefixes for every test. + `db.supabase_client` resolves `LEDGER_SCHEMA` at import — during collection, + before any fixture — so `tests/test_chronicle_namespace.py` re-imports it + under the cleared environment instead of asserting the constant it bound at + collection time. + +`storage.previous_r2` is a sibling key, chosen because every reader +(`inventory-artifacts`, `publish-raw`, `source_package._artifact_content`, the +suite's raw-R2-link acceptance check) reads `storage.r2` alone, and +`publish-raw` already spreads the rest of the `storage` block when it writes +back, so a revision survives publication untouched. All 180 tracked manifest +entries that carry a `storage.r2` block are content-addressed and agree with +their declared `sha256` and `filename`, so the identity check never fires on +tracked data. + ## Verification - `uv run pytest -q`: green. @@ -56,6 +92,10 @@ Lane C5's handoff notes previously lived here; its durable record is files are unformatted on `main` already and are byte-identical here; CI runs `ruff check` only, so they are pre-existing and out of scope. - CI's db CLI gate (`chronicle init` / `load all` / `stats`): passes. +- `CHRONICLE_R2_RAW_BUCKET=zzz CHRONICLE_SCHEMA=zzz uv run pytest -q`: green. + Before the shared fixture it failed five tests — four bucket-default + assertions in `tests/test_chronicle_artifacts.py` and the collection-time + schema constant in `tests/test_chronicle_namespace.py`. ## Next From 7e9a58d39efd84e3089c7d63f6d872ede3d23372 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:33:18 -0400 Subject: [PATCH 10/42] Read the recorded object's identity from its URI too A storage.r2 block written by Chronicle always carries both key and uri, but a hand-written one need not. The identity read now falls back to the uri, whose last two segments are the same {sha256}/{filename}, so a block that records only where the object lives is still recognized as history for the bytes it holds instead of being refused as a revision. Also covers registering a revision without an upload: the entry is left with no storage.r2 at all rather than a pointer to bytes R2 does not hold, the superseded object stays in storage.previous_r2, and publish-raw completes the registration. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 22 +++++----- tests/test_chronicle_artifacts.py | 69 +++++++++++++++++++++++++++++++ 2 files changed, 81 insertions(+), 10 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 8bf9c7cd..9421ab08 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1140,17 +1140,19 @@ def _recorded_r2(spec: Any) -> dict[str, Any]: return recorded if isinstance(recorded, dict) else {} -def _r2_key_identity(key: Any) -> tuple[str, str]: - """Return the ``(sha256, filename)`` a raw R2 key is addressed by. +def _r2_key_identity(recorded_r2: dict[str, Any]) -> tuple[str, str]: + """Return the ``(sha256, filename)`` a recorded R2 object is addressed by. Raw keys are ``{prefix}/{source_id}/{package_id}/{year}/{sha256}/{filename}`` (see :func:`build_r2_key`), so the last two segments say which bytes the - object holds. A key in any other shape yields empty strings and therefore - never matches fetched bytes. + object holds; the URI ends in the same two segments and stands in for a + block that records only that. A locator in any other shape yields empty + strings and therefore never matches fetched bytes. """ - if not isinstance(key, str): + locator = recorded_r2.get("key") or recorded_r2.get("uri") + if not isinstance(locator, str): return ("", "") - parts = [part for part in key.split("/") if part] + parts = [part for part in locator.split("/") if part] if len(parts) < 2: return ("", "") return (parts[-2], parts[-1]) @@ -1163,7 +1165,7 @@ def _r2_holds_these_bytes( filename: str, ) -> bool: """Whether a recorded ``storage.r2`` block addresses exactly these bytes.""" - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) return bool(recorded_sha256) and (recorded_sha256, recorded_filename) == ( sha256, Path(filename).name, @@ -1182,7 +1184,7 @@ def _revision_error_message( r2_bucket: str, ) -> str: """Explain a refused fetch: recorded identity, fetched identity, next step.""" - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) declared_sha256 = recorded_spec.get("sha256") recorded_size = ( recorded_spec.get("size_bytes") if declared_sha256 == recorded_sha256 else None @@ -1264,7 +1266,7 @@ def _superseding_storage( recorded_r2 = _recorded_r2(recorded_spec) if recorded_r2: entry = dict(recorded_r2) - recorded_sha256, _recorded_filename = _r2_key_identity(recorded_r2.get("key")) + recorded_sha256, _recorded_filename = _r2_key_identity(recorded_r2) if recorded_sha256: entry["sha256"] = recorded_sha256 if recorded_spec.get("sha256") == recorded_sha256: @@ -1446,7 +1448,7 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: # key that misdescribes its content or restate a URI that belongs to # the superseded bytes. Registering a publisher revision is # `fetch-artifact --record-revision`, not a publish-time rewrite. - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2.get("key")) + recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) return refuse( "recorded_r2_identity_mismatch:" f"recorded_sha256={recorded_sha256 or 'unknown'}:" diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 7346584c..329ac1cf 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1114,3 +1114,72 @@ def test_fetch_artifact_cli_refuses_a_revision_then_records_it_on_request( assert recorded["r2_location"]["key"].endswith( f"/{recorded['sha256']}/{REPUBLISHED_FILENAME}" ) + + +def test_record_revision_without_an_upload_records_no_current_object( + tmp_path, monkeypatch +): + """An offline revision keeps history without claiming the new bytes exist. + + Registering a revision without ``--upload-r2`` leaves the entry with no + ``storage.r2`` at all rather than a pointer to bytes R2 does not hold. The + superseded object stays addressable, and a later publish-raw completes the + registration. + """ + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + first_report = _fetch_republished(output_dir, wrangler) + + _serve(monkeypatch, SECOND_PUBLICATION) + _fetch_republished(output_dir, wrangler, upload_r2=False, record_revision=True) + registered = yaml.safe_load(manifest_path.read_text())["files"][2022] + + assert "r2" not in registered["storage"] + assert [entry["sha256"] for entry in registered["storage"]["previous_r2"]] == [ + first_report.sha256 + ] + + report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) + published = yaml.safe_load(manifest_path.read_text())["files"][2022] + revised_sha256 = hashlib.sha256(SECOND_PUBLICATION).hexdigest() + + assert report.valid + assert published["storage"]["r2"]["key"].endswith( + f"/{revised_sha256}/{REPUBLISHED_FILENAME}" + ) + assert [entry["sha256"] for entry in published["storage"]["previous_r2"]] == [ + first_report.sha256 + ] + + +def test_a_recorded_block_that_only_carries_a_uri_is_still_recognized( + tmp_path, monkeypatch +): + """Identity reads the URI when a hand-written block records no key.""" + output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + wrangler = _wrangler_stub(tmp_path, tmp_path / "wrangler.log") + manifest_path = output_dir / "manifest.yaml" + _serve(monkeypatch, FIRST_PUBLICATION) + _fetch_republished(output_dir, wrangler) + manifest = yaml.safe_load(manifest_path.read_text()) + recorded = manifest["files"][2022]["storage"]["r2"] + manifest["files"][2022]["storage"]["r2"] = { + "provider": recorded["provider"], + "bucket": recorded["bucket"], + "uri": recorded["uri"], + } + manifest_path.write_text(yaml.safe_dump(manifest, sort_keys=False)) + uri_only = manifest["files"][2022]["storage"]["r2"] + + report = _fetch_republished(output_dir, wrangler) + preserved = yaml.safe_load(manifest_path.read_text())["files"][2022] + + assert report.valid + assert preserved["storage"]["r2"] == uri_only + + _serve(monkeypatch, SECOND_PUBLICATION) + with pytest.raises(SourceArtifactRevisionError): + _fetch_republished(output_dir, wrangler) From 6cb181b5fb43e241ea0c32ee6ac672af66c55241 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:35:35 -0400 Subject: [PATCH 11/42] Pin the preserved block's field order too Dict equality would pass on a block that was rebuilt with the same values in a different order, which is not what "preserved" means for a manifest that gets dumped back to YAML. Co-Authored-By: Claude Fable 5.1 --- tests/test_chronicle_artifacts.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 329ac1cf..93b4d429 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -897,6 +897,8 @@ def test_repeated_fetch_of_identical_bytes_preserves_the_recorded_block( assert report.r2_location.bucket == "chronicle-raw" assert "chronicle-raw" in log.read_text() assert second["storage"] == first["storage"] + # Field order too, so the block is byte-for-byte identical once dumped. + assert list(second["storage"]["r2"].items()) == list(first["storage"]["r2"].items()) assert second["storage"]["r2"]["bucket"] == "ledger-raw" assert "previous_r2" not in second["storage"] assert second["sha256"] == first["sha256"] From a9591fcd6704d369a0153ffd7ba000b09cc2e724 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:37:48 -0400 Subject: [PATCH 12/42] Say what the reproduction actually ran The hashes named the fix lane's two-body fixture, not IRS bytes. Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/PROGRESS.md b/PROGRESS.md index 27b1babe..38e669e2 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -55,9 +55,11 @@ The Fable+Sol gate requested changes; both findings are applied on this branch. - **[high] `fetch-artifact` could attach a recorded R2 URI to new bytes.** The preserve rule keyed on the bucket, so a repeated fetch that did not re-upload into the same bucket kept the recorded `storage.r2` block while rewriting the - entry's `sha256`/`size_bytes`. Reproduced against this branch's parent: the - entry ends up declaring `109dcf49…` with a key addressed by `c63744a4…`, both - when the fetch only registers the bytes and when the bucket default has moved. + entry's `sha256`/`size_bytes`. Reproduced against this branch's parent by + serving two different bodies from one URL: the entry ends up declaring the + fetched bytes' `sha256` under a key addressed by the superseded bytes' one, + both when the fetch only registers the bytes and when the bucket default has + moved. The rule now keys on identity — the recorded key's `{sha256}/{filename}` tail against the fetched bytes. Identical preserves the block exactly; different raises `SourceArtifactRevisionError` before the cached artifact or its From 7101fb10d9acc3549cefc35ab44adc890c4b0c27 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:09:03 -0400 Subject: [PATCH 13/42] Record the round-2 gate findings and the corpus scan The seven findings, in the order the fixes depend on each other, plus what a scan of every tracked manifest says about how strict the locator check can be: 187 entries, all content-addressed, no contradictory field. Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 33 +++++++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index 38e669e2..e51ea2ca 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -86,6 +86,39 @@ entries that carry a `storage.r2` block are content-addressed and agree with their declared `sha256` and `filename`, so the identity check never fires on tracked data. +## Review fixes (gate round 2) + +The second Fable+Sol gate requested changes again. Seven findings, each fixed +with a regression test on this branch. Plan, in dependency order: + +1. **[high] `CHRONICLE_SCHEMA` does not reach the Supabase mirror writer.** + `chronicle/harness.py` and `chronicle/mirror.py` default the schema to the + literal `"ledger"`; only `db.supabase_client` reads the renamed variable. + Resolve through the shared helper whenever no explicit `--schema` is given. +2. **[high] The derived-fact boundary check is not rename-safe.** + `chronicle/consumer_contract.py` matches the `.ledger_derived` suffix only. +3. **[high] `fetch-artifact` cannot address a package's non-default manifest.** + Seven tracked packages keep a `manifest_*_source_package.yaml`; three + directories keep two. A fetch into one of them writes a third manifest and + never sees the recorded block. +4. **[high] Revision protection vanishes when the entry has no `storage.r2`.** +5. **[medium] Recorded-R2 locator fields must be cross-checked**, not read as + key-or-URI, before a block is preserved or published. +6. **[medium] `_read_manifest` must reject a malformed document**, not treat a + non-mapping YAML payload as an absent manifest. +7. **[low] Schema resolution must be lazy** so no legacy variable is read at + collection, before the autouse isolation fixture runs. + +## State (round 2) + +- Read both gate rounds on PolicyEngine/chronicle#226 and the code each finding + names. +- Scanned all 154 tracked manifest files (187 `files` entries, every one + carrying `storage.r2`): every recorded block supplies provider, bucket, key + and uri; every key is content-addressed; every declared `sha256`/`filename` + agrees with its key tail; no `uri` contradicts its `key`. Strict locator + validation therefore refuses nothing that is tracked today. + ## Verification - `uv run pytest -q`: green. From b46bee13f7acae2e608007a0cbb3007431b18cc8 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:09:55 -0400 Subject: [PATCH 14/42] Reject the chronicle spelling of the derived-row marker The consumer boundary matched the `.ledger_derived` suffix literally, so a producer that renamed its derived rows with everything else would have walked a downstream target fact straight through the guard. Match the whole final dot-segment against both spellings instead. Without the fix the two new cases fail: the chronicle-spelled record id validates clean. Co-Authored-By: Claude Fable 5.1 --- chronicle/consumer_contract.py | 16 ++++++- tests/test_chronicle_consumer_contract.py | 52 +++++++++++++++++++++++ 2 files changed, 67 insertions(+), 1 deletion(-) diff --git a/chronicle/consumer_contract.py b/chronicle/consumer_contract.py index f9c93f3a..17315a13 100644 --- a/chronicle/consumer_contract.py +++ b/chronicle/consumer_contract.py @@ -476,6 +476,20 @@ def _r2_uri_parts(uri: str) -> tuple[str, str]: return bucket, key +# The marker a downstream target row carries in its source_record_id. It moves +# with the rename window: producers write `ledger_derived` today and +# `chronicle_derived` once they migrate (PolicyEngine/chronicle#143, mechanism +# 3), so the boundary has to reject both spellings identically or the guard +# stops firing the moment a producer renames. +DERIVED_SOURCE_RECORD_SUFFIXES = frozenset({"ledger_derived", "chronicle_derived"}) + + +def _is_derived_source_record_id(source_record_id: str) -> bool: + """Whether a source_record_id marks a downstream derived target row.""" + _, separator, suffix = source_record_id.rpartition(".") + return bool(separator) and suffix in DERIVED_SOURCE_RECORD_SUFFIXES + + def _points_at_derived(bucket: str, key: str) -> bool: """Whether an R2 bucket/key pair addresses derived build output. @@ -520,7 +534,7 @@ def _derived_source_provenance_issue(fact: AggregateFact) -> str | None: "Chronicle consumer facts must point at raw source artifacts, not " "derived build artifacts." ) - if source_record_id.endswith(".ledger_derived"): + if _is_derived_source_record_id(source_record_id): return ( "Chronicle source_record_id must identify a publisher-backed row, not " "a downstream derived target row." diff --git a/tests/test_chronicle_consumer_contract.py b/tests/test_chronicle_consumer_contract.py index 30ac88d5..9cc730c1 100644 --- a/tests/test_chronicle_consumer_contract.py +++ b/tests/test_chronicle_consumer_contract.py @@ -909,6 +909,18 @@ def test_export_consumer_facts_cli_rejects_contract_invalid_facts(tmp_path, caps }, "publisher.raw.fact", ), + # The derived-row marker renames with everything else, so the guard has + # to reject the chronicle spelling the same way it rejects the ledger + # one (PolicyEngine/chronicle#143, mechanism 3). + ( + { + "source_name": "irs_soi", + "source_file": "publisher.xlsx", + "raw_r2_bucket": "ledger-raw", + "raw_r2_uri": "r2://ledger-raw/raw/source/publisher.xlsx", + }, + "irs_soi.ty2024.table.us.taxable_interest_amount.chronicle_derived", + ), ], ) def test_consumer_contract_rejects_downstream_derived_target_facts( @@ -928,6 +940,46 @@ def test_consumer_contract_rejects_downstream_derived_target_facts( assert "derived_fact_provenance" in {error.code for error in report.errors} +def test_derived_record_marker_is_rejected_in_either_spelling(): + """Both rename-window spellings produce the identical boundary error.""" + fact = _soi_agi_fact() + base = "irs_soi.ty2024.table.us.taxable_interest_amount" + + reports = { + suffix: validate_consumer_fact_contract( + [replace(fact, source_record_id=f"{base}.{suffix}")] + ) + for suffix in ("ledger_derived", "chronicle_derived") + } + + ledger_errors = [ + (error.code, error.message) for error in reports["ledger_derived"].errors + ] + chronicle_errors = [ + (error.code, error.message) for error in reports["chronicle_derived"].errors + ] + assert ledger_errors == chronicle_errors + assert "derived_fact_provenance" in {code for code, _ in ledger_errors} + + +@pytest.mark.parametrize( + "source_record_id", + [ + # A publisher-backed row that merely contains the marker as a word, or + # carries it without the separating dot, is not a derived target row. + "irs_soi.ty2024.table.us.chronicle_derived_totals", + "irs_soi.ty2024.table.us.ledger_derived_totals", + "chronicle_derived", + ], +) +def test_derived_record_marker_matches_the_whole_final_segment(source_record_id): + fact = replace(_soi_agi_fact(), source_record_id=source_record_id) + + report = validate_consumer_fact_contract([fact]) + + assert report.valid + + def test_export_consumer_facts_cli_preserves_decimal_values(tmp_path, capsys): input_path = tmp_path / "facts.jsonl" output_path = tmp_path / "consumer_facts.jsonl" From d9476ef67431d9c7276e308eb0d72bedaa6a455d Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:15:03 -0400 Subject: [PATCH 15/42] Let CHRONICLE_SCHEMA configure the mirror writer, at call time The hosted mirror loader is the primary writer into the schema, and both it and its CLI defaulted to the literal "ledger": setting CHRONICLE_SCHEMA to rehearse a cutover moved the read-side client and left the writer pointed at production. The default now resolves through the shared helper -- one home for the CHRONICLE_SCHEMA -> POLICYENGINE_LEDGER_SCHEMA -> LEDGER_SCHEMA -> "ledger" ladder -- whenever no explicit --schema is supplied. Defaults are unchanged. Resolution is a function rather than a module constant, so db.supabase_client no longer binds the schema while being imported. That import happens at collection, before any fixture, which is why the namespace test needed a reload dance to assert the defaults; it now just calls the resolver. A pytest_configure hook clears the rename window before collection too, so no module can read (or warn from) an operator's shell on the way in. Six tests fail against the previous code: the loader and its CLI ignore the variable, and the namespace assertions see the collection-time constant. Co-Authored-By: Claude Fable 5.1 --- chronicle/env.py | 23 ++++++ chronicle/harness.py | 16 +++- chronicle/mirror.py | 13 +++- db/supabase_client.py | 43 ++++++++--- tests/conftest.py | 17 ++++ tests/test_chronicle_env.py | 45 ++++++++--- tests/test_chronicle_mirror.py | 124 ++++++++++++++++++++++++++++++ tests/test_chronicle_namespace.py | 34 ++++---- 8 files changed, 276 insertions(+), 39 deletions(-) diff --git a/chronicle/env.py b/chronicle/env.py index 3a2f7af6..0d4bbbad 100644 --- a/chronicle/env.py +++ b/chronicle/env.py @@ -19,8 +19,11 @@ __all__ = [ "CHRONICLE_ENV_PREFIX", + "CHRONICLE_SCHEMA_ENV", "ChronicleEnvDeprecationWarning", + "DEFAULT_CHRONICLE_SCHEMA", "LEGACY_ENV_PREFIXES", + "default_chronicle_schema", "env_flag", "env_names", "env_value", @@ -34,6 +37,14 @@ TRUTHY_ENV_VALUES = frozenset({"1", "true", "yes", "on"}) +CHRONICLE_SCHEMA_ENV = "CHRONICLE_SCHEMA" + +# The hosted Postgres schema is still named "ledger". Renaming the schema value +# is a later slice of PolicyEngine/chronicle#143, coordinated with the CI +# writers that already target it; only the variable that overrides the name has +# moved to the chronicle prefix. +DEFAULT_CHRONICLE_SCHEMA = "ledger" + class ChronicleEnvDeprecationWarning(FutureWarning): """A ledger-era environment variable supplied a Chronicle setting. @@ -126,6 +137,18 @@ def env_value(*names: str, default: _Default = None) -> str | _Default: return default if value is None else value +def default_chronicle_schema() -> str: + """Resolve the Chronicle schema: ``$CHRONICLE_SCHEMA``, else the default. + + Every reader of the setting goes through this function so the lookup ladder + and the default have one home. It resolves at call time rather than at + import: a module-level constant binds whatever the shell held when the + module was first imported, which for a library means an arbitrary moment + the caller cannot control, and for the test suite means collection. + """ + return env_value(CHRONICLE_SCHEMA_ENV, default=DEFAULT_CHRONICLE_SCHEMA) + + def env_flag(*names: str) -> bool: """Return whether the first set value across ``names`` reads as true. diff --git a/chronicle/harness.py b/chronicle/harness.py index dc9df8ba..a5646553 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -37,6 +37,7 @@ ) from chronicle.core import AggregateFact, ValidationReport, validate_facts from chronicle.database import ChronicleDbBuildReport, build_chronicle_db +from chronicle.env import DEFAULT_CHRONICLE_SCHEMA from chronicle.mirror import ( ChronicleMirrorExportReport, SupabaseMirrorLoadReport, @@ -452,12 +453,16 @@ def export_chronicle_db_table_files( def load_supabase_mirror_files( input_dir: str | Path, *, - schema: str = "ledger", + schema: str | None = None, batch_size: int = 500, dry_run: bool = False, build_artifacts_path: str | Path | None = None, ) -> SupabaseMirrorLoadReport: - """Load exported Chronicle JSONL mirror files into Supabase/Postgres.""" + """Load exported Chronicle JSONL mirror files into Supabase/Postgres. + + ``schema`` of None resolves to ``$CHRONICLE_SCHEMA``, else the default + schema, so the hosted mirror writer answers to the renamed variable. + """ table_paths = ( {"build_artifacts": Path(build_artifacts_path)} if build_artifacts_path is not None @@ -1112,8 +1117,11 @@ def main(argv: list[str] | None = None) -> int: ) mirror_load_parser.add_argument( "--schema", - default="ledger", - help="Supabase/Postgres schema to load into.", + default=None, + help=( + "Supabase/Postgres schema to load into. Defaults to " + f"$CHRONICLE_SCHEMA, else {DEFAULT_CHRONICLE_SCHEMA}." + ), ) mirror_load_parser.add_argument( "--batch-size", diff --git a/chronicle/mirror.py b/chronicle/mirror.py index 7a3ccaf3..f9c95dc2 100644 --- a/chronicle/mirror.py +++ b/chronicle/mirror.py @@ -11,6 +11,8 @@ from pathlib import Path from typing import Any +from chronicle.env import default_chronicle_schema + LEDGER_MIRROR_TABLES = ( "ledger_builds", "build_artifacts", @@ -200,15 +202,22 @@ def export_chronicle_db_tables( def load_supabase_mirror( input_dir: str | Path, *, - schema: str = "ledger", + schema: str | None = None, batch_size: int = 500, dry_run: bool = False, table_paths: dict[str, str | Path] | None = None, client: Any | None = None, ) -> SupabaseMirrorLoadReport: - """Load exported Chronicle JSONL mirror files into Supabase/Postgres.""" + """Load exported Chronicle JSONL mirror files into Supabase/Postgres. + + ``schema`` defaults to :func:`chronicle.env.default_chronicle_schema`, so + the writer that owns the hosted mirror answers to ``CHRONICLE_SCHEMA`` (and + the ledger-era names behind it) exactly like every other reader of the + setting. The resolved name is reported back in the load report. + """ if batch_size < 1: raise ValueError("batch_size must be at least 1.") + schema = schema or default_chronicle_schema() input_path = Path(input_dir) tables: list[SupabaseTableLoad] = [] errors: list[str] = [] diff --git a/db/supabase_client.py b/db/supabase_client.py index 50ce4f30..190d8b7e 100644 --- a/db/supabase_client.py +++ b/db/supabase_client.py @@ -15,14 +15,33 @@ from supabase import create_client, Client -from chronicle.env import env_value +from chronicle.env import default_chronicle_schema, env_value -# The hosted Postgres schema is still named "ledger"; only the environment -# variable that overrides it has moved to the chronicle prefix. Renaming the -# schema itself is a later slice of PolicyEngine/chronicle#143, coordinated -# with the CI writers that already target the ledger schema. -LEDGER_SCHEMA = env_value("CHRONICLE_SCHEMA") or "ledger" -TARGETS_SCHEMA = env_value("POLICYENGINE_TARGETS_SCHEMA") or "targets" +TARGETS_SCHEMA_ENV = "POLICYENGINE_TARGETS_SCHEMA" +DEFAULT_TARGETS_SCHEMA = "targets" + + +def chronicle_schema() -> str: + """Resolve the hosted Chronicle schema for a query. + + Read at call time, not bound at import: an import-time constant fixes the + schema at whatever the environment held when this module was first + imported, which the caller does not control (in the test suite that moment + is collection, before any fixture has isolated the environment). The + hosted schema is still named "ledger" -- only the variable that overrides + it has moved to the chronicle prefix, and renaming the schema value is a + later slice of PolicyEngine/chronicle#143. + """ + return default_chronicle_schema() + + +def targets_schema() -> str: + """Resolve the hosted targets schema. Read at call time, as above. + + ``POLICYENGINE_TARGETS_SCHEMA`` names a surface outside the ledger rename + window, so it is read literally. + """ + return env_value(TARGETS_SCHEMA_ENV, default=DEFAULT_TARGETS_SCHEMA) @dataclass @@ -109,7 +128,7 @@ def query_sources( List of source records """ client = get_supabase_client() - query = _table(client, LEDGER_SCHEMA, "sources").select("*") + query = _table(client, chronicle_schema(), "sources").select("*") if jurisdiction: query = query.eq("jurisdiction", jurisdiction) @@ -138,7 +157,9 @@ def query_strata( List of strata records with nested constraints """ client = get_supabase_client() - query = _table(client, TARGETS_SCHEMA, "strata").select("*, stratum_constraints(*)") + query = _table(client, targets_schema(), "strata").select( + "*, stratum_constraints(*)" + ) if jurisdiction: query = query.eq("jurisdiction", jurisdiction) @@ -167,7 +188,7 @@ def query_targets( """ client = get_supabase_client() # Nested join: strata with their stratum_constraints - query = _table(client, TARGETS_SCHEMA, "targets").select( + query = _table(client, targets_schema(), "targets").select( "*, strata(*, stratum_constraints(*)), sources(*)" ) @@ -214,7 +235,7 @@ def insert_targets_batch( for i in range(0, len(targets), chunk_size): chunk = targets[i : i + chunk_size] - _table(client, TARGETS_SCHEMA, "targets").insert(chunk).execute() + _table(client, targets_schema(), "targets").insert(chunk).execute() total += len(chunk) return total diff --git a/tests/conftest.py b/tests/conftest.py index 378fa0b3..3d315441 100644 --- a/tests/conftest.py +++ b/tests/conftest.py @@ -22,6 +22,23 @@ RENAME_WINDOW_PREFIXES = (CHRONICLE_ENV_PREFIX, *LEGACY_ENV_PREFIXES) +def pytest_configure(config): + """Strip the rename window before collection imports a single module. + + The autouse fixture below runs per test, which is too late for anything a + module does while being imported. Collection happens after this hook, so + clearing here means no module can read an operator's ``LEDGER_SCHEMA`` + (warning as it goes, or freezing it into a constant) before a fixture has + had the chance to isolate it. Modules should resolve settings at call time + rather than at import; this hook makes that property testable instead of + depending on which shell ran pytest. + """ + for name in list(os.environ): + if name.startswith(RENAME_WINDOW_PREFIXES): + del os.environ[name] + reset_env_deprecation_state() + + @pytest.fixture(autouse=True) def isolated_rename_window_env(monkeypatch): """Run every test with no rename-window variable inherited from the shell.""" diff --git a/tests/test_chronicle_env.py b/tests/test_chronicle_env.py index 65029d14..202d86f5 100644 --- a/tests/test_chronicle_env.py +++ b/tests/test_chronicle_env.py @@ -24,7 +24,9 @@ from chronicle.env import ( CHRONICLE_ENV_PREFIX, ChronicleEnvDeprecationWarning, + DEFAULT_CHRONICLE_SCHEMA, LEGACY_ENV_PREFIXES, + default_chronicle_schema, env_flag, env_names, env_value, @@ -252,17 +254,18 @@ def test_db_cli_parser_builds_with_the_env_backed_defaults(monkeypatch, capsys): assert "Manage Chronicle target input data" in capsys.readouterr().out -@pytest.mark.parametrize("name", ["CHRONICLE_SCHEMA", "POLICYENGINE_LEDGER_SCHEMA"]) -def test_supabase_schema_honors_both_names(monkeypatch, name): +@pytest.mark.parametrize( + "name", + ["CHRONICLE_SCHEMA", "POLICYENGINE_LEDGER_SCHEMA", "LEDGER_SCHEMA"], +) +def test_supabase_schema_honors_every_name_in_the_window(monkeypatch, name): + """Set after import and still honored: the schema is read at call time.""" import db.supabase_client monkeypatch.setenv(name, "chronicle_probe") - try: - reloaded = importlib.reload(db.supabase_client) - assert reloaded.LEDGER_SCHEMA == "chronicle_probe" - finally: - monkeypatch.delenv(name, raising=False) - importlib.reload(db.supabase_client) + + assert db.supabase_client.chronicle_schema() == "chronicle_probe" + assert default_chronicle_schema() == "chronicle_probe" def test_supabase_schema_default_is_unchanged(): @@ -270,7 +273,31 @@ def test_supabase_schema_default_is_unchanged(): # The hosted schema name itself is out of this slice; only the variable # that overrides it moved. - assert db.supabase_client.LEDGER_SCHEMA == "ledger" + assert DEFAULT_CHRONICLE_SCHEMA == "ledger" + assert default_chronicle_schema() == "ledger" + assert db.supabase_client.chronicle_schema() == "ledger" + assert db.supabase_client.targets_schema() == "targets" + + +def test_supabase_schema_is_not_bound_at_import(monkeypatch): + """No module-level constant may freeze the schema at import time. + + A reload under a set variable is the pre-fix behavior this guards against: + it proves nothing about a module that resolved the value once, at + collection, and answers with the stale constant forever after. + """ + import db.supabase_client + + assert not [ + name + for name, value in vars(db.supabase_client).items() + if name.isupper() and value == "ledger" + ] + + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + unreloaded = importlib.import_module("db.supabase_client") + + assert unreloaded.chronicle_schema() == "chronicle_probe" # --------------------------------------------------------------------------- diff --git a/tests/test_chronicle_mirror.py b/tests/test_chronicle_mirror.py index c06a33bd..6b2e694d 100644 --- a/tests/test_chronicle_mirror.py +++ b/tests/test_chronicle_mirror.py @@ -4,6 +4,9 @@ import json +import pytest + +from chronicle.env import ChronicleEnvDeprecationWarning from chronicle.harness import main as harness_main from chronicle.mirror import ( LEDGER_MIRROR_TABLES, @@ -189,6 +192,127 @@ def test_load_supabase_mirror_cli_dry_run(tmp_path, capsys): assert payload["table_count"] == len(LEDGER_MIRROR_TABLES) +# --------------------------------------------------------------------------- +# Schema configuration +# +# The mirror loader is the primary writer into the hosted schema, so it is the +# call site CHRONICLE_SCHEMA has to reach (PolicyEngine/chronicle#143, +# mechanism 3). It defaulted to the literal "ledger" while only the read-side +# client honored the renamed variable, which would have sent a rehearsal load +# into production the moment an operator set it. +# --------------------------------------------------------------------------- + + +def _empty_mirror(tmp_path): + mirror_dir = tmp_path / "mirror" + mirror_dir.mkdir() + for table in LEDGER_MIRROR_TABLES: + (mirror_dir / f"{table}.jsonl").write_text("") + return mirror_dir + + +def _one_build_artifact(tmp_path): + path = tmp_path / "build_artifacts.jsonl" + path.write_text( + json.dumps( + { + "build_artifact_key": "ledger.build_artifact.v1:test", + "build_id": "ledger.build.v1:test", + "artifact_kind": "json", + "artifact_name": "reports/build_summary.json", + "sha256": "abc", + "size_bytes": 3, + "r2_bucket": "ledger-derived", + "r2_key": "derived/test", + "r2_uri": "r2://ledger-derived/derived/test", + }, + sort_keys=True, + ) + + "\n" + ) + return path + + +def _load_into_fake_client(tmp_path, **kwargs): + client = _FakeSupabaseClient() + report = load_supabase_mirror( + _empty_mirror(tmp_path), + table_paths={"build_artifacts": _one_build_artifact(tmp_path)}, + client=client, + **kwargs, + ) + return report, client + + +def test_load_supabase_mirror_defaults_to_the_ledger_schema(tmp_path): + report, client = _load_into_fake_client(tmp_path) + + assert report.schema == "ledger" + assert [upsert[0] for upsert in client.upserts] == ["ledger"] + + +def test_load_supabase_mirror_writes_to_the_chronicle_schema(tmp_path, monkeypatch): + """The renamed variable configures the writer, not just the reader.""" + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + + report, client = _load_into_fake_client(tmp_path) + + assert report.schema == "chronicle_probe" + assert [upsert[0] for upsert in client.upserts] == ["chronicle_probe"] + + +@pytest.mark.parametrize("name", ["POLICYENGINE_LEDGER_SCHEMA", "LEDGER_SCHEMA"]) +def test_load_supabase_mirror_honors_a_ledger_era_schema_name( + tmp_path, monkeypatch, name +): + monkeypatch.setenv(name, "legacy_probe") + + with pytest.warns(ChronicleEnvDeprecationWarning): + report, client = _load_into_fake_client(tmp_path) + + assert report.schema == "legacy_probe" + assert [upsert[0] for upsert in client.upserts] == ["legacy_probe"] + + +def test_an_explicit_schema_still_wins_over_the_environment(tmp_path, monkeypatch): + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + + report, client = _load_into_fake_client(tmp_path, schema="explicit_probe") + + assert report.schema == "explicit_probe" + assert [upsert[0] for upsert in client.upserts] == ["explicit_probe"] + + +def test_load_supabase_mirror_cli_writes_to_the_configured_schema( + tmp_path, monkeypatch, capsys +): + """The CLI resolves the same way when no --schema is supplied.""" + client = _FakeSupabaseClient() + monkeypatch.setattr("chronicle.mirror._get_supabase_client", lambda: client) + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + argv = [ + "load-supabase-mirror", + "--dir", + str(_empty_mirror(tmp_path)), + "--build-artifacts", + str(_one_build_artifact(tmp_path)), + ] + + exit_code = harness_main(argv) + payload = json.loads(capsys.readouterr().out) + + assert exit_code == 0 + assert payload["schema"] == "chronicle_probe" + assert [upsert[0] for upsert in client.upserts] == ["chronicle_probe"] + + assert harness_main([*argv, "--schema", "explicit_probe"]) == 0 + assert json.loads(capsys.readouterr().out)["schema"] == "explicit_probe" + assert [upsert[0] for upsert in client.upserts] == [ + "chronicle_probe", + "explicit_probe", + ] + + class _FakeSupabaseClient: def __init__(self): self.upserts = [] diff --git a/tests/test_chronicle_namespace.py b/tests/test_chronicle_namespace.py index e8890948..5dad695e 100644 --- a/tests/test_chronicle_namespace.py +++ b/tests/test_chronicle_namespace.py @@ -1,7 +1,5 @@ """Tests for the Chronicle namespace.""" -import importlib - from chronicle.client import get_supabase_client from chronicle.normalization import convert_units from chronicle.targets import ( @@ -27,20 +25,30 @@ def test_chronicle_client_reexports_supabase_client(): def test_chronicle_supabase_schema_boundaries_are_defaulted(): - """The schema names are import-time constants, so re-read them here. - - ``db.supabase_client`` resolves them from the environment when it is first - imported, which happens at collection — before the suite-wide - ``isolated_rename_window_env`` fixture clears an operator's - ``CHRONICLE_SCHEMA``. Reloading under the cleared environment is what makes - this a test of the defaults rather than of the shell. + """The schema names resolve per call, so this reads the cleared window. + + ``db.supabase_client`` is imported at collection, before any fixture runs. + Resolving the schema there — as an import-time constant — would bind an + operator's ``CHRONICLE_SCHEMA`` (or a ledger-era name, warning as it went) + into the module for the whole session, and no fixture could take it back. + Reading at call time is what makes this a test of the defaults rather than + of the shell. """ - import db.supabase_client + from db import supabase_client + + assert supabase_client.chronicle_schema() == "ledger" + assert supabase_client.targets_schema() == "targets" + + +def test_chronicle_supabase_schema_follows_the_environment(monkeypatch): + """The renamed variable reaches the client after it has been imported.""" + from db import supabase_client - supabase_client = importlib.reload(db.supabase_client) + monkeypatch.setenv("CHRONICLE_SCHEMA", "chronicle_probe") + monkeypatch.setenv("POLICYENGINE_TARGETS_SCHEMA", "targets_probe") - assert supabase_client.LEDGER_SCHEMA == "ledger" - assert supabase_client.TARGETS_SCHEMA == "targets" + assert supabase_client.chronicle_schema() == "chronicle_probe" + assert supabase_client.targets_schema() == "targets_probe" def test_chronicle_normalization_exports_helpers(): From e7d1cd1216c0b795a1b6dfb11d3bbf9f4530f9b4 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:26:41 -0400 Subject: [PATCH 16/42] Address the manifest a fetch is actually revising, and read it strictly Four defects in the state fetch-artifact reads before it writes: - It always targeted manifest.yaml. Seven tracked packages keep a manifest_*_source_package.yaml instead, and three publisher directories keep two of them, so the IRA revision workflow the docs cite would have written a third manifest beside the real ones and never seen the recorded block. A --manifest filename now selects it; the name has to stay inside the package. - Revision protection vanished when the entry had no storage.r2 -- a fetch that only registered bytes, or one whose upload failed, which is the state #225 landed in. A manifest entry identifies its bytes by its declared sha256 whether or not it has been published, and a fetch of different bytes over either identity is refused unless --record-revision opts in. - Recorded R2 validation read the key or the uri, whichever came first. A block whose key and uri named different objects was preserved verbatim, so the entry kept publishing a URI for bytes it no longer described. Every supplied locator field is now cross-checked against every other and against the content-addressed key shape; a contradiction is an error at fetch time and a refusal at publish time, never a silent preserve. - A manifest that parsed as anything but a mapping was treated as absent, so the fetch would replace it with a single entry. It is now refused before the publisher is read at all, and inventory-artifacts and publish-raw report it rather than crashing on it. The refusals share a SourceArtifactManifestError base, so the CLI reports all of them as an exit-1 message with nothing written. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 525 +++++++++++++++++++++++------- chronicle/harness.py | 29 +- tests/test_chronicle_artifacts.py | 416 +++++++++++++++++++++++ 3 files changed, 846 insertions(+), 124 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 9421ab08..1b3f202a 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -9,6 +9,7 @@ from datetime import UTC, datetime from pathlib import Path import posixpath +import re import shlex import sqlite3 import subprocess @@ -42,6 +43,27 @@ DEFAULT_R2_PREFIX = "raw" DEFAULT_R2_DERIVED_PREFIX = "derived" +# Most packages keep one manifest.yaml. Publisher directories that feed several +# source packages keep one manifest each -- db/data/irs_soi/ira_contributions +# holds manifest_traditional_source_package.yaml beside the Roth one -- so the +# name is an input, not a constant, wherever a caller addresses a package. +DEFAULT_MANIFEST_FILENAME = "manifest.yaml" + + +def _manifest_path(output: Path, manifest_filename: str) -> Path: + """Return the named manifest inside ``output``. + + The name is a filename, not a path: it selects among the manifests a + package directory keeps, and must not reach outside it. + """ + name = manifest_filename.strip() + if not name or name in (".", "..") or name != Path(name).name: + raise ValueError( + "Manifest must name a file inside the package directory, not " + f"{manifest_filename!r}." + ) + return output / name + def default_r2_raw_bucket() -> str: """Resolve the raw bucket: ``$CHRONICLE_R2_RAW_BUCKET`` or the default.""" @@ -53,15 +75,44 @@ def default_r2_derived_bucket() -> str: return env_value(R2_DERIVED_BUCKET_ENV, default=DEFAULT_R2_DERIVED_BUCKET) -class SourceArtifactRevisionError(RuntimeError): - """Fetched bytes are not the bytes the recorded R2 object holds. +class SourceArtifactManifestError(RuntimeError): + """A manifest refuses the write a fetch is about to make. + + Every subclass is raised before anything is downloaded, cached, uploaded or + rewritten, so a refusal leaves the package exactly as it was. + """ + + +class SourceArtifactRevisionError(SourceArtifactManifestError): + """Fetched bytes are not the bytes the manifest entry identifies. + + A manifest entry identifies specific bytes: by its declared ``sha256``, and + -- once published -- by a content-addressed R2 key that repeats them. When + a publisher re-publishes under the same URL and vintage, rewriting that + entry would attach its provenance, and any recorded URI, to bytes it never + described. Chronicle refuses instead: same vintage plus new bytes is a new + release revision (docs/adr-chronicle-fact-identity-v2.md), registered with + ``fetch-artifact --record-revision``. + """ + + +class MalformedManifestError(SourceArtifactManifestError): + """A manifest document, or a block inside one, is not a mapping. + + Reading such a file as an absent manifest would let a fetch replace it with + a single entry, dropping whatever the unreadable document recorded. + """ + - Raw R2 keys are content-addressed, so a recorded ``storage.r2`` block is a - claim about specific bytes. When a publisher re-publishes under the same - URL and vintage, keeping that block would attach its provenance to bytes it - never described. Chronicle refuses instead: same vintage plus new bytes is a - new release revision (docs/adr-chronicle-fact-identity-v2.md), registered - with ``fetch-artifact --record-revision``. +class RecordedR2LocatorError(SourceArtifactManifestError): + """A recorded ``storage.r2`` block does not locate exactly one object. + + ``provider``, ``bucket``, ``key`` and ``uri`` all describe the same object, + so any that are supplied have to agree, and the key has to carry the + ``{sha256}/{filename}`` tail that says which bytes it holds. A block whose + fields contradict each other has no single answer to "which bytes does this + entry claim R2 holds", and preserving or publishing under it would ship + whichever field the reader happened to consult. """ @@ -114,6 +165,60 @@ def to_dict(self) -> dict[str, str]: } +# A raw key ends in {sha256}/{filename} (see build_r2_key), so the segment +# before the filename is what says which bytes the object holds. +_SHA256_KEY_SEGMENT = re.compile(r"[0-9a-f]{64}") + + +@dataclass(frozen=True) +class RecordedR2Object: + """The R2 object a manifest entry's ``storage.r2`` block claims exists. + + Built only by :func:`_validated_recorded_r2`, so every instance names one + object whose locator fields agree with each other. + """ + + provider: str + bucket: str + key: str + sha256: str + filename: str + + @property + def uri(self) -> str: + """Return the storage URI the recorded fields spell out.""" + return f"{self.provider}://{self.bucket}/{self.key}" + + +@dataclass(frozen=True) +class RecordedIdentity: + """The bytes a manifest entry says its vintage currently holds. + + From the recorded object's content-addressed key once the entry has been + published, and from the entry's own declared ``sha256``/``filename`` before + that. ``r2`` is None in the second case: protection does not wait for an + upload to have happened. + """ + + sha256: str + filename: str + size_bytes: int | None + declared_sha256: str | None + r2: RecordedR2Object | None + + def holds(self, *, sha256: str, filename: str) -> bool: + """Whether this identity is exactly the given bytes under that name. + + The filename participates only when the entry records one: a published + key always carries it, an entry that declares bytes and no name does + not, and inventing a mismatch there would refuse a re-fetch of the very + bytes the entry describes. + """ + if self.sha256 != sha256: + return False + return not self.filename or self.filename == Path(filename).name + + @dataclass(frozen=True) class ArtifactCommandResult: """Result from a storage command.""" @@ -430,6 +535,7 @@ def fetch_source_artifact( source_page: str | None = None, table: str | None = None, filename: str | None = None, + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -438,20 +544,36 @@ def fetch_source_artifact( ) -> ArtifactFetchReport: """Fetch/register a source artifact and optionally upload it to R2. + ``manifest_filename`` names the manifest inside ``output_dir`` the entry + belongs to. Packages that split one publisher directory across several + source packages keep one manifest each, so a fetch that always wrote + ``manifest.yaml`` would write a fresh manifest beside the real ones and + never see the entry it is revising. + ``record_revision`` opts into registering a publisher revision: the fetched bytes get their own content-addressed key under the configured bucket and the superseded object moves to ``storage.previous_r2``. Without it, bytes - that disagree with the recorded object raise + that disagree with the entry's recorded identity raise :class:`SourceArtifactRevisionError` before anything is overwritten. """ r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) + manifest_path = _manifest_path(output, manifest_filename) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, default_prefix=DEFAULT_R2_PREFIX, source_id=source_id, package_path=output, ) + # Read and validate the entry being written before anything is fetched: a + # manifest Chronicle cannot read, or a recorded block that names two + # different objects, is a refusal that need not touch the publisher. + recorded_identity = _recorded_identity( + _manifest_file_spec(_read_manifest(manifest_path), year), + manifest_path=manifest_path, + year=year, + ) + fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() content, inferred_filename = _read_artifact(source_url) artifact_filename = filename or inferred_filename @@ -460,12 +582,12 @@ def fetch_source_artifact( sha256 = hashlib.sha256(content).hexdigest() size_bytes = len(content) - manifest_path = output / "manifest.yaml" # Guard before the cached artifact is touched. A rejected fetch must leave # the recorded bytes and their manifest entry exactly as they were. - _assert_recorded_object_holds_these_bytes( - manifest_path, + _assert_recorded_identity_holds_these_bytes( + recorded_identity, + manifest_path=manifest_path, year=year, filename=artifact_filename, sha256=sha256, @@ -676,7 +798,7 @@ def publish_derived_artifacts( def publish_source_artifacts( root: str | Path, *, - manifest_filename: str = "manifest.yaml", + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, source_id: str | None = None, package_id: str | None = None, r2_bucket: str | None = None, @@ -697,8 +819,8 @@ def publish_source_artifacts( errors: list[str] = [] for manifest_path in sorted(root_path.rglob(manifest_filename)): try: - manifest = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - except (OSError, yaml.YAMLError) as exc: + manifest = _read_manifest(manifest_path) + except (OSError, MalformedManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue @@ -801,7 +923,7 @@ def write_build_artifacts_jsonl( def inventory_source_artifacts( root: str | Path, *, - manifest_filename: str = "manifest.yaml", + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, ) -> ArtifactInventoryReport: """Inventory manifest-declared source artifacts under a root directory.""" root_path = Path(root) @@ -824,9 +946,8 @@ def inventory_source_artifacts( manifests = sorted(root_path.rglob(manifest_filename)) for manifest_path in manifests: try: - manifest = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - files = manifest.get("files") or {} - except (OSError, yaml.YAMLError) as exc: + files = _read_manifest(manifest_path).get("files") or {} + except (OSError, MalformedManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue if not isinstance(files, dict): @@ -1110,11 +1231,31 @@ def _filename_from_url(source_url: str) -> str: def _read_manifest(manifest_path: Path) -> dict[str, Any]: - """Return a manifest's parsed payload, or an empty mapping.""" + """Return a manifest's parsed payload, refusing a document it cannot read. + + An absent or empty manifest reads as an empty mapping: ``fetch-artifact`` + writes the first entry into a package that has none. A document that parses + as anything else -- a list, a scalar, a truncated or half-merged file -- is + not an absent manifest, and treating it as one would let the fetch replace + it with a single entry and drop everything it recorded. + """ if not manifest_path.exists(): return {} - payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - return payload if isinstance(payload, dict) else {} + try: + payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) + except yaml.YAMLError as exc: + raise MalformedManifestError(f"{manifest_path} is not valid YAML: {exc}") from ( + exc + ) + if payload is None: + return {} + if not isinstance(payload, dict): + raise MalformedManifestError( + f"{manifest_path} must be a YAML mapping; it parses as a " + f"{type(payload).__name__}. Chronicle will not overwrite a manifest " + "it cannot read." + ) + return payload def _manifest_file_spec(payload: dict[str, Any], year: Any) -> dict[str, Any]: @@ -1135,40 +1276,182 @@ def _recorded_storage(spec: Any) -> dict[str, Any]: def _recorded_r2(spec: Any) -> dict[str, Any]: - """Return a manifest file spec's recorded ``storage.r2`` block, if any.""" + """Return the recorded ``storage.r2`` block verbatim, if any. + + Raw access, for callers that carry the block forward as history. Callers + that reason about which object it names go through + :func:`_validated_recorded_r2` instead. + """ recorded = _recorded_storage(spec).get("r2") return recorded if isinstance(recorded, dict) else {} -def _r2_key_identity(recorded_r2: dict[str, Any]) -> tuple[str, str]: - """Return the ``(sha256, filename)`` a recorded R2 object is addressed by. +def _split_r2_uri(uri: str) -> tuple[str, str, str] | None: + """Split ``provider://bucket/key`` into its three parts, or None.""" + provider, separator, remainder = uri.partition("://") + if not separator or not provider: + return None + bucket, separator, key = remainder.partition("/") + if not separator or not bucket or not key: + return None + return (provider, bucket, key) - Raw keys are ``{prefix}/{source_id}/{package_id}/{year}/{sha256}/{filename}`` - (see :func:`build_r2_key`), so the last two segments say which bytes the - object holds; the URI ends in the same two segments and stands in for a - block that records only that. A locator in any other shape yields empty - strings and therefore never matches fetched bytes. + +def _validated_recorded_r2( + spec: Any, + *, + manifest_path: Path, + year: Any, +) -> RecordedR2Object | None: + """Return the object a recorded ``storage.r2`` block names, or None. + + Every locator field the block supplies is cross-checked against every + other: ``key`` against the URI's path, ``bucket`` against its authority, + ``provider`` against its scheme, and the resulting key against the + canonical content-addressed shape :func:`build_r2_key` writes. Reading one + field and trusting the rest is what lets a block that says two different + things survive a preserve or a publish. """ - locator = recorded_r2.get("key") or recorded_r2.get("uri") - if not isinstance(locator, str): - return ("", "") - parts = [part for part in locator.split("/") if part] - if len(parts) < 2: - return ("", "") - return (parts[-2], parts[-1]) + storage = _recorded_storage_block(spec, manifest_path=manifest_path, year=year) + if "r2" not in storage: + return None + block = storage["r2"] + where = f"{manifest_path} entry {year!r} storage.r2" + if not isinstance(block, dict): + raise MalformedManifestError( + f"{where} must be a mapping; it is a {type(block).__name__}." + ) + + supplied: dict[str, str] = {} + for field in ("provider", "bucket", "key", "uri"): + value = block.get(field) + if value is None: + continue + if not isinstance(value, str) or not value.strip(): + raise RecordedR2LocatorError( + f"{where}: {field} must be a non-empty string, not {value!r}." + ) + supplied[field] = value + + provider = supplied.get("provider") + bucket = supplied.get("bucket") + key = supplied.get("key") + uri = supplied.get("uri") + if uri is not None: + parts = _split_r2_uri(uri) + if parts is None: + raise RecordedR2LocatorError( + f"{where}: uri {uri!r} is not provider://bucket/key." + ) + for field, value, from_uri in zip( + ("provider", "bucket", "key"), (provider, bucket, key), parts + ): + if value is not None and value != from_uri: + raise RecordedR2LocatorError( + f"{where}: {field}={value!r} contradicts uri {uri!r}, which " + f"names {from_uri!r}. The block records two different " + "objects, so Chronicle cannot say which bytes it claims." + ) + provider, bucket, key = ( + provider or parts[0], + bucket or parts[1], + key or parts[2], + ) + + missing = [ + field + for field, value in ( + ("provider", provider), + ("bucket", bucket), + ("key", key), + ) + if not value + ] + if missing: + raise RecordedR2LocatorError( + f"{where}: records no {', '.join(missing)}. A recorded block has to " + "locate its object, by key and bucket or by uri." + ) + + segments = key.split("/") + if ( + len(segments) < 2 + or not all(segments) + or not _SHA256_KEY_SEGMENT.fullmatch(segments[-2]) + ): + raise RecordedR2LocatorError( + f"{where}: key {key!r} is not content-addressed. A raw key ends in " + "{sha256}/{filename}, which is what says the object holds the " + "entry's bytes; Chronicle will not guess for a key that does not." + ) + return RecordedR2Object( + provider=provider, + bucket=bucket, + key=key, + sha256=segments[-2], + filename=segments[-1], + ) -def _r2_holds_these_bytes( - recorded_r2: dict[str, Any], +def _recorded_storage_block( + spec: Any, *, - sha256: str, - filename: str, -) -> bool: - """Whether a recorded ``storage.r2`` block addresses exactly these bytes.""" - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) - return bool(recorded_sha256) and (recorded_sha256, recorded_filename) == ( - sha256, - Path(filename).name, + manifest_path: Path, + year: Any, +) -> dict[str, Any]: + """Return the entry's ``storage`` mapping, refusing a malformed one.""" + if not isinstance(spec, dict) or "storage" not in spec: + return {} + storage = spec["storage"] + if not isinstance(storage, dict): + raise MalformedManifestError( + f"{manifest_path} entry {year!r} storage must be a mapping; it is a " + f"{type(storage).__name__}." + ) + return storage + + +def _recorded_identity( + spec: Any, + *, + manifest_path: Path, + year: Any, +) -> RecordedIdentity | None: + """Return what a manifest entry says its vintage holds, if anything. + + A published entry is identified by its recorded object's content-addressed + key. An entry that has not been published yet -- registered without an + upload, or left behind by a failed one -- is identified by its own declared + ``sha256`` and ``filename``. Both are recorded identities, and a fetch of + different bytes over either one is a publisher revision. + """ + recorded_r2 = _validated_recorded_r2(spec, manifest_path=manifest_path, year=year) + declared_sha256 = spec.get("sha256") if isinstance(spec, dict) else None + declared_sha256 = declared_sha256 if isinstance(declared_sha256, str) else None + declared_filename = spec.get("filename") if isinstance(spec, dict) else None + declared_filename = ( + declared_filename if isinstance(declared_filename, str) else None + ) + size_bytes = spec.get("size_bytes") if isinstance(spec, dict) else None + size_bytes = size_bytes if isinstance(size_bytes, int) else None + if recorded_r2 is not None: + return RecordedIdentity( + sha256=recorded_r2.sha256, + filename=recorded_r2.filename, + # Only report a size the recorded key agrees with: an entry can + # arrive here already describing the new bytes. + size_bytes=size_bytes if declared_sha256 == recorded_r2.sha256 else None, + declared_sha256=declared_sha256, + r2=recorded_r2, + ) + if not declared_sha256: + return None + return RecordedIdentity( + sha256=declared_sha256, + filename=Path(declared_filename).name if declared_filename else "", + size_bytes=size_bytes, + declared_sha256=declared_sha256, + r2=None, ) @@ -1177,33 +1460,35 @@ def _revision_error_message( manifest_path: Path, year: Any, filename: str, - recorded_spec: dict[str, Any], - recorded_r2: dict[str, Any], + identity: RecordedIdentity, sha256: str, size_bytes: int, r2_bucket: str, ) -> str: """Explain a refused fetch: recorded identity, fetched identity, next step.""" - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) - declared_sha256 = recorded_spec.get("sha256") - recorded_size = ( - recorded_spec.get("size_bytes") if declared_sha256 == recorded_sha256 else None + records = ( + f"already records the R2 object {identity.r2.uri}, which holds" + if identity.r2 is not None + else "already records" ) message = ( - f"{manifest_path} entry {year!r} already records the R2 object " - f"{recorded_r2.get('uri') or recorded_r2.get('key')}, which holds " - f"sha256={recorded_sha256 or 'unknown'} " - f"filename={recorded_filename or 'unknown'} " - f"size_bytes={recorded_size if recorded_size is not None else 'unknown'}. " + f"{manifest_path} entry {year!r} {records} " + f"sha256={identity.sha256} " + f"filename={identity.filename or 'unknown'} " + f"size_bytes=" + f"{identity.size_bytes if identity.size_bytes is not None else 'unknown'}. " f"The fetched bytes are sha256={sha256} filename={Path(filename).name} " - f"size_bytes={size_bytes}. Chronicle will not attach a recorded, " - "content-addressed R2 URI to bytes it does not describe." + f"size_bytes={size_bytes}. Chronicle will not rewrite a vintage that " + "identifies specific bytes to describe bytes it never identified." ) - if declared_sha256 and declared_sha256 != recorded_sha256: + if identity.r2 is not None and identity.declared_sha256 not in ( + None, + identity.sha256, + ): message += ( - f" (The entry also declares sha256={declared_sha256}, which its own " - "R2 key contradicts: an earlier fetch rewrote the hash without " - "moving the object.)" + f" (The entry also declares sha256={identity.declared_sha256}, which " + "its own R2 key contradicts: an earlier fetch rewrote the hash " + "without moving the object.)" ) return message + ( " The same vintage with new bytes is a new release revision " @@ -1214,9 +1499,10 @@ def _revision_error_message( ) -def _assert_recorded_object_holds_these_bytes( - manifest_path: Path, +def _assert_recorded_identity_holds_these_bytes( + identity: RecordedIdentity | None, *, + manifest_path: Path, year: Any, filename: str, sha256: str, @@ -1225,21 +1511,16 @@ def _assert_recorded_object_holds_these_bytes( record_revision: bool, ) -> None: """Refuse a publisher revision that has not been opted into.""" - if record_revision: - return - recorded_spec = _manifest_file_spec(_read_manifest(manifest_path), year) - recorded_r2 = _recorded_r2(recorded_spec) - if not recorded_r2: + if record_revision or identity is None: return - if _r2_holds_these_bytes(recorded_r2, sha256=sha256, filename=filename): + if identity.holds(sha256=sha256, filename=filename): return raise SourceArtifactRevisionError( _revision_error_message( manifest_path=manifest_path, year=year, filename=filename, - recorded_spec=recorded_spec, - recorded_r2=recorded_r2, + identity=identity, sha256=sha256, size_bytes=size_bytes, r2_bucket=r2_bucket, @@ -1250,6 +1531,7 @@ def _assert_recorded_object_holds_these_bytes( def _superseding_storage( recorded_spec: dict[str, Any], *, + recorded_r2: RecordedR2Object | None, new_r2: dict[str, Any] | None, superseded_at: str, ) -> dict[str, Any]: @@ -1258,18 +1540,16 @@ def _superseding_storage( ``storage.r2`` only ever names the object that holds the entry's current bytes. The superseded block is appended, oldest first, to ``storage.previous_r2`` so the earlier bytes stay addressable by the URI - archived witness records already pin. + archived witness records already pin. An entry that was never published has + no object to supersede, and gets no ``previous_r2`` key. """ storage = dict(_recorded_storage(recorded_spec)) previous = storage.get("previous_r2") entries = list(previous) if isinstance(previous, list) else [] - recorded_r2 = _recorded_r2(recorded_spec) - if recorded_r2: - entry = dict(recorded_r2) - recorded_sha256, _recorded_filename = _r2_key_identity(recorded_r2) - if recorded_sha256: - entry["sha256"] = recorded_sha256 - if recorded_spec.get("sha256") == recorded_sha256: + if recorded_r2 is not None: + entry = dict(_recorded_r2(recorded_spec)) + entry["sha256"] = recorded_r2.sha256 + if recorded_spec.get("sha256") == recorded_r2.sha256: # Only carry metadata the superseded key agrees with: a manifest # can arrive here already describing the new bytes. for field in ("size_bytes", "fetched_at", "source_url"): @@ -1278,7 +1558,8 @@ def _superseding_storage( entry[field] = value entry["superseded_at"] = superseded_at entries.append(entry) - storage["previous_r2"] = entries + if entries: + storage["previous_r2"] = entries if new_r2 is None: storage.pop("r2", None) else: @@ -1319,42 +1600,45 @@ def _upsert_manifest( } recorded_spec = _manifest_file_spec(payload, year) recorded_storage = _recorded_storage(recorded_spec) - recorded_r2 = _recorded_r2(recorded_spec) + identity = _recorded_identity(recorded_spec, manifest_path=manifest_path, year=year) new_r2 = r2_location.to_dict() if r2_location is not None else None - if recorded_r2 and _r2_holds_these_bytes( - recorded_r2, sha256=sha256, filename=filename - ): - # A recorded storage.r2 block for these exact bytes is historical - # truth: archived witness records pin raw R2 URLs by hash. Re-fetching - # under a renamed bucket copies bytes; it does not restate where the - # bytes were first published (PolicyEngine/chronicle#143, mechanism 3). - file_entry["storage"] = {**recorded_storage, "r2": recorded_r2} - elif recorded_r2: + holds = identity is not None and identity.holds(sha256=sha256, filename=filename) + if identity is not None and not holds and not record_revision: # Different bytes under the same vintage. The guard in # fetch_source_artifact refuses this without --record-revision; repeat # the check here so no caller can reach a false-provenance write. - if not record_revision: - raise SourceArtifactRevisionError( - _revision_error_message( - manifest_path=manifest_path, - year=year, - filename=filename, - recorded_spec=recorded_spec, - recorded_r2=recorded_r2, - sha256=sha256, - size_bytes=size_bytes, - r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), - ) + raise SourceArtifactRevisionError( + _revision_error_message( + manifest_path=manifest_path, + year=year, + filename=filename, + identity=identity, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), ) - file_entry["storage"] = _superseding_storage( + ) + if holds and identity.r2 is not None: + # A recorded storage.r2 block for these exact bytes is historical + # truth: archived witness records pin raw R2 URLs by hash. Re-fetching + # under a renamed bucket copies bytes; it does not restate where the + # bytes were first published (PolicyEngine/chronicle#143, mechanism 3). + storage = {**recorded_storage, "r2": _recorded_r2(recorded_spec)} + elif identity is not None and not holds: + storage = _superseding_storage( recorded_spec, + recorded_r2=identity.r2, new_r2=new_r2, superseded_at=fetched_at, ) elif new_r2 is not None: - file_entry["storage"] = {**recorded_storage, "r2": new_r2} - elif recorded_storage: - file_entry["storage"] = dict(recorded_storage) + storage = {**recorded_storage, "r2": new_r2} + else: + storage = dict(recorded_storage) + # An entry that has no storage to record carries no empty block: a + # revision over a never-published entry supersedes nothing. + if storage: + file_entry["storage"] = storage payload["files"][year] = file_entry manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False), @@ -1439,20 +1723,27 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: if errors: return refuse() - recorded_r2 = _recorded_r2(spec) - if recorded_r2 and not _r2_holds_these_bytes( - recorded_r2, sha256=sha256_actual or "", filename=filename + try: + recorded_r2 = _validated_recorded_r2( + spec, manifest_path=manifest_path, year=year + ) + except SourceArtifactManifestError as error: + # A block that does not name one object cannot be treated as history, + # and publishing under it would ship whichever field was read. + return refuse(f"recorded_r2_locator_invalid:{error}") + if recorded_r2 is not None and (recorded_r2.sha256, recorded_r2.filename) != ( + sha256_actual or "", + Path(filename).name, ): # The recorded object is addressed by different bytes, so it is not # this file's history. Uploading anyway would either publish under a # key that misdescribes its content or restate a URI that belongs to # the superseded bytes. Registering a publisher revision is # `fetch-artifact --record-revision`, not a publish-time rewrite. - recorded_sha256, recorded_filename = _r2_key_identity(recorded_r2) return refuse( "recorded_r2_identity_mismatch:" - f"recorded_sha256={recorded_sha256 or 'unknown'}:" - f"recorded_filename={recorded_filename or 'unknown'}:" + f"recorded_sha256={recorded_r2.sha256}:" + f"recorded_filename={recorded_r2.filename}:" f"local_sha256={sha256_actual}:" f"local_filename={Path(filename).name}" ) @@ -1470,7 +1761,7 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: package_path=manifest_path, ), ) - recorded_bucket = recorded_r2.get("bucket") + recorded_bucket = recorded_r2.bucket if recorded_r2 is not None else None if recorded_bucket and recorded_bucket != location.bucket: # The recorded bucket is preserved history. Publishing the same bytes # into a renamed bucket is a backfill copy, not a restatement, so the @@ -1479,7 +1770,7 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: "recorded_r2_bucket_is_preserved_history:" f"recorded={recorded_bucket}:requested={location.bucket}" ) - recorded_key = recorded_r2.get("key") + recorded_key = recorded_r2.key if recorded_r2 is not None else None if recorded_key and recorded_key != location.key: return refuse( "recorded_r2_key_disagrees_with_country_prefix:" diff --git a/chronicle/harness.py b/chronicle/harness.py index a5646553..906cac55 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -9,6 +9,7 @@ from pathlib import Path from chronicle.artifacts import ( + DEFAULT_MANIFEST_FILENAME, DEFAULT_R2_DERIVED_BUCKET, DEFAULT_R2_RAW_BUCKET, ArtifactFetchReport, @@ -16,7 +17,7 @@ DerivedArtifactPublishReport, R2BootstrapReport, RawArtifactPublishReport, - SourceArtifactRevisionError, + SourceArtifactManifestError, bootstrap_r2_buckets, fetch_source_artifact, inventory_source_artifacts, @@ -339,6 +340,7 @@ def fetch_artifact_file( source_page: str | None = None, table: str | None = None, filename: str | None = None, + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -347,8 +349,9 @@ def fetch_artifact_file( ) -> ArtifactFetchReport: """Fetch/register a raw source artifact and optionally upload it to R2. - Raises :class:`SourceArtifactRevisionError` when the fetched bytes are not - the bytes the manifest's recorded R2 object holds, unless + ``manifest_filename`` selects which of the package directory's manifests + the entry belongs to. Raises :class:`SourceArtifactRevisionError` when the + fetched bytes are not the bytes that manifest's entry identifies, unless ``record_revision`` opts into registering the publisher revision. """ return fetch_source_artifact( @@ -361,6 +364,7 @@ def fetch_artifact_file( source_page=source_page, table=table, filename=filename, + manifest_filename=manifest_filename, upload_r2=upload_r2, record_revision=record_revision, r2_bucket=r2_bucket, @@ -852,7 +856,7 @@ def main(argv: list[str] | None = None) -> int: artifact_parser = subparsers.add_parser( "fetch-artifact", - help="Fetch/register a raw source artifact and update manifest.yaml", + help="Fetch/register a raw source artifact and update its manifest", ) artifact_parser.add_argument( "--url", @@ -873,13 +877,23 @@ def main(argv: list[str] | None = None) -> int: "--year", type=int, required=True, - help="Artifact vintage year to record in manifest.yaml", + help="Artifact vintage year to record in the manifest", ) artifact_parser.add_argument( "--out-dir", type=Path, required=True, - help="Directory where the raw artifact and manifest.yaml should live", + help="Directory where the raw artifact and its manifest should live", + ) + artifact_parser.add_argument( + "--manifest", + default=DEFAULT_MANIFEST_FILENAME, + help=( + "Manifest filename inside --out-dir. A publisher directory that " + "feeds several source packages keeps one manifest each, and the " + "entry being revised lives in exactly one of them. Defaults to " + f"{DEFAULT_MANIFEST_FILENAME}." + ), ) artifact_parser.add_argument( "--dataset", @@ -1361,13 +1375,14 @@ def main(argv: list[str] | None = None) -> int: source_page=args.source_page, table=args.table, filename=args.filename, + manifest_filename=args.manifest, upload_r2=args.upload_r2, record_revision=args.record_revision, r2_bucket=args.r2_bucket, r2_prefix=args.r2_prefix, wrangler_command=args.wrangler_command, ) - except SourceArtifactRevisionError as error: + except SourceArtifactManifestError as error: print(f"error: {error}", file=sys.stderr) return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 93b4d429..4629e06c 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -11,6 +11,8 @@ from chronicle.cli import main as cli_main from chronicle.artifacts import ( + MalformedManifestError, + RecordedR2LocatorError, SourceArtifactRevisionError, build_artifact_key, build_artifact_rows, @@ -1185,3 +1187,417 @@ def test_a_recorded_block_that_only_carries_a_uri_is_still_recognized( _serve(monkeypatch, SECOND_PUBLICATION) with pytest.raises(SourceArtifactRevisionError): _fetch_republished(output_dir, wrangler) + + +# --------------------------------------------------------------------------- +# Manifest addressing, entry identity, and recorded locators +# +# Everything below concerns the state a fetch reads before it writes: which +# manifest it reads, what that manifest's entry says its vintage holds, and +# whether the recorded R2 block names one object or two. +# --------------------------------------------------------------------------- + +TRADITIONAL_MANIFEST = "manifest_traditional_source_package.yaml" +ROTH_MANIFEST = "manifest_roth_source_package.yaml" + + +def _publish(tmp_path, name, content): + """Write bytes a fetch can read as a local publisher path.""" + path = tmp_path / "publisher" / name + path.parent.mkdir(parents=True, exist_ok=True) + path.write_bytes(content) + return path + + +def _fetch_local(output_dir, source_path, *, package_id="soi-table-5", **kwargs): + return fetch_source_artifact( + str(source_path), + source_id="irs_soi", + package_id=package_id, + year=2022, + output_dir=output_dir, + **kwargs, + ) + + +def _entry(manifest_path): + return yaml.safe_load(manifest_path.read_text())["files"][2022] + + +def test_fetch_artifact_writes_the_manifest_it_was_given(tmp_path): + """One publisher directory, two source packages, two manifests. + + db/data/irs_soi/ira_contributions keeps the traditional and Roth IRA + packages side by side. A fetch that always wrote manifest.yaml would write + a third manifest neither package reads. + """ + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + roth = _publish(tmp_path, "22in06ira.xlsx", b"roth IRA table") + + _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + manifest_filename=TRADITIONAL_MANIFEST, + ) + _fetch_local( + package, + roth, + package_id="soi-ira-roth-contributions-2022", + manifest_filename=ROTH_MANIFEST, + ) + + assert sorted(path.name for path in package.glob("manifest*.yaml")) == [ + ROTH_MANIFEST, + TRADITIONAL_MANIFEST, + ] + assert not (package / "manifest.yaml").exists() + assert _entry(package / TRADITIONAL_MANIFEST)["filename"] == "22in05ira.xlsx" + assert _entry(package / ROTH_MANIFEST)["filename"] == "22in06ira.xlsx" + assert _entry(package / TRADITIONAL_MANIFEST)["sha256"] == ( + hashlib.sha256(b"traditional IRA table").hexdigest() + ) + + +def test_a_revision_is_refused_in_the_manifest_that_records_it(tmp_path): + """The IRA revision workflow the docs cite, on a two-manifest package.""" + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + manifest_filename=TRADITIONAL_MANIFEST, + ) + recorded = (package / TRADITIONAL_MANIFEST).read_bytes() + + # The IRS re-publishes under the same URL and vintage. + traditional.write_bytes(b"traditional IRA table, revised rows") + + with pytest.raises(SourceArtifactRevisionError) as raised: + _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + manifest_filename=TRADITIONAL_MANIFEST, + ) + + assert TRADITIONAL_MANIFEST in str(raised.value) + assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded + assert (package / "22in05ira.xlsx").read_bytes() == b"traditional IRA table" + + # Without the flag the same fetch addresses a manifest that has no entry to + # protect -- which is exactly why the flag exists. + report = _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + ) + + assert report.valid + assert report.manifest_path.endswith("manifest.yaml") + assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded + + +@pytest.mark.parametrize( + "manifest_filename", + ["../manifest.yaml", "nested/manifest.yaml", "", " ", ".", ".."], +) +def test_a_manifest_name_must_stay_inside_the_package(tmp_path, manifest_filename): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "table.xlsx", b"table") + + with pytest.raises(ValueError, match="inside the package directory"): + _fetch_local(package, source, manifest_filename=manifest_filename) + + assert not package.exists() + + +def test_fetch_artifact_cli_targets_the_named_manifest(tmp_path, capsys): + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + argv = [ + "fetch-artifact", + "--url", + str(traditional), + "--source-id", + "irs_soi", + "--package-id", + "soi-ira-traditional-contributions-2022", + "--year", + "2022", + "--out-dir", + str(package), + "--manifest", + TRADITIONAL_MANIFEST, + ] + + assert harness_main(argv) == 0 + payload = json.loads(capsys.readouterr().out) + + assert payload["manifest_path"].endswith(TRADITIONAL_MANIFEST) + assert not (package / "manifest.yaml").exists() + + traditional.write_bytes(b"traditional IRA table, revised rows") + + assert harness_main(argv) == 1 + assert TRADITIONAL_MANIFEST in capsys.readouterr().err + + +# --------------------------------------------------------------------------- +# Identity without a recorded R2 object +# --------------------------------------------------------------------------- + + +def _failing_wrangler(tmp_path, log): + wrangler = tmp_path / "failing-wrangler" + wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\nexit 1\n") + wrangler.chmod(0o755) + return wrangler + + +def test_a_registered_entry_is_protected_before_it_is_ever_published(tmp_path): + """No storage.r2 yet is not no identity: the entry declares its bytes.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5, first publication") + first = _fetch_local(package, source, upload_r2=False) + recorded = (package / "manifest.yaml").read_bytes() + + assert "storage" not in _entry(package / "manifest.yaml") + + # Same bytes: an ordinary repeated fetch, not a revision. + assert _fetch_local(package, source, upload_r2=False).sha256 == first.sha256 + + source.write_bytes(b"IRA table 5, silently re-published") + with pytest.raises(SourceArtifactRevisionError) as raised: + _fetch_local(package, source, upload_r2=False) + + message = str(raised.value) + assert first.sha256 in message + assert hashlib.sha256(b"IRA table 5, silently re-published").hexdigest() in message + assert "size_bytes=30" in message + assert "--record-revision" in message + assert (package / "manifest.yaml").read_bytes() == recorded + assert (package / "22in05ira.xlsx").read_bytes() == ( + b"IRA table 5, first publication" + ) + + +def test_a_failed_upload_does_not_disable_revision_protection(tmp_path): + """The state #225 hit: bytes registered, upload failed, no storage.r2.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + log = tmp_path / "wrangler.log" + wrangler = _failing_wrangler(tmp_path, log) + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5, first publication") + + report = _fetch_local( + package, source, upload_r2=True, wrangler_command=str(wrangler) + ) + recorded = (package / "manifest.yaml").read_bytes() + + assert report.errors == ("r2_upload_failed",) + assert "storage" not in _entry(package / "manifest.yaml") + + source.write_bytes(b"IRA table 5, silently re-published") + with pytest.raises(SourceArtifactRevisionError): + _fetch_local(package, source, upload_r2=True, wrangler_command=str(wrangler)) + + assert (package / "manifest.yaml").read_bytes() == recorded + + +def test_record_revision_over_an_unpublished_entry_supersedes_nothing(tmp_path): + """There is no object to keep, so the entry gets no previous_r2 key.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5, first publication") + _fetch_local(package, source, upload_r2=False) + + source.write_bytes(b"IRA table 5, silently re-published") + report = _fetch_local(package, source, upload_r2=False, record_revision=True) + revised = _entry(package / "manifest.yaml") + + assert report.valid + assert revised["sha256"] == ( + hashlib.sha256(b"IRA table 5, silently re-published").hexdigest() + ) + assert "storage" not in revised + + +# --------------------------------------------------------------------------- +# Recorded locator cross-checks +# --------------------------------------------------------------------------- + + +def _recorded_package(tmp_path, content=b"IRA table 5, first publication"): + """A package whose entry records a published, content-addressed object.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + wrangler = _wrangler_stub(tmp_path, tmp_path / "wrangler.log") + source = _publish(tmp_path, "22in05ira.xlsx", content) + report = _fetch_local( + package, source, upload_r2=True, wrangler_command=str(wrangler) + ) + return package, source, report + + +def _rewrite_recorded_r2(package, mutate, manifest="manifest.yaml"): + manifest_path = package / manifest + payload = yaml.safe_load(manifest_path.read_text()) + mutate(payload["files"][2022]["storage"]) + manifest_path.write_text(yaml.safe_dump(payload, sort_keys=False)) + return manifest_path + + +def _other_sha256(): + return hashlib.sha256(b"some other object entirely").hexdigest() + + +def _contradict_key(storage): + key = storage["r2"]["key"] + storage["r2"]["key"] = key.replace(key.split("/")[-2], _other_sha256()) + + +def _contradict_bucket(storage): + storage["r2"]["bucket"] = "some-other-bucket" + + +def _contradict_provider(storage): + storage["r2"]["provider"] = "s3" + + +def _mangle_uri(storage): + storage["r2"]["uri"] = "r2:/ledger-raw-missing-a-slash" + + +def _drop_the_locator(storage): + storage["r2"] = {"provider": "r2", "bucket": "ledger-raw"} + + +def _flatten_the_key(storage): + storage["r2"]["key"] = "raw/irs_soi/22in05ira.xlsx" + storage["r2"]["uri"] = f"r2://ledger-raw/{storage['r2']['key']}" + + +@pytest.mark.parametrize( + ("mutate", "expected"), + [ + pytest.param(_contradict_key, "contradicts uri", id="key-vs-uri"), + pytest.param(_contradict_bucket, "contradicts uri", id="bucket-vs-uri"), + pytest.param(_contradict_provider, "contradicts uri", id="provider-vs-uri"), + pytest.param(_mangle_uri, "is not provider://bucket/key", id="uri-shape"), + pytest.param(_drop_the_locator, "records no key", id="no-locator"), + pytest.param( + _flatten_the_key, "is not content-addressed", id="not-content-addressed" + ), + ], +) +def test_a_recorded_block_that_names_two_objects_is_refused(tmp_path, mutate, expected): + """A contradictory locator is an error, never a silently preserved block. + + The key-vs-uri case is the one that used to pass: identity was read from + the key alone, so a block whose uri named different bytes was carried + forward verbatim, and the manifest kept publishing a URI for an object it + no longer described. + """ + package, source, _ = _recorded_package(tmp_path) + manifest_path = _rewrite_recorded_r2(package, mutate) + recorded = manifest_path.read_bytes() + + # Identical bytes: the fetch would otherwise preserve the recorded block. + with pytest.raises(RecordedR2LocatorError) as raised: + _fetch_local(package, source, upload_r2=False) + + assert expected in str(raised.value) + assert manifest_path.read_bytes() == recorded + + +def test_a_malformed_storage_block_is_not_treated_as_absent(tmp_path): + package, source, _ = _recorded_package(tmp_path) + manifest_path = _rewrite_recorded_r2( + package, lambda storage: storage.update({"r2": ["r2://ledger-raw/raw/key"]}) + ) + recorded = manifest_path.read_bytes() + + with pytest.raises(MalformedManifestError, match="must be a mapping"): + _fetch_local(package, source, upload_r2=False) + + assert manifest_path.read_bytes() == recorded + + +def test_publish_raw_refuses_a_contradictory_recorded_block(tmp_path): + """Nothing is uploaded under a block that does not name one object.""" + package, _, _ = _recorded_package(tmp_path) + log = tmp_path / "publish.log" + wrangler = _wrangler_stub(tmp_path, log) + manifest_path = _rewrite_recorded_r2(package, _contradict_key) + recorded = manifest_path.read_bytes() + + report = publish_source_artifacts(package, wrangler_command=str(wrangler)) + + assert not report.valid + assert report.entries[0].upload is None + assert report.entries[0].errors[0].startswith("recorded_r2_locator_invalid:") + assert "contradicts uri" in report.entries[0].errors[0] + assert not log.exists() + assert manifest_path.read_bytes() == recorded + + +# --------------------------------------------------------------------------- +# Malformed manifests +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize( + "document", + [ + pytest.param("- one entry\n- another\n", id="list"), + pytest.param("a bare scalar\n", id="scalar"), + pytest.param("files: [\n", id="unparseable"), + ], +) +def test_a_malformed_manifest_is_refused_before_anything_is_fetched(tmp_path, document): + """Not an absent manifest: refusing it protects what it still records. + + The publisher path does not exist, so reaching the fetch at all would raise + FileNotFoundError instead. Getting MalformedManifestError is what says the + manifest was read and refused first. + """ + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text(document) + + with pytest.raises(MalformedManifestError): + _fetch_local(package, tmp_path / "publisher" / "never-read.xlsx") + + assert manifest_path.read_text() == document + assert list(package.iterdir()) == [manifest_path] + + +@pytest.mark.parametrize("document", ["", "\n", "{}\n", "# only a comment\n"]) +def test_an_empty_manifest_still_reads_as_absent(tmp_path, document): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + (package / "manifest.yaml").write_text(document) + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5") + + report = _fetch_local(package, source, upload_r2=False) + + assert report.valid + assert _entry(package / "manifest.yaml")["filename"] == "22in05ira.xlsx" + + +def test_a_malformed_manifest_is_reported_by_inventory_and_publish(tmp_path): + """Neither sweep may crash on, or silently skip, a document it cannot read.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + (package / "manifest.yaml").write_text("- not a mapping\n") + + inventory = inventory_source_artifacts(package) + published = publish_source_artifacts(package) + + assert not inventory.valid + assert inventory.entries == () + assert "must be a YAML mapping" in inventory.errors[0] + assert not published.valid + assert published.entries == () + assert "must be a YAML mapping" in published.errors[0] From 03bbe5ce456adbb5cfe08ca14dfcae2864aeb562 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:38:08 -0400 Subject: [PATCH 17/42] Document identity, manifest selection and locator checks What a manifest entry identifies (declared sha256 from registration, the content-addressed key once published), which manifest a fetch addresses in a package that keeps more than one, what a recorded storage.r2 block has to say for itself, and that load-supabase-mirror takes its schema default from CHRONICLE_SCHEMA like every other reader. Co-Authored-By: Claude Fable 5.1 --- README.md | 8 ++++-- docs/agent-source-package-harness.md | 11 +++++--- docs/storage-architecture.md | 41 ++++++++++++++++++++++++++-- 3 files changed, 51 insertions(+), 9 deletions(-) diff --git a/README.md b/README.md index 6b6e170c..7c526a97 100644 --- a/README.md +++ b/README.md @@ -356,7 +356,9 @@ bunx wrangler login uv run chronicle bootstrap-r2 # Fetch/register a source artifact, write db/data/.../manifest.yaml, and upload -# the exact bytes to R2 when Wrangler is authenticated: +# the exact bytes to R2 when Wrangler is authenticated. Pass --manifest when the +# package directory keeps more than one manifest (ira_contributions keeps a +# traditional and a Roth one): uv run chronicle fetch-artifact \ --url https://www.irs.gov/pub/irs-soi/23in12ms.xls \ --source-id irs_soi \ @@ -368,7 +370,9 @@ uv run chronicle fetch-artifact \ --upload-r2 # Re-fetching is safe: identical bytes keep the recorded storage.r2 block, and -# bytes that disagree with it are refused. When a publisher has re-published +# bytes that disagree with what the entry identifies -- its declared sha256, or +# its recorded content-addressed key once published -- are refused. When a +# publisher has re-published # under the same URL and vintage, register the revision explicitly — the new # bytes get their own content-addressed key and the superseded object is kept # in storage.previous_r2: diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 11200ebb..56d65bb7 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -16,15 +16,18 @@ register raw source files with `uv run chronicle fetch-artifact` before authorin selectors. This writes the local artifact, captures checksum and retrieval metadata in `manifest.yaml`, and can upload the exact bytes to the private raw R2 bucket (`ledger-raw` today; overridable with `CHRONICLE_R2_RAW_BUCKET`) when -Wrangler is authenticated. Agents can audit the local +Wrangler is authenticated. A publisher directory that feeds several source +packages keeps one manifest each, so pass `--manifest ` to address +the right one. Agents can audit the local artifact registry with `uv run chronicle inventory-artifacts --root db/data`. For already-downloaded manifest artifacts, agents should run `uv run chronicle publish-raw --root db/data` to upload checksum-verified bytes to R2 and write `storage.r2` metadata back into each manifest entry. -Both commands treat a recorded `storage.r2` block as a claim about specific -bytes, because raw keys are content-addressed. Re-fetching or publishing bytes -the recorded object does not hold is refused; when a publisher has re-published +Both commands treat a manifest entry as a claim about specific bytes: by its +declared `sha256` from the moment it is registered, and by the content-addressed +key of its recorded `storage.r2` block once it is published. Re-fetching or +publishing bytes the entry does not identify is refused; when a publisher has re-published under the same URL and vintage, register the revision with `uv run chronicle fetch-artifact ... --record-revision`, which stores the new bytes under their own key and keeps the superseded object in diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index 6f4cfb00..3607e82b 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -99,8 +99,12 @@ name. Publishers do not always honor that: on 2026-09-02 the IRS re-published `22in05ira.xlsx` and `22in06ira.xlsx` under their existing URLs (PolicyEngine/chronicle#225). -`fetch-artifact` therefore compares the recorded key's `{sha256}/{filename}` -tail with the bytes it just fetched, before it writes anything: +`fetch-artifact` therefore compares the entry's recorded identity with the +bytes it just fetched, before it writes anything. That identity is the recorded +key's `{sha256}/{filename}` tail once the entry has been published, and the +entry's own declared `sha256` before then — an entry that was registered +without an upload, or whose upload failed, still identifies its bytes, and gets +the same protection: - **Identical** — the recorded block is preserved exactly, whichever bucket is configured now. Re-fetching after the bucket rename copies bytes; it does not @@ -153,6 +157,31 @@ as history. A local file the recorded object does not hold is reported as `recorded_r2_identity_mismatch` and nothing is uploaded: registering a revision is a fetch-time decision, not a publish-time rewrite. +### Which manifest + +Most packages keep one `manifest.yaml`. A publisher directory that feeds +several source packages keeps one manifest each — +`db/data/irs_soi/ira_contributions/` holds +`manifest_traditional_source_package.yaml` beside +`manifest_roth_source_package.yaml` — and the entry being revised lives in +exactly one of them. `fetch-artifact --manifest ` selects it; +defaulting to `manifest.yaml` there would write a third manifest neither +package reads, and the recorded block would never be compared at all. The name +must be a filename inside `--out-dir`, not a path. + +### What a recorded block has to say + +A `storage.r2` block's `provider`, `bucket`, `key` and `uri` all describe one +object, so every field that is present is cross-checked against every other: +the key against the URI's path, the bucket against its authority, the provider +against its scheme, and the resulting key against the content-addressed +`{sha256}/{filename}` shape. A block whose fields disagree does not answer +"which bytes does this entry claim R2 holds", so it is an error rather than +something to preserve or publish under. Likewise a manifest that parses as +anything other than a mapping is refused rather than treated as absent — +reading it as absent would let the next fetch replace the file with a single +entry. + ## Relational Registry Contract The hosted `chronicle` schema should be the lookup surface for Chronicle, not the place @@ -254,7 +283,7 @@ what a naive fallback would do: | `CHRONICLE_SOURCE_ARTIFACT_FETCH` | `LEDGER_SOURCE_ARTIFACT_FETCH` | Fetch a missing manifest artifact from its `source_url` during a build | | `CHRONICLE_PE_US_DATA_ROOT` | `LEDGER_PE_US_DATA_ROOT` | Local checkout root for PE US source inventory | | `CHRONICLE_PE_UK_DATA_ROOT` | `LEDGER_PE_UK_DATA_ROOT` | Local checkout root for PE UK source inventory | -| `CHRONICLE_SCHEMA` | `POLICYENGINE_LEDGER_SCHEMA` | Postgres schema the Supabase client reads and writes | +| `CHRONICLE_SCHEMA` | `POLICYENGINE_LEDGER_SCHEMA`, `LEDGER_SCHEMA` | Postgres schema the Supabase client reads and `load-supabase-mirror` writes; defaults to `ledger` | | `CHRONICLE_R2_RAW_BUCKET` | `LEDGER_R2_RAW_BUCKET` | Raw R2 archive bucket; defaults to `ledger-raw` | | `CHRONICLE_R2_DERIVED_BUCKET` | `LEDGER_R2_DERIVED_BUCKET` | Derived R2 archive bucket; defaults to `ledger-derived` | @@ -272,6 +301,12 @@ the variable that overrides the schema; the schema still defaults to `ledger`, and the mirror table names are unchanged. Those move in a later slice coordinated with the CI writers. +Every reader of the setting resolves it the same way, at call time. That +includes the writer: `load-supabase-mirror` takes its `--schema` default from +`CHRONICLE_SCHEMA`, so setting the variable to rehearse a cutover moves the +mirror load with the client rather than leaving it pointed at `ledger`. An +explicit `--schema` still wins. + ## Bucket Cutover Chronicle's operational stores migrate by dual-run From 6f1ea254b9188dec9d7ba26977e7617d723717ea Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:39:25 -0400 Subject: [PATCH 18/42] Name the strict storage reader for the pair it belongs to _validated_recorded_storage sits next to _validated_recorded_r2 and above its only caller, instead of a _recorded_storage_block that read like a variant of the lenient accessor. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 44 +++++++++++++++++++++--------------------- 1 file changed, 22 insertions(+), 22 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 1b3f202a..1a81b918 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1244,9 +1244,9 @@ def _read_manifest(manifest_path: Path) -> dict[str, Any]: try: payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) except yaml.YAMLError as exc: - raise MalformedManifestError(f"{manifest_path} is not valid YAML: {exc}") from ( - exc - ) + raise MalformedManifestError( + f"{manifest_path} is not valid YAML: {exc}" + ) from exc if payload is None: return {} if not isinstance(payload, dict): @@ -1297,6 +1297,24 @@ def _split_r2_uri(uri: str) -> tuple[str, str, str] | None: return (provider, bucket, key) +def _validated_recorded_storage( + spec: Any, + *, + manifest_path: Path, + year: Any, +) -> dict[str, Any]: + """Return the entry's ``storage`` mapping, refusing a malformed one.""" + if not isinstance(spec, dict) or "storage" not in spec: + return {} + storage = spec["storage"] + if not isinstance(storage, dict): + raise MalformedManifestError( + f"{manifest_path} entry {year!r} storage must be a mapping; it is a " + f"{type(storage).__name__}." + ) + return storage + + def _validated_recorded_r2( spec: Any, *, @@ -1312,7 +1330,7 @@ def _validated_recorded_r2( field and trusting the rest is what lets a block that says two different things survive a preserve or a publish. """ - storage = _recorded_storage_block(spec, manifest_path=manifest_path, year=year) + storage = _validated_recorded_storage(spec, manifest_path=manifest_path, year=year) if "r2" not in storage: return None block = storage["r2"] @@ -1393,24 +1411,6 @@ def _validated_recorded_r2( ) -def _recorded_storage_block( - spec: Any, - *, - manifest_path: Path, - year: Any, -) -> dict[str, Any]: - """Return the entry's ``storage`` mapping, refusing a malformed one.""" - if not isinstance(spec, dict) or "storage" not in spec: - return {} - storage = spec["storage"] - if not isinstance(storage, dict): - raise MalformedManifestError( - f"{manifest_path} entry {year!r} storage must be a mapping; it is a " - f"{type(storage).__name__}." - ) - return storage - - def _recorded_identity( spec: Any, *, From f742d08adbfe01fc74b0a016c122e206b6f05232 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 13:40:10 -0400 Subject: [PATCH 19/42] Record what each round-2 fix does and how it was reproduced Co-Authored-By: Claude Fable 5.1 --- PROGRESS.md | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 50 insertions(+) diff --git a/PROGRESS.md b/PROGRESS.md index e51ea2ca..79bafcd9 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -118,6 +118,56 @@ with a regression test on this branch. Plan, in dependency order: and uri; every key is content-addressed; every declared `sha256`/`filename` agrees with its key tail; no `uri` contradicts its `key`. Strict locator validation therefore refuses nothing that is tracked today. +- All seven findings are applied, each with a regression test, and each + reproduced against this branch's previous head (`34d1d0f`) first. + +### What each fix does + +1. `chronicle/env.py` gains `default_chronicle_schema()`: one home for the + `CHRONICLE_SCHEMA` -> `POLICYENGINE_LEDGER_SCHEMA` -> `LEDGER_SCHEMA` -> + `"ledger"` ladder. `load_supabase_mirror`, its harness wrapper and the + `--schema` CLI default all resolve through it when no schema is supplied; + an explicit `--schema` still wins. Defaults unchanged. +2. `chronicle/consumer_contract.py` matches the whole final dot-segment of a + `source_record_id` against both `ledger_derived` and `chronicle_derived`. +3. `fetch-artifact --manifest ` selects which of a package's + manifests the entry belongs to (default `manifest.yaml`); the name must be + a filename inside `--out-dir`. +4. Revision protection now compares against the entry's recorded identity -- + the recorded key's `{sha256}/{filename}` once published, the declared + `sha256` before that -- so a registered-but-unpublished entry, or one whose + upload failed, is protected exactly like a published one. +5. `_validated_recorded_r2` cross-checks every supplied locator field against + every other and against the content-addressed key shape. A contradiction is + `RecordedR2LocatorError` at fetch time and `recorded_r2_locator_invalid` at + publish time, never a preserved block. +6. `_read_manifest` refuses a non-mapping or unparseable document + (`MalformedManifestError`) before the publisher is read at all; + `inventory-artifacts` and `publish-raw` report it instead of crashing. +7. `db.supabase_client` resolves both schemas per call rather than at import, + and `tests/conftest.py` strips the rename window in `pytest_configure`, so + no module can read or warn from an operator's shell during collection. + +All four refusals share a `SourceArtifactManifestError` base, so the +`fetch-artifact` CLI reports every one as exit 1 with nothing written. + +### Reproduced against `34d1d0f` (the round-1 head) + +Running the same operations against a checkout of the previous head: + +1. `load_supabase_mirror` default `schema='ledger'`; with + `CHRONICLE_SCHEMA=chronicle_probe` the load still reports `schema='ledger'`. +2. `'.chronicle_derived'.endswith('.ledger_derived')` is False: the boundary + never fired for the chronicle spelling. +3. `fetch_source_artifact()` rejects `manifest_filename` as an unexpected + keyword; a fetch into `ira_contributions/` writes `manifest.yaml`. +4. A fetch of different bytes over a registered (unpublished) entry was + accepted silently: the entry's `sha256` was rewritten with no refusal. +5. A block whose `key` and `uri` named different objects was preserved + verbatim, key sha `c63744a4...` beside uri sha `1e9b3fdb...`. +6. A list-valued `manifest.yaml` was overwritten by the fetch. +7. Importing `db.supabase_client` under `LEDGER_SCHEMA=zzz` bound + `LEDGER_SCHEMA='zzz'` and emitted a `FutureWarning` at collection. ## Verification From ee9bdb356a4c33f65592531d636947050b45671c Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 14:15:39 -0400 Subject: [PATCH 20/42] Say which fields a recorded block actually needs The message named key and bucket while the check also requires provider, so a hand-written block missing it read as a contradiction between the error and the rule. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 1a81b918..00807ade 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -1388,7 +1388,8 @@ def _validated_recorded_r2( if missing: raise RecordedR2LocatorError( f"{where}: records no {', '.join(missing)}. A recorded block has to " - "locate its object, by key and bucket or by uri." + "locate its object: provider, bucket and key, or a uri that " + "supplies them." ) segments = key.split("/") From 53d038ae0b43b021f1c462e2cd76cf344f51ebf0 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 14:16:15 -0400 Subject: [PATCH 21/42] Say what --record-revision is an opt-in over The flag's help still described the recorded R2 object alone, which is no longer the only identity a fetch is refused against. Co-Authored-By: Claude Fable 5.1 --- chronicle/harness.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/chronicle/harness.py b/chronicle/harness.py index 906cac55..ee7f6ef9 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -923,8 +923,9 @@ def main(argv: list[str] | None = None) -> int: "Register a publisher revision: the fetched bytes get their own " "content-addressed key under the configured bucket and the " "superseded object moves to storage.previous_r2. Without this " - "flag, bytes that disagree with the recorded R2 object are " - "refused." + "flag, bytes that disagree with what the entry identifies -- its " + "declared sha256, or its recorded content-addressed key once " + "published -- are refused." ), ) artifact_parser.add_argument( From 3fceb40f96da35786a573b53ae6340e08ec5f1d7 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 14:16:41 -0400 Subject: [PATCH 22/42] Say when the manifest guards actually run They run before the publisher is read, and again before the manifest is rewritten; the base class claimed the second one also predated the cache write. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 00807ade..58437120 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -78,8 +78,10 @@ def default_r2_derived_bucket() -> str: class SourceArtifactManifestError(RuntimeError): """A manifest refuses the write a fetch is about to make. - Every subclass is raised before anything is downloaded, cached, uploaded or - rewritten, so a refusal leaves the package exactly as it was. + The checks that raise these run before the publisher is read, so an + ordinary refusal costs nothing and leaves the package exactly as it was. + They are repeated immediately before the manifest is rewritten, so no + caller can reach a false-provenance write by another route. """ From aa768ddf0045ae52f83463d3189263fe7411aa2b Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 00:22:35 -0400 Subject: [PATCH 23/42] Drop a blank line ruff format rejects after the rebase onto #228 --- chronicle/artifacts.py | 1 - 1 file changed, 1 deletion(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 58437120..0a746ca7 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -28,7 +28,6 @@ from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain - R2_RAW_BUCKET_ENV = "CHRONICLE_R2_RAW_BUCKET" R2_DERIVED_BUCKET_ENV = "CHRONICLE_R2_DERIVED_BUCKET" From 70e46f9d6a0ca057f71ac33f679cd9089d4ee9b0 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 00:57:46 -0400 Subject: [PATCH 24/42] Gate round: manifest-name refusals reach the CLI, files blocks are validated before I/O, renames are refused by name, preserved-bucket objects are skipped not red, stray default manifests are refused - `_manifest_path` raises ManifestNameError (a SourceArtifactManifestError that is still a ValueError), so `chronicle fetch-artifact --manifest ../x` prints the ordinary `error:` line and exits 1 instead of a traceback. - A `files` block that is not a mapping is refused by the fetch before the publisher is read, with the same wording inventory-artifacts and publish-raw use; nothing is overwritten or uploaded first. - Identical bytes under a different filename are refused as a rename, naming both filenames and the --filename that keeps the recorded identity; the refusal holds with --record-revision too, since a rename is not a revision. - publish-raw reports an entry whose recorded object (in another bucket) holds exactly the local bytes as `skipped`, with its recorded location and no upload or rewrite; the report stays valid and gains `skipped_count`, so the documented post-flip sweep exits 0. Docs updated (Bucket Cutover, step 5). - Fetching with the default manifest name into a directory that keeps manifest_*.yaml files and no manifest.yaml is refused (AmbiguousManifestError) naming the candidates; the #225 path is closed without the flag, on the API and the CLI. - previous_r2 doc example carries source_url as the code does. --- chronicle/artifacts.py | 144 ++++++++++++++++++++++++--- docs/storage-architecture.md | 12 ++- tests/test_chronicle_artifacts.py | 159 +++++++++++++++++++++++++++--- 3 files changed, 283 insertions(+), 32 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 0a746ca7..7c3032f0 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -57,13 +57,66 @@ def _manifest_path(output: Path, manifest_filename: str) -> Path: """ name = manifest_filename.strip() if not name or name in (".", "..") or name != Path(name).name: - raise ValueError( + raise ManifestNameError( "Manifest must name a file inside the package directory, not " f"{manifest_filename!r}." ) return output / name +def _sibling_manifests(output: Path) -> list[str]: + """Return the ``manifest_*.yaml`` files a package directory keeps.""" + if not output.is_dir(): + return [] + return sorted( + path.name + for pattern in ("manifest_*.yaml", "manifest_*.yml") + for path in output.glob(pattern) + if path.is_file() + ) + + +def _refuse_a_stray_default_manifest(output: Path, manifest_path: Path) -> None: + """Refuse to create ``manifest.yaml`` beside a package's named manifests. + + A publisher directory that feeds several source packages keeps one + ``manifest_.yaml`` per package and no ``manifest.yaml``. A fetch + that omits ``--manifest`` there would create a third manifest none of the + packages read, and would bypass the revision guard of the one it should + have addressed (PolicyEngine/chronicle#225). + """ + if manifest_path.name != DEFAULT_MANIFEST_FILENAME or manifest_path.exists(): + return + siblings = _sibling_manifests(output) + if not siblings: + return + raise AmbiguousManifestError( + f"{output} keeps {', '.join(siblings)} and no {DEFAULT_MANIFEST_FILENAME}; " + "pass --manifest to name the manifest this fetch records into rather " + f"than creating {DEFAULT_MANIFEST_FILENAME} beside them." + ) + + +def _manifest_files(payload: dict[str, Any], manifest_path: Path) -> dict[str, Any]: + """Return a manifest's ``files`` block, refusing one that is not a mapping. + + ``inventory-artifacts`` and ``publish-raw`` report the same document as + ``files must be a mapping``; a fetch must refuse it before reading the + publisher, or the write fails only after the local artifact has been + overwritten and any upload has run. + """ + files = payload.get("files") + if files is None: + return {} + if not isinstance(files, dict): + raise MalformedManifestError( + f"{manifest_path} files must be a mapping; it parses as a " + f"{type(files).__name__}. Chronicle will not overwrite a manifest " + "it cannot read." + ) + return files + + def default_r2_raw_bucket() -> str: """Resolve the raw bucket: ``$CHRONICLE_R2_RAW_BUCKET`` or the default.""" return env_value(R2_RAW_BUCKET_ENV, default=DEFAULT_R2_RAW_BUCKET) @@ -97,6 +150,19 @@ class SourceArtifactRevisionError(SourceArtifactManifestError): """ +class ManifestNameError(SourceArtifactManifestError, ValueError): + """A manifest name is not a bare filename inside the package directory. + + Also a :class:`ValueError` for callers that validated the name that way + before the CLI learned to report it as an ordinary manifest refusal. + """ + + +class AmbiguousManifestError(SourceArtifactManifestError): + """The default manifest name would create a manifest beside the ones a + package already keeps (PolicyEngine/chronicle#225).""" + + class MalformedManifestError(SourceArtifactManifestError): """A manifest document, or a block inside one, is not a mapping. @@ -366,16 +432,24 @@ class RawArtifactPublishEntry: r2_location: ArtifactStorageLocation | None upload: ArtifactCommandResult | None errors: tuple[str, ...] = () + skipped: str | None = None + + @property + def uploaded(self) -> bool: + """Whether this run uploaded the artifact.""" + return self.upload is not None and self.upload.ok @property def valid(self) -> bool: - """Whether this raw artifact uploaded and was registered.""" - return not self.errors and self.upload is not None and self.upload.ok + """Whether this raw artifact is published: uploaded now, or already + held by the recorded object in a preserved bucket (``skipped``).""" + return not self.errors and (self.skipped is not None or self.uploaded) def to_dict(self) -> dict[str, Any]: """Return a JSON-serializable entry.""" return { "valid": self.valid, + "skipped": self.skipped, "manifest_path": self.manifest_path, "source_id": self.source_id, "package_id": self.package_id, @@ -410,7 +484,10 @@ def counts(self) -> dict[str, int]: return { "manifest_count": len(manifest_paths), "artifact_count": len(self.entries), - "uploaded_count": sum(1 for entry in self.entries if entry.valid), + "uploaded_count": sum(1 for entry in self.entries if entry.uploaded), + "skipped_count": sum( + 1 for entry in self.entries if entry.skipped is not None + ), "failed_count": sum(1 for entry in self.entries if not entry.valid), "r2_link_count": sum( 1 for entry in self.entries if entry.r2_location is not None @@ -569,8 +646,11 @@ def fetch_source_artifact( # Read and validate the entry being written before anything is fetched: a # manifest Chronicle cannot read, or a recorded block that names two # different objects, is a refusal that need not touch the publisher. + _refuse_a_stray_default_manifest(output, manifest_path) + existing_manifest = _read_manifest(manifest_path) + _manifest_files(existing_manifest, manifest_path) recorded_identity = _recorded_identity( - _manifest_file_spec(_read_manifest(manifest_path), year), + _manifest_file_spec(existing_manifest, year), manifest_path=manifest_path, year=year, ) @@ -1513,9 +1593,21 @@ def _assert_recorded_identity_holds_these_bytes( record_revision: bool, ) -> None: """Refuse a publisher revision that has not been opted into.""" - if record_revision or identity is None: + if identity is None or identity.holds(sha256=sha256, filename=filename): return - if identity.holds(sha256=sha256, filename=filename): + if identity.sha256 == sha256: + # The recorded object holds exactly these bytes under another name. A + # rename is not a publisher revision, so --record-revision does not + # apply, and silently adopting the new name would leave the entry's + # filename disagreeing with the key its own storage block records. + raise SourceArtifactRevisionError( + f"{manifest_path} entry {year!r} already records these exact bytes " + f"(sha256={sha256}) as filename={identity.filename}; this fetch " + f"names them {Path(filename).name}. A rename is not a release " + "revision, so --record-revision does not apply. Re-run with " + f"--filename {identity.filename} to keep the recorded identity." + ) + if record_revision: return raise SourceArtifactRevisionError( _revision_error_message( @@ -1764,13 +1856,37 @@ def refuse(reason: str | None = None) -> tuple[RawArtifactPublishEntry, None]: ), ) recorded_bucket = recorded_r2.bucket if recorded_r2 is not None else None - if recorded_bucket and recorded_bucket != location.bucket: - # The recorded bucket is preserved history. Publishing the same bytes - # into a renamed bucket is a backfill copy, not a restatement, so the - # manifest must not be rewritten to point at the new bucket. - return refuse( - "recorded_r2_bucket_is_preserved_history:" - f"recorded={recorded_bucket}:requested={location.bucket}" + if recorded_r2 is not None and recorded_bucket != location.bucket: + # The recorded bucket is preserved history and, per the identity check + # above, its object holds exactly these bytes: the artifact is already + # published. Restating it under the configured bucket would rewrite + # where the bytes were first published (a backfill copy is not a + # restatement), so the entry is reported as skipped with nothing + # uploaded or rewritten. After the bucket-default flip every entry + # published before it takes this path, and the sweep stays green. + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(artifact_path), + sha256=sha256_actual, + size_bytes=size_bytes, + r2_location=ArtifactStorageLocation( + provider="r2", + bucket=recorded_r2.bucket, + key=recorded_r2.key, + ), + upload=None, + errors=(), + skipped=( + "recorded_r2_bucket_is_preserved_history:" + f"recorded={recorded_bucket}:requested={location.bucket}" + ), + ), + None, ) recorded_key = recorded_r2.key if recorded_r2 is not None else None if recorded_key and recorded_key != location.key: diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index 3607e82b..b91d06fa 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -141,6 +141,7 @@ files: sha256: size_bytes: fetched_at: "2026-06-11T14:22:05+00:00" + source_url: https://www.irs.gov/pub/irs-soi/22in05ira.xlsx superseded_at: "2026-09-02T17:04:11+00:00" ``` @@ -318,8 +319,10 @@ deleted, and manifests keep the `storage.r2` URIs they already recorded as historical truth. A backfill copies bytes into the new bucket; it never rewrites where those bytes were first published. `publish-raw` and `fetch-artifact` enforce that: a recorded block that addresses the bytes in hand is preserved -whichever bucket is configured, and `publish-raw` refuses to restate it under a -different one. Bytes that the recorded object does not hold are not that +whichever bucket is configured, and `publish-raw` reports such an entry as +`skipped` (already published under the recorded bucket) rather than restating +it under a different one, so a sweep over a fully published tree stays green +after the flip. Bytes that the recorded object does not hold are not that object's history at all; see [Publisher Revisions](#publisher-revisions). The cutover therefore has one irreversible-looking step that is in fact additive @@ -410,7 +413,10 @@ export CHRONICLE_R2_DERIVED_BUCKET=chronicle-derived New raw publications land in the new bucket from that point. Manifests written before the flip keep pointing at `ledger-raw`, which is why the old bucket stays -readable. +readable. A `publish-raw --root db/data` sweep after the flip reports every +already-published entry as `skipped` with its recorded `ledger-raw` location +(`skipped_count` in the report) and exits 0; only bytes that no recorded object +holds are uploaded, into `chronicle-raw`. ### 6. Set the ledger-era buckets read-only diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 4629e06c..224e73bd 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -11,6 +11,7 @@ from chronicle.cli import main as cli_main from chronicle.artifacts import ( + AmbiguousManifestError, MalformedManifestError, RecordedR2LocatorError, SourceArtifactRevisionError, @@ -246,6 +247,7 @@ def test_publish_source_artifacts_uploads_manifest_entries(tmp_path): "failed_count": 0, "manifest_count": 1, "r2_link_count": 1, + "skipped_count": 0, "uploaded_count": 1, } assert storage["bucket"] == "ledger-raw" @@ -740,11 +742,15 @@ def test_publish_derived_uses_the_configured_bucket(tmp_path, monkeypatch): assert "chronicle-derived/derived/irs_soi/" in log.read_text() -def test_publish_raw_refuses_to_restate_a_recorded_bucket(tmp_path, monkeypatch): +def test_publish_raw_skips_an_object_already_held_by_a_preserved_bucket( + tmp_path, monkeypatch +): """A recorded storage.r2 bucket is preserved history, not a publish target. Archived witness records pin raw R2 URLs by hash, so backfilling the same - bytes into a renamed bucket must not rewrite the manifest. + bytes into a renamed bucket must not rewrite the manifest. The entry is + already published, so the sweep reports it skipped and stays green: after + the bucket-default flip every entry published before it takes this path. """ monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") output_dir = tmp_path / "db" / "data" / "irs_soi" / "soi-table-1-1" @@ -777,17 +783,26 @@ def test_publish_raw_refuses_to_restate_a_recorded_bucket(tmp_path, monkeypatch) wrangler.write_text(f"#!/bin/sh\nprintf '%s\\n' \"$*\" >> {log}\necho ok\n") wrangler.chmod(0o755) + before = manifest_path.read_bytes() report = publish_source_artifacts(output_dir, wrangler_command=str(wrangler)) - unchanged = yaml.safe_load(manifest_path.read_text()) + entry = report.entries[0] - assert not report.valid - assert ( - report.entries[0] - .errors[0] - .startswith("recorded_r2_bucket_is_preserved_history:") + assert report.valid + assert entry.errors == () + assert entry.upload is None + assert entry.skipped == ( + "recorded_r2_bucket_is_preserved_history:" + "recorded=ledger-raw:requested=chronicle-raw" ) + assert entry.r2_location is not None + assert entry.r2_location.bucket == "ledger-raw" + assert entry.r2_location.key == recorded_key + assert entry.to_dict()["skipped"] == entry.skipped + assert report.counts["skipped_count"] == 1 + assert report.counts["uploaded_count"] == 0 + assert report.counts["failed_count"] == 0 assert not log.exists() - assert unchanged["files"][2023]["storage"]["r2"]["bucket"] == "ledger-raw" + assert manifest_path.read_bytes() == before def test_fetch_artifact_keeps_an_already_recorded_bucket(tmp_path, monkeypatch): @@ -1287,17 +1302,76 @@ def test_a_revision_is_refused_in_the_manifest_that_records_it(tmp_path): assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded assert (package / "22in05ira.xlsx").read_bytes() == b"traditional IRA table" - # Without the flag the same fetch addresses a manifest that has no entry to - # protect -- which is exactly why the flag exists. - report = _fetch_local( + # Without the flag the same fetch would address a manifest.yaml that no + # package reads and that protects nothing: the #225 path. It is refused, + # naming the manifests the directory keeps, and nothing is written. + with pytest.raises(AmbiguousManifestError) as stray: + _fetch_local( + package, + traditional, + package_id="soi-ira-traditional-contributions-2022", + ) + + assert TRADITIONAL_MANIFEST in str(stray.value) + assert "--manifest" in str(stray.value) + assert not (package / "manifest.yaml").exists() + assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded + assert (package / "22in05ira.xlsx").read_bytes() == b"traditional IRA table" + + +def test_fetch_artifact_cli_refuses_a_stray_default_manifest(tmp_path, capsys): + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") + _fetch_local( package, traditional, package_id="soi-ira-traditional-contributions-2022", + manifest_filename=TRADITIONAL_MANIFEST, ) + argv = [ + "fetch-artifact", + "--url", + str(traditional), + "--source-id", + "irs_soi", + "--package-id", + "soi-ira-traditional-contributions-2022", + "--year", + "2022", + "--out-dir", + str(package), + ] - assert report.valid - assert report.manifest_path.endswith("manifest.yaml") - assert (package / TRADITIONAL_MANIFEST).read_bytes() == recorded + assert harness_main(argv) == 1 + + err = capsys.readouterr().err + assert err.startswith("error: ") + assert TRADITIONAL_MANIFEST in err + assert not (package / "manifest.yaml").exists() + + +def test_a_same_bytes_rename_is_refused_by_name_not_as_a_revision(tmp_path): + """Identical bytes under another filename are neither a revision nor a + re-fetch: the entry's filename must keep agreeing with its recorded key.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "22in05ira.xlsx", b"IRA table 5") + _fetch_local(package, source, filename="table-5.xlsx", upload_r2=False) + recorded = (package / "manifest.yaml").read_bytes() + + for record_revision in (False, True): + with pytest.raises(SourceArtifactRevisionError) as raised: + _fetch_local( + package, source, upload_r2=False, record_revision=record_revision + ) + message = str(raised.value) + assert "rename is not a release revision" in message + assert "filename=table-5.xlsx" in message + assert "names them 22in05ira.xlsx" in message + assert "--filename table-5.xlsx" in message + + assert (package / "manifest.yaml").read_bytes() == recorded + assert not (package / "22in05ira.xlsx").exists() + assert (package / "table-5.xlsx").read_bytes() == b"IRA table 5" @pytest.mark.parametrize( @@ -1314,6 +1388,36 @@ def test_a_manifest_name_must_stay_inside_the_package(tmp_path, manifest_filenam assert not package.exists() +def test_fetch_artifact_cli_reports_a_manifest_name_outside_the_package( + tmp_path, capsys +): + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + source = _publish(tmp_path, "table.xlsx", b"table") + argv = [ + "fetch-artifact", + "--url", + str(source), + "--source-id", + "irs_soi", + "--package-id", + "soi-table-5", + "--year", + "2022", + "--out-dir", + str(package), + "--manifest", + "../manifest.yaml", + ] + + assert harness_main(argv) == 1 + + err = capsys.readouterr().err + assert err.startswith("error: ") + assert "inside the package directory" in err + assert not package.exists() + assert not (tmp_path / "db" / "data" / "irs_soi" / "manifest.yaml").exists() + + def test_fetch_artifact_cli_targets_the_named_manifest(tmp_path, capsys): package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" traditional = _publish(tmp_path, "22in05ira.xlsx", b"traditional IRA table") @@ -1573,6 +1677,31 @@ def test_a_malformed_manifest_is_refused_before_anything_is_fetched(tmp_path, do assert list(package.iterdir()) == [manifest_path] +@pytest.mark.parametrize( + "document", + [ + "files:\n- not a mapping\n", + "files: 3\n", + "source_id: irs_soi\nfiles: text\n", + ], +) +def test_a_non_mapping_files_block_is_refused_before_anything_is_fetched( + tmp_path, document +): + """The same document inventory-artifacts and publish-raw report as + 'files must be a mapping'; a fetch must not overwrite the artifact first.""" + package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" + package.mkdir(parents=True) + manifest_path = package / "manifest.yaml" + manifest_path.write_text(document) + + with pytest.raises(MalformedManifestError, match="files must be a mapping"): + _fetch_local(package, tmp_path / "publisher" / "never-read.xlsx") + + assert manifest_path.read_text() == document + assert list(package.iterdir()) == [manifest_path] + + @pytest.mark.parametrize("document", ["", "\n", "{}\n", "# only a comment\n"]) def test_an_empty_manifest_still_reads_as_absent(tmp_path, document): package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" From fb1bc1dfc33ad60f2467764df645d02f9a28cb61 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 01:53:56 -0400 Subject: [PATCH 25/42] Record into a fresh mapping when a manifest's files block is an explicit null A bare `files:` line parses as None. `_manifest_files` reads it as no entries (like an absent block) and `_upsert_manifest` now replaces the null with a mapping before recording, instead of `setdefault` keeping the null and the item assignment failing after the artifact was written and any upload ran. Covered by two new `files:` cases in test_an_empty_manifest_still_reads_as_absent. --- chronicle/artifacts.py | 7 ++++++- tests/test_chronicle_artifacts.py | 14 +++++++++++++- 2 files changed, 19 insertions(+), 2 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 7c3032f0..e67f151d 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -107,6 +107,8 @@ def _manifest_files(payload: dict[str, Any], manifest_path: Path) -> dict[str, A """ files = payload.get("files") if files is None: + # A bare ``files:`` line parses as None: no entries, like an absent + # block. The writer normalizes it to a mapping before recording into it. return {} if not isinstance(files, dict): raise MalformedManifestError( @@ -1684,7 +1686,10 @@ def _upsert_manifest( payload.setdefault("dataset", dataset) payload.setdefault("source_page", source_page) payload.setdefault("table", table) - payload.setdefault("files", {}) + if payload.get("files") is None: + # setdefault keeps an explicit null (a bare ``files:`` line); the + # entry below needs a mapping to record into. + payload["files"] = {} file_entry: dict[str, Any] = { "filename": filename, "source_url": source_url, diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index 224e73bd..f41f3938 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -1702,8 +1702,20 @@ def test_a_non_mapping_files_block_is_refused_before_anything_is_fetched( assert list(package.iterdir()) == [manifest_path] -@pytest.mark.parametrize("document", ["", "\n", "{}\n", "# only a comment\n"]) +@pytest.mark.parametrize( + "document", + [ + "", + "\n", + "{}\n", + "# only a comment\n", + "files:\n", + "source_id: irs_soi\nfiles:\n", + ], +) def test_an_empty_manifest_still_reads_as_absent(tmp_path, document): + """Including a bare ``files:`` line, which parses as an explicit null: the + fetch records into a fresh mapping rather than failing after the write.""" package = tmp_path / "db" / "data" / "irs_soi" / "soi-table-5" package.mkdir(parents=True) (package / "manifest.yaml").write_text(document) From 31f7aa1dcc9aeca217769643d8d1d64cb6948cb7 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:18:15 -0400 Subject: [PATCH 26/42] Add licence/access fields and hash-only artifact registration Introduce chronicle/registration.py as the leaf module owning the access class closed set (public/licensed/restricted), the manifest kind closed set (publisher_table/microdata_release), file-entry validation, and register_hash_only_artifact. Wire it through artifacts.py (fetch refuses hash-only access, inventory accepts entries with no local file, publish refuses to upload them), source_package.py (microdata releases are never parsed), and the harness/CLI (new `chronicle register-artifact`, plus --access/--licence on fetch-artifact and --skip-hash-only on publish-raw). Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 306 ++++++++++++++++-- chronicle/cli.py | 1 + chronicle/harness.py | 241 ++++++++++++++ chronicle/registration.py | 616 ++++++++++++++++++++++++++++++++++++ chronicle/source_package.py | 61 +++- 5 files changed, 1189 insertions(+), 36 deletions(-) create mode 100644 chronicle/registration.py diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index e67f151d..1015c367 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -26,6 +26,19 @@ ) from chronicle.env import env_value from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain +from chronicle.registration import ( + ACCESS_PUBLIC, + MICRODATA_RELEASE_KIND, + ListSpecRejected, + ManifestAccessError, + is_hash_only, + iter_file_specs, + normalize_access, + recorded_r2, + safe_entry_access, + safe_manifest_kind, + validate_file_entry, +) R2_RAW_BUCKET_ENV = "CHRONICLE_R2_RAW_BUCKET" @@ -42,6 +55,10 @@ DEFAULT_R2_PREFIX = "raw" DEFAULT_R2_DERIVED_PREFIX = "derived" +#: ``RawArtifactPublishEntry.skipped`` prefix for a licensed or restricted +#: registration: nothing to upload, because no Chronicle store holds its bytes. +HASH_ONLY_SKIP_PREFIX = "hash_only_access:" + # Most packages keep one manifest.yaml. Publisher directories that feed several # source packages keep one manifest each -- db/data/irs_soi/ira_contributions # holds manifest_traditional_source_package.yaml beside the Roth one -- so the @@ -370,10 +387,13 @@ class ArtifactInventoryEntry: source_url: str | None r2: dict[str, Any] | None errors: tuple[str, ...] + access: str = ACCESS_PUBLIC + licence: str | None = None + hash_only: bool = False @property def valid(self) -> bool: - """Whether this artifact is locally available and checksum-valid.""" + """Whether this registration is complete and, if public, available.""" return not self.errors def to_dict(self) -> dict[str, Any]: @@ -385,6 +405,9 @@ def to_dict(self) -> dict[str, Any]: "filename": self.filename, "local_path": self.local_path, "exists": self.exists, + "access": self.access, + "licence": self.licence, + "hash_only": self.hash_only, "sha256_expected": self.sha256_expected, "sha256_actual": self.sha256_actual, "size_bytes": self.size_bytes, @@ -441,10 +464,18 @@ def uploaded(self) -> bool: """Whether this run uploaded the artifact.""" return self.upload is not None and self.upload.ok + @property + def hash_only_refused(self) -> bool: + """Whether this entry was left alone because its access is hash-only.""" + return self.skipped is not None and self.skipped.startswith( + HASH_ONLY_SKIP_PREFIX + ) + @property def valid(self) -> bool: - """Whether this raw artifact is published: uploaded now, or already - held by the recorded object in a preserved bucket (``skipped``).""" + """Whether this raw artifact is published or correctly not uploaded: + uploaded now, already held by the recorded object in a preserved + bucket, or skipped as a hash-only registration (``skipped``).""" return not self.errors and (self.skipped is not None or self.uploaded) def to_dict(self) -> dict[str, Any]: @@ -462,6 +493,7 @@ def to_dict(self) -> dict[str, Any]: "size_bytes": self.size_bytes, "r2_location": (self.r2_location.to_dict() if self.r2_location else None), "upload": self.upload.to_dict() if self.upload else None, + "skipped": self.skipped, "errors": list(self.errors), } @@ -491,6 +523,9 @@ def counts(self) -> dict[str, int]: 1 for entry in self.entries if entry.skipped is not None ), "failed_count": sum(1 for entry in self.entries if not entry.valid), + "hash_only_refused_count": sum( + 1 for entry in self.entries if entry.hash_only_refused + ), "r2_link_count": sum( 1 for entry in self.entries if entry.r2_location is not None ), @@ -616,6 +651,8 @@ def fetch_source_artifact( table: str | None = None, filename: str | None = None, manifest_filename: str = DEFAULT_MANIFEST_FILENAME, + access: str = ACCESS_PUBLIC, + licence: str | None = None, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -635,10 +672,22 @@ def fetch_source_artifact( the superseded object moves to ``storage.previous_r2``. Without it, bytes that disagree with the entry's recorded identity raise :class:`SourceArtifactRevisionError` before anything is overwritten. + + Only ``public`` artifacts travel this path: it writes bytes into the package + tree and can upload them to the raw bucket. Licensed and restricted + artifacts are registered hash-only with + :func:`chronicle.registration.register_hash_only_artifact`. """ r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) manifest_path = _manifest_path(output, manifest_filename) + access_class = normalize_access(access) + if is_hash_only(access_class): + raise ManifestAccessError( + f"fetch-artifact stores bytes and refuses access={access_class!r}. " + "Register a licensed or restricted artifact by identity with " + "`chronicle register-artifact`." + ) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, default_prefix=DEFAULT_R2_PREFIX, @@ -656,12 +705,28 @@ def fetch_source_artifact( manifest_path=manifest_path, year=year, ) + licence_text = licence.strip() if isinstance(licence, str) else None + if ( + safe_manifest_kind(existing_manifest)[0] == MICRODATA_RELEASE_KIND + and not licence_text + ): + raise ManifestAccessError( + f"{manifest_path} registers a microdata release, so every entry " + "must record its publisher licence; pass --licence." + ) + _assert_no_hash_only_entry(existing_manifest, manifest_path, year, filename) fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() content, inferred_filename = _read_artifact(source_url) artifact_filename = filename or inferred_filename if not artifact_filename: raise ValueError("Could not infer artifact filename; pass --filename.") + _assert_no_hash_only_entry( + existing_manifest, + manifest_path, + year, + artifact_filename, + ) sha256 = hashlib.sha256(content).hexdigest() size_bytes = len(content) @@ -720,6 +785,8 @@ def fetch_source_artifact( sha256=sha256, size_bytes=size_bytes, fetched_at=fetched_at, + access=access_class, + licence=licence_text, r2_location=(r2_location if upload_r2 and r2_upload and r2_upload.ok else None), record_revision=record_revision, ) @@ -887,8 +954,15 @@ def publish_source_artifacts( r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", + skip_hash_only: bool = False, ) -> RawArtifactPublishReport: - """Upload manifest-declared raw source artifacts and record R2 locations.""" + """Upload manifest-declared raw source artifacts and record R2 locations. + + Only ``public`` artifacts are uploaded. A licensed or restricted entry is + refused: no bytes are read or sent, and the entry carries a + ``hash_only_access_refuses_bytes`` error unless ``skip_hash_only`` marks the + scan as deliberately mixed. + """ r2_bucket = r2_bucket or default_r2_raw_bucket() root_path = Path(root) if not root_path.exists(): @@ -931,22 +1005,29 @@ def publish_source_artifacts( errors.append(f"Could not resolve R2 prefix for {manifest_path}: {exc}") continue + kind, kind_error = safe_manifest_kind(manifest) + if kind_error: + errors.append(f"{kind_error}: {manifest_path}") updated = False for year, spec in files.items(): - entry, updated_spec = _publish_raw_manifest_entry( - manifest_path, - manifest_source_id, - manifest_package_id, - year, - spec, - r2_bucket=r2_bucket, - r2_prefix=resolved_r2_prefix, - wrangler_command=wrangler_command, - ) - entries.append(entry) - if updated_spec is not None and isinstance(spec, dict): - spec.update(updated_spec) - updated = True + for file_spec in iter_file_specs(spec, kind=kind): + entry, updated_spec = _publish_raw_manifest_entry( + manifest_path, + manifest_source_id, + manifest_package_id, + year, + file_spec, + manifest=manifest, + kind=kind, + r2_bucket=r2_bucket, + r2_prefix=resolved_r2_prefix, + wrangler_command=wrangler_command, + skip_hash_only=skip_hash_only, + ) + entries.append(entry) + if updated_spec is not None and isinstance(file_spec, dict): + file_spec.update(updated_spec) + updated = True if updated: manifest.setdefault("source_id", manifest_source_id) manifest.setdefault("package_id", manifest_package_id) @@ -1020,6 +1101,7 @@ def inventory_source_artifacts( "artifact_count": 0, "missing_count": 0, "checksum_mismatch_count": 0, + "hash_only_count": 0, "r2_link_count": 0, }, entries=(), @@ -1029,23 +1111,37 @@ def inventory_source_artifacts( manifests = sorted(root_path.rglob(manifest_filename)) for manifest_path in manifests: try: - files = _read_manifest(manifest_path).get("files") or {} + manifest = _read_manifest(manifest_path) except (OSError, MalformedManifestError) as exc: errors.append(f"Could not read {manifest_path}: {exc}") continue + files = manifest.get("files") or {} if not isinstance(files, dict): errors.append(f"Manifest files must be a mapping: {manifest_path}") continue + kind, kind_error = safe_manifest_kind(manifest) + if kind_error: + errors.append(f"{kind_error}: {manifest_path}") for year, spec in files.items(): - entries.append(_inventory_entry(manifest_path, year, spec)) + for file_spec in iter_file_specs(spec, kind=kind): + entries.append( + _inventory_entry( + manifest_path, + year, + file_spec, + manifest=manifest, + kind=kind, + ) + ) counts = { "manifest_count": len(manifests), "artifact_count": len(entries), - "missing_count": sum(1 for entry in entries if not entry.exists), + "missing_count": sum(1 for entry in entries if "missing_file" in entry.errors), "checksum_mismatch_count": sum( 1 for entry in entries if "checksum_mismatch" in entry.errors ), + "hash_only_count": sum(1 for entry in entries if entry.hash_only), "r2_link_count": sum(1 for entry in entries if entry.r2 is not None), } return ArtifactInventoryReport( @@ -1677,6 +1773,8 @@ def _upsert_manifest( sha256: str, size_bytes: int, fetched_at: str, + access: str, + licence: str | None, r2_location: ArtifactStorageLocation | None, record_revision: bool = False, ) -> None: @@ -1690,13 +1788,25 @@ def _upsert_manifest( # setdefault keeps an explicit null (a bare ``files:`` line); the # entry below needs a mapping to record into. payload["files"] = {} + # Access is written explicitly on every entry this command touches, so a + # manifest never relies on the inferred ``public`` default once rewritten. file_entry: dict[str, Any] = { "filename": filename, "source_url": source_url, - "sha256": sha256, - "size_bytes": size_bytes, - "fetched_at": fetched_at, + "access": access, } + if licence: + file_entry["licence"] = licence + file_entry.update( + { + "sha256": sha256, + "size_bytes": size_bytes, + "fetched_at": fetched_at, + } + ) + existing = ( + payload["files"].get(year) if isinstance(payload["files"], dict) else None + ) recorded_spec = _manifest_file_spec(payload, year) recorded_storage = _recorded_storage(recorded_spec) identity = _recorded_identity(recorded_spec, manifest_path=manifest_path, year=year) @@ -1738,13 +1848,58 @@ def _upsert_manifest( # revision over a never-published entry supersedes nothing. if storage: file_entry["storage"] = storage - payload["files"][year] = file_entry + if isinstance(existing, list): + replaced = [ + entry + for entry in existing + if not (isinstance(entry, dict) and entry.get("filename") == filename) + ] + payload["files"][year] = [*replaced, file_entry] + else: + payload["files"][year] = file_entry manifest_path.write_text( - yaml.safe_dump(payload, sort_keys=False), + yaml.safe_dump(payload, sort_keys=False, allow_unicode=True), encoding="utf-8", ) +def _load_manifest_payload(manifest_path: Path) -> dict[str, Any]: + """Load a manifest mapping, or an empty mapping when absent or unreadable.""" + if not manifest_path.exists(): + return {} + payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} + if not isinstance(payload, dict): + raise ValueError(f"Manifest must be a mapping: {manifest_path}") + return payload + + +def _assert_no_hash_only_entry( + manifest: dict[str, Any], + manifest_path: Path, + year: Any, + filename: str | None, +) -> None: + """Refuse to fetch bytes over an existing hash-only registration.""" + if not filename: + return + files = manifest.get("files") + if not isinstance(files, dict): + return + kind, _kind_error = safe_manifest_kind(manifest) + for spec in iter_file_specs(files.get(year), kind=kind): + if not isinstance(spec, dict): + continue + if spec.get("filename") != filename: + continue + access = safe_entry_access(spec) + if is_hash_only(access): + raise ManifestAccessError( + f"{manifest_path} registers {filename!r} for {year} as " + f"access={access!r}. Its bytes must not enter a Chronicle " + "store; keep the hash-only registration." + ) + + def _upload_r2_object( location: ArtifactStorageLocation, local_path: Path, @@ -1774,15 +1929,70 @@ def _publish_raw_manifest_entry( year: Any, spec: Any, *, + manifest: dict[str, Any] | None = None, + kind: str | None = None, r2_bucket: str, r2_prefix: str, wrangler_command: str, + skip_hash_only: bool = False, ) -> tuple[RawArtifactPublishEntry, dict[str, Any] | None]: errors: list[str] = [] + if isinstance(spec, ListSpecRejected): + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename="", + local_path=str(manifest_path.parent), + sha256=None, + size_bytes=None, + r2_location=None, + upload=None, + errors=("list_file_spec_requires_microdata_release_kind",), + ), + None, + ) if not isinstance(spec, dict): spec = {} errors.append("malformed_file_spec") filename = str(spec.get("filename") or "") + access = safe_entry_access(spec) + if is_hash_only(access): + # Refuse before touching bytes: no Chronicle store holds a licensed or + # restricted artifact, so there is nothing here to upload. + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(manifest_path.parent / filename), + sha256=spec.get("sha256"), + size_bytes=spec.get("size_bytes"), + r2_location=None, + upload=None, + errors=( + () + if skip_hash_only + else (f"hash_only_access_refuses_bytes:{access}",) + ), + skipped=f"{HASH_ONLY_SKIP_PREFIX}{access}", + ), + None, + ) + errors.extend( + validate_file_entry( + spec, + kind=kind or safe_manifest_kind(manifest)[0], + manifest=manifest, + local_file_exists=(manifest_path.parent / filename).exists() + if filename + else False, + ) + ) artifact_path = manifest_path.parent / filename sha256_expected = spec.get("sha256") sha256_actual = None @@ -1942,19 +2152,51 @@ def _inventory_entry( manifest_path: Path, year: Any, spec: Any, + *, + manifest: dict[str, Any] | None = None, + kind: str | None = None, ) -> ArtifactInventoryEntry: errors: list[str] = [] + original_spec = spec + if isinstance(spec, ListSpecRejected): + return ArtifactInventoryEntry( + manifest_path=str(manifest_path), + year=str(year), + filename="", + local_path=str(manifest_path.parent), + exists=False, + sha256_expected=None, + sha256_actual=None, + size_bytes=None, + source_url=None, + r2=None, + errors=("list_file_spec_requires_microdata_release_kind",), + ) if not isinstance(spec, dict): spec = {} errors.append("malformed_file_spec") filename = str(spec.get("filename") or "") artifact_path = manifest_path.parent / filename exists = bool(filename) and artifact_path.exists() + access = safe_entry_access(spec) + hash_only = is_hash_only(access) + errors.extend( + validate_file_entry( + original_spec, + kind=kind or safe_manifest_kind(manifest)[0], + manifest=manifest, + local_file_exists=exists, + ) + ) sha256_expected = spec.get("sha256") sha256_actual = None - size_bytes = None + size_bytes = spec.get("size_bytes") if hash_only else None if not filename: errors.append("missing_filename") + elif hash_only: + # A licensed or restricted registration is identity only: Chronicle + # never holds the bytes, so a missing local file is the correct state. + pass elif not exists: errors.append("missing_file") else: @@ -1963,8 +2205,7 @@ def _inventory_entry( size_bytes = len(content) if sha256_expected and sha256_actual != sha256_expected: errors.append("checksum_mismatch") - storage = spec.get("storage") if isinstance(spec, dict) else None - r2 = storage.get("r2") if isinstance(storage, dict) else None + r2 = recorded_r2(spec) return ArtifactInventoryEntry( manifest_path=str(manifest_path), year=str(year), @@ -1975,8 +2216,11 @@ def _inventory_entry( sha256_actual=sha256_actual, size_bytes=size_bytes, source_url=spec.get("source_url"), - r2=r2, - errors=tuple(errors), + r2=dict(r2) if r2 is not None else None, + errors=tuple(dict.fromkeys(errors)), + access=access, + licence=spec.get("licence"), + hash_only=hash_only, ) diff --git a/chronicle/cli.py b/chronicle/cli.py index b9098d7f..bf032e5f 100644 --- a/chronicle/cli.py +++ b/chronicle/cli.py @@ -26,6 +26,7 @@ def main() -> None: ["plan-pe-sources"], ["publish-derived"], ["publish-raw"], + ["register-artifact"], ["scaffold-package"], ["validate-concept-alignments"], ["validate-package"], diff --git a/chronicle/harness.py b/chronicle/harness.py index ee7f6ef9..acbbc4dc 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -51,6 +51,11 @@ write_pe_source_plan_json, write_pe_source_plan_markdown, ) +from chronicle.registration import ( + ACCESS_CLASSES, + ArtifactRegistrationReport, + register_hash_only_artifact, +) from chronicle.sources.cells import ( SourceCell, SourceCellReport, @@ -341,6 +346,8 @@ def fetch_artifact_file( table: str | None = None, filename: str | None = None, manifest_filename: str = DEFAULT_MANIFEST_FILENAME, + access: str = "public", + licence: str | None = None, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -365,6 +372,8 @@ def fetch_artifact_file( table=table, filename=filename, manifest_filename=manifest_filename, + access=access, + licence=licence, upload_r2=upload_r2, record_revision=record_revision, r2_bucket=r2_bucket, @@ -373,6 +382,60 @@ def fetch_artifact_file( ) +def register_artifact_file( + *, + source_id: str, + package_id: str, + year: int, + output_dir: str | Path, + filename: str, + sha256: str, + licence: str, + access: str, + vintage: str, + size_bytes: int | None = None, + source_page: str | None = None, + source_url: str | None = None, + access_route: str | None = None, + doi: str | None = None, + study: str | None = None, + dataset: str | None = None, + table: str | None = None, + publisher: str | None = None, + fetched_at: str | None = None, + verified_at: str | None = None, + hash_source: str | None = None, + notes: str | None = None, + allow_reissue: bool = False, +) -> ArtifactRegistrationReport: + """Register a licensed or restricted artifact by identity, without bytes.""" + return register_hash_only_artifact( + source_id=source_id, + package_id=package_id, + year=year, + output_dir=output_dir, + filename=filename, + sha256=sha256, + licence=licence, + access=access, + vintage=vintage, + size_bytes=size_bytes, + source_page=source_page, + source_url=source_url, + access_route=access_route, + doi=doi, + study=study, + dataset=dataset, + table=table, + publisher=publisher, + fetched_at=fetched_at, + verified_at=verified_at, + hash_source=hash_source, + notes=notes, + allow_reissue=allow_reissue, + ) + + def inventory_artifact_files( root: str | Path, *, @@ -391,6 +454,7 @@ def publish_raw_artifact_files( r2_bucket: str | None = None, r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", + skip_hash_only: bool = False, ) -> RawArtifactPublishReport: """Publish manifest-declared raw source artifacts to R2.""" return publish_source_artifacts( @@ -401,6 +465,7 @@ def publish_raw_artifact_files( r2_bucket=r2_bucket, r2_prefix=r2_prefix, wrangler_command=wrangler_command, + skip_hash_only=skip_hash_only, ) @@ -911,6 +976,23 @@ def main(argv: list[str] | None = None) -> int: "--filename", help="Override artifact filename inferred from URL/path.", ) + artifact_parser.add_argument( + "--access", + default="public", + choices=list(ACCESS_CLASSES), + help=( + "Publisher access class. Only public artifacts may be fetched; " + "licensed and restricted artifacts are registered hash-only with " + "`chronicle register-artifact`." + ), + ) + artifact_parser.add_argument( + "--licence", + help=( + "Publisher terms identifier or URL. Required for entries in a " + "microdata-release manifest." + ), + ) artifact_parser.add_argument( "--upload-r2", action="store_true", @@ -950,6 +1032,126 @@ def main(argv: list[str] | None = None) -> int: help="Wrangler command prefix to use for R2 uploads.", ) + registration_parser = subparsers.add_parser( + "register-artifact", + help="Register a licensed or restricted artifact by identity, no bytes", + description=( + "Register a raw artifact Chronicle may not redistribute. The " + "manifest records the checksum, size, vintage, licence, and access " + "route; no bytes are fetched, stored, or uploaded, and no R2 key " + "is recorded." + ), + ) + registration_parser.add_argument( + "--source-id", + required=True, + help="Stable source ID, such as dwp.", + ) + registration_parser.add_argument( + "--package-id", + required=True, + help="Stable package ID, such as dwp-frs-2023-24.", + ) + registration_parser.add_argument( + "--year", + type=int, + required=True, + help="Artifact vintage year to record in manifest.yaml", + ) + registration_parser.add_argument( + "--out-dir", + type=Path, + required=True, + help="Directory where manifest.yaml should live. No bytes are written.", + ) + registration_parser.add_argument( + "--filename", + required=True, + help="Publisher filename this registration identifies.", + ) + registration_parser.add_argument( + "--sha256", + required=True, + help="Lowercase 64-character SHA-256 of the publisher bytes.", + ) + registration_parser.add_argument( + "--size-bytes", + type=int, + help="Size of the publisher bytes, when known.", + ) + registration_parser.add_argument( + "--vintage", + required=True, + help="Publisher vintage label, such as 2023_24.", + ) + registration_parser.add_argument( + "--licence", + required=True, + help="Publisher terms identifier or URL, such as a UKDS licence.", + ) + registration_parser.add_argument( + "--access", + required=True, + choices=["licensed", "restricted"], + help="Access class. Public artifacts are registered with fetch-artifact.", + ) + registration_parser.add_argument( + "--source-page", + help="Publisher or archive landing page for the release.", + ) + registration_parser.add_argument( + "--source-url", + help="Direct publisher URL, when the release has one behind its licence.", + ) + registration_parser.add_argument( + "--access-route", + help="How an authorized consumer obtains the bytes.", + ) + registration_parser.add_argument( + "--doi", + help="Persistent identifier for the study, such as a UKDS DOI.", + ) + registration_parser.add_argument( + "--study", + help="Archive study reference, such as a UK Data Service study number.", + ) + registration_parser.add_argument( + "--dataset", + help="Manifest dataset ID. Defaults to _.", + ) + registration_parser.add_argument( + "--table", + help="Human-readable release title.", + ) + registration_parser.add_argument( + "--publisher", + help="Publishing body, when it differs from the source ID.", + ) + registration_parser.add_argument( + "--fetched-at", + help="When the authorized environment fetched the bytes, if known.", + ) + registration_parser.add_argument( + "--verified-at", + help="When this checksum was verified against the reviewed pin.", + ) + registration_parser.add_argument( + "--hash-source", + help="Where the checksum came from, such as a consumer source manifest.", + ) + registration_parser.add_argument( + "--notes", + help="Free-text provenance notes for the registration.", + ) + registration_parser.add_argument( + "--allow-reissue", + action="store_true", + help=( + "Register different bytes for a filename already registered in " + "this year, keeping both registrations." + ), + ) + artifact_inventory_parser = subparsers.add_parser( "inventory-artifacts", help="Inventory local manifest-declared source artifacts", @@ -981,6 +1183,14 @@ def main(argv: list[str] | None = None) -> int: default="manifest.yaml", help="Manifest filename to scan for.", ) + raw_publish_parser.add_argument( + "--skip-hash-only", + action="store_true", + help=( + "Treat licensed and restricted registrations as deliberately " + "skipped rather than refused, so a mixed tree can be published." + ), + ) raw_publish_parser.add_argument( "--source-id", help="Override manifest source_id for scanned artifacts.", @@ -1377,6 +1587,8 @@ def main(argv: list[str] | None = None) -> int: table=args.table, filename=args.filename, manifest_filename=args.manifest, + access=args.access, + licence=args.licence, upload_r2=args.upload_r2, record_revision=args.record_revision, r2_bucket=args.r2_bucket, @@ -1388,6 +1600,34 @@ def main(argv: list[str] | None = None) -> int: return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 + if args.command == "register-artifact": + registration = register_artifact_file( + source_id=args.source_id, + package_id=args.package_id, + year=args.year, + output_dir=args.out_dir, + filename=args.filename, + sha256=args.sha256, + licence=args.licence, + access=args.access, + vintage=args.vintage, + size_bytes=args.size_bytes, + source_page=args.source_page, + source_url=args.source_url, + access_route=args.access_route, + doi=args.doi, + study=args.study, + dataset=args.dataset, + table=args.table, + publisher=args.publisher, + fetched_at=args.fetched_at, + verified_at=args.verified_at, + hash_source=args.hash_source, + notes=args.notes, + allow_reissue=args.allow_reissue, + ) + print(json.dumps(registration.to_dict(), indent=2, sort_keys=True)) + return 0 if registration.valid else 1 if args.command == "inventory-artifacts": report = inventory_artifact_files( args.root, @@ -1404,6 +1644,7 @@ def main(argv: list[str] | None = None) -> int: r2_bucket=args.r2_bucket, r2_prefix=args.r2_prefix, wrangler_command=args.wrangler_command, + skip_hash_only=args.skip_hash_only, ) print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 diff --git a/chronicle/registration.py b/chronicle/registration.py new file mode 100644 index 00000000..6044af00 --- /dev/null +++ b/chronicle/registration.py @@ -0,0 +1,616 @@ +"""Access classification and hash-only registration for Chronicle artifacts. + +Chronicle registers every raw artifact its consumers build from and stores the +bytes of only those the publisher permits it to redistribute. Two manifest file +fields carry that split: + +``licence`` + The publisher's terms, as an identifier or a URL. + +``access`` + A closed class: ``public``, ``licensed``, or ``restricted``. + +``public`` artifacts keep the existing fetch/publish path: bytes are archived in +the raw R2 bucket under the content-addressed key +``raw/{source_id}/{package_id}/{year}/{sha256}/{filename}``. ``licensed`` and +``restricted`` artifacts are registered *hash-only*: the manifest records the +checksum, vintage, licence, and access route, and no Chronicle store ever holds +the bytes. That key exists only for ``public`` artifacts. + +A registration is identified by ``{source_id, package_id, year, sha256, +filename}``. Consumers reference a registration by exactly that tuple. + +See ``docs/adr-chronicle-raw-microdata-identity.md``. +""" + +from __future__ import annotations + +from collections.abc import Iterable, Mapping +from dataclasses import dataclass +from pathlib import Path +import re +from typing import Any + +import yaml + + +ACCESS_PUBLIC = "public" +ACCESS_LICENSED = "licensed" +ACCESS_RESTRICTED = "restricted" +#: The closed set of access classes a manifest file entry may declare. +ACCESS_CLASSES: tuple[str, ...] = (ACCESS_PUBLIC, ACCESS_LICENSED, ACCESS_RESTRICTED) +#: Access class inferred for an entry that does not declare one. +DEFAULT_ACCESS = ACCESS_PUBLIC + +PUBLISHER_TABLE_KIND = "publisher_table" +MICRODATA_RELEASE_KIND = "microdata_release" +#: The closed set of manifest kinds. Manifests without ``kind`` are tables. +MANIFEST_KINDS: tuple[str, ...] = (PUBLISHER_TABLE_KIND, MICRODATA_RELEASE_KIND) +DEFAULT_MANIFEST_KIND = PUBLISHER_TABLE_KIND + +_SHA256_RE = re.compile(r"^[0-9a-f]{64}$") + +# Registration entry key order, so emitted manifests are byte-stable. +_REGISTRATION_FIELD_ORDER: tuple[str, ...] = ( + "filename", + "access", + "licence", + "vintage", + "sha256", + "size_bytes", + "source_url", + "access_route", + "doi", + "study", + "fetched_at", + "verified_at", + "hash_source", + "notes", +) + + +class ManifestAccessError(ValueError): + """Raised when a manifest declares an unusable access class or kind.""" + + +class HashOnlyRegistrationError(ValueError): + """Raised when a hash-only registration is malformed or would store bytes.""" + + +class MicrodataReleaseNotParseableError(ValueError): + """Raised when a source package points at a registered microdata release. + + Registration is manifest-level: no source package parses a microdata + release, and no microdata row, cell, or fact enters Chronicle. + """ + + +@dataclass(frozen=True) +class ListSpecRejected: + """Marker for a list ``files[year]`` value outside a microdata release.""" + + spec: Any + + +def manifest_kind(manifest: Mapping[str, Any] | None) -> str: + """Return a manifest's declared kind, defaulting to ``publisher_table``.""" + if not isinstance(manifest, Mapping): + return DEFAULT_MANIFEST_KIND + declared = manifest.get("kind") + if declared is None: + return DEFAULT_MANIFEST_KIND + kind = str(declared) + if kind not in MANIFEST_KINDS: + raise ManifestAccessError( + f"Unknown manifest kind {kind!r}; expected one of {list(MANIFEST_KINDS)}." + ) + return kind + + +def safe_manifest_kind(manifest: Mapping[str, Any] | None) -> tuple[str, str | None]: + """Return ``(kind, error_code)`` without raising on an unknown kind.""" + try: + return manifest_kind(manifest), None + except ManifestAccessError: + declared = manifest.get("kind") if isinstance(manifest, Mapping) else None + return DEFAULT_MANIFEST_KIND, f"unknown_manifest_kind:{declared}" + + +def is_microdata_release(manifest: Mapping[str, Any] | None) -> bool: + """Whether a manifest registers a microdata release rather than a table.""" + return manifest_kind(manifest) == MICRODATA_RELEASE_KIND + + +def normalize_access(access: str | None) -> str: + """Return a validated access class, defaulting to ``public``.""" + if access is None: + return DEFAULT_ACCESS + value = str(access) + if value not in ACCESS_CLASSES: + raise ManifestAccessError( + f"Unknown access class {value!r}; expected one of {list(ACCESS_CLASSES)}." + ) + return value + + +def entry_access(spec: Any) -> str: + """Return the access class a manifest file entry declares or inherits.""" + if not isinstance(spec, Mapping): + return DEFAULT_ACCESS + return normalize_access(spec.get("access")) + + +def safe_entry_access(spec: Any) -> str: + """Return an entry's access class, falling back to ``public`` if unknown. + + An unparseable class is reported by :func:`validate_file_entry`; treating it + as ``public`` here would be unsafe, so it is treated as ``restricted`` and + therefore never uploaded. + """ + if not isinstance(spec, Mapping): + return DEFAULT_ACCESS + try: + return normalize_access(spec.get("access")) + except ManifestAccessError: + return ACCESS_RESTRICTED + + +def stores_bytes(access: str) -> bool: + """Whether Chronicle may hold this access class's bytes.""" + return normalize_access(access) == ACCESS_PUBLIC + + +def is_hash_only(access: str) -> bool: + """Whether this access class must be registered without bytes.""" + return not stores_bytes(access) + + +def registration_id( + *, + source_id: str, + package_id: str, + year: Any, + sha256: str, + filename: str, +) -> str: + """Return the registration identity tuple as a stable string.""" + return f"{source_id}/{package_id}/{year}/{sha256}/{filename}" + + +def iter_file_specs(spec: Any, *, kind: str) -> tuple[Any, ...]: + """Expand one ``files[year]`` value into individual file entries. + + A microdata release registers many files under one vintage — the 14 FRS + 2023-24 tabs share ``{source_id, package_id, year}`` and differ only by + ``filename`` and ``sha256`` — so its ``files[year]`` value may be a list. + Publisher-table manifests keep the single-mapping shape, and a list there is + surfaced as a rejected entry rather than silently expanded. + """ + if isinstance(spec, list): + if kind != MICRODATA_RELEASE_KIND: + return (ListSpecRejected(spec),) + return tuple(spec) + return (spec,) + + +def validate_file_entry( + spec: Any, + *, + kind: str, + manifest: Mapping[str, Any] | None, + local_file_exists: bool, +) -> tuple[str, ...]: + """Return stable error codes for one manifest file entry. + + The codes are the refusal vocabulary shared by ``inventory-artifacts``, + ``publish-raw``, and ``register-artifact``. + """ + if isinstance(spec, ListSpecRejected): + return ("list_file_spec_requires_microdata_release_kind",) + if not isinstance(spec, Mapping): + return () + + errors: list[str] = [] + declared_access = spec.get("access") + if declared_access is None: + if kind == MICRODATA_RELEASE_KIND: + errors.append("missing_access") + access = DEFAULT_ACCESS + else: + try: + access = normalize_access(declared_access) + except ManifestAccessError: + return (*errors, f"unknown_access_class:{declared_access}") + + if kind == MICRODATA_RELEASE_KIND and not _text(spec.get("licence")): + errors.append("missing_licence") + + if is_hash_only(access): + errors.extend( + _hash_only_entry_errors( + spec, + manifest=manifest, + local_file_exists=local_file_exists, + ) + ) + return tuple(_dedupe(errors)) + + +def _hash_only_entry_errors( + spec: Mapping[str, Any], + *, + manifest: Mapping[str, Any] | None, + local_file_exists: bool, +) -> list[str]: + """Return refusal codes for a licensed or restricted registration.""" + errors: list[str] = [] + if not _text(spec.get("licence")): + errors.append("missing_licence") + sha256 = _text(spec.get("sha256")) + if not sha256: + errors.append("missing_sha256") + elif not _SHA256_RE.match(sha256): + errors.append("malformed_sha256") + if not _text(spec.get("vintage")): + errors.append("missing_vintage") + if not _access_route(spec, manifest): + errors.append("missing_access_route") + if not (_text(spec.get("verified_at")) or _text(spec.get("fetched_at"))): + errors.append("missing_verification_timestamp") + if local_file_exists: + errors.append("bytes_present_for_hash_only_entry") + if recorded_r2(spec): + errors.append("r2_location_for_hash_only_entry") + return errors + + +def _access_route( + spec: Mapping[str, Any], + manifest: Mapping[str, Any] | None, +) -> str | None: + """Return the recorded route to the bytes, from the entry or the manifest.""" + for key in ("access_route", "source_url", "source_page", "doi"): + value = _text(spec.get(key)) + if value: + return value + if isinstance(manifest, Mapping): + for key in ("source_page", "access_route"): + value = _text(manifest.get(key)) + if value: + return value + return None + + +def recorded_r2(spec: Any) -> Mapping[str, Any] | None: + """Return a recorded ``storage.r2`` mapping, if the entry carries one.""" + if not isinstance(spec, Mapping): + return None + storage = spec.get("storage") + if not isinstance(storage, Mapping): + return None + r2 = storage.get("r2") + return r2 if isinstance(r2, Mapping) else None + + +@dataclass(frozen=True) +class ArtifactRegistrationReport: + """Report from registering one hash-only source artifact.""" + + manifest_path: str + source_id: str + package_id: str + year: int + filename: str + sha256: str + size_bytes: int | None + vintage: str + licence: str + access: str + registration: str + replaced: bool + errors: tuple[str, ...] = () + + @property + def valid(self) -> bool: + """Whether the registration was written without refusals.""" + return not self.errors + + def to_dict(self) -> dict[str, Any]: + """Return a JSON-serializable report.""" + return { + "valid": self.valid, + "manifest_path": self.manifest_path, + "source_id": self.source_id, + "package_id": self.package_id, + "year": self.year, + "filename": self.filename, + "sha256": self.sha256, + "size_bytes": self.size_bytes, + "vintage": self.vintage, + "licence": self.licence, + "access": self.access, + "registration": self.registration, + "replaced": self.replaced, + "r2_location": None, + "errors": list(self.errors), + } + + +def register_hash_only_artifact( + *, + source_id: str, + package_id: str, + year: int, + output_dir: str | Path, + filename: str, + sha256: str, + licence: str, + access: str, + vintage: str, + size_bytes: int | None = None, + source_page: str | None = None, + source_url: str | None = None, + access_route: str | None = None, + doi: str | None = None, + study: str | None = None, + dataset: str | None = None, + table: str | None = None, + publisher: str | None = None, + fetched_at: str | None = None, + verified_at: str | None = None, + hash_source: str | None = None, + notes: str | None = None, + allow_reissue: bool = False, +) -> ArtifactRegistrationReport: + """Register a licensed or restricted artifact by identity, without bytes. + + Writes (or updates) a ``kind: microdata_release`` manifest entry carrying the + checksum, size, vintage, licence, access route, and verification timestamp. + No bytes are read, written, or uploaded, and no R2 key is recorded. + """ + access_class = normalize_access(access) + if stores_bytes(access_class): + raise HashOnlyRegistrationError( + "register-artifact records identity without bytes and refuses " + f"access={ACCESS_PUBLIC!r}. Register a public artifact with its " + "bytes using fetch-artifact." + ) + checksum = _text(sha256) + if not checksum or not _SHA256_RE.match(checksum): + raise HashOnlyRegistrationError( + "A registration needs a lowercase 64-character SHA-256; refusing to " + f"register {filename!r} with sha256={sha256!r}. Never invent a hash." + ) + if not _text(licence): + raise HashOnlyRegistrationError( + f"A {access_class} registration must record the publisher licence." + ) + if not _text(vintage): + raise HashOnlyRegistrationError( + f"A {access_class} registration must record the artifact vintage." + ) + artifact_name = _text(filename) + if not artifact_name or Path(artifact_name).name != artifact_name: + raise HashOnlyRegistrationError( + f"Registration filename must be a bare filename; got {filename!r}." + ) + if not (_text(verified_at) or _text(fetched_at)): + raise HashOnlyRegistrationError( + "A hash-only registration must record when the checksum was " + "verified; pass --verified-at." + ) + + output = Path(output_dir) + local_path = output / artifact_name + if local_path.exists(): + raise HashOnlyRegistrationError( + f"Refusing to register {artifact_name!r} hash-only while its bytes " + f"are present at {local_path}. A {access_class} artifact's bytes " + "must not live in a Chronicle store." + ) + + manifest_path = output / "manifest.yaml" + payload = _load_manifest(manifest_path) + existing_kind = manifest_kind(payload) + if payload and existing_kind != MICRODATA_RELEASE_KIND: + raise HashOnlyRegistrationError( + f"{manifest_path} is a {existing_kind} manifest; hash-only " + "registrations belong in a kind: microdata_release manifest." + ) + + entry = _registration_entry( + filename=artifact_name, + access=access_class, + licence=str(licence), + vintage=str(vintage), + sha256=checksum, + size_bytes=size_bytes, + source_url=source_url, + access_route=access_route, + doi=doi, + study=study, + fetched_at=fetched_at, + verified_at=verified_at, + hash_source=hash_source, + notes=notes, + ) + route_context = dict(payload) + if source_page: + route_context["source_page"] = source_page + if not _access_route(entry, route_context): + raise HashOnlyRegistrationError( + "A hash-only registration must record how the bytes are reached; " + "pass --access-route, --source-url, --doi, or --source-page." + ) + + _assert_manifest_identity(payload, manifest_path, "source_id", source_id) + _assert_manifest_identity(payload, manifest_path, "package_id", package_id) + payload.setdefault("source_id", source_id) + payload.setdefault("package_id", package_id) + payload["kind"] = MICRODATA_RELEASE_KIND + payload.setdefault("dataset", dataset or f"{source_id}_{package_id}") + if publisher: + payload.setdefault("publisher", publisher) + if source_page: + payload.setdefault("source_page", source_page) + if table: + payload.setdefault("table", table) + payload.setdefault("files", {}) + + entries = _existing_entries(payload["files"], year) + replaced = False + for index, existing in enumerate(entries): + if not isinstance(existing, Mapping): + continue + if _text(existing.get("filename")) != artifact_name: + continue + if _text(existing.get("sha256")) == checksum: + entries[index] = entry + replaced = True + break + if not allow_reissue: + raise HashOnlyRegistrationError( + f"{manifest_path} already registers {artifact_name!r} for " + f"{year} with sha256={existing.get('sha256')!r}. Different " + "bytes are a new publisher release, not a pin replacement; " + "pass --allow-reissue to register both." + ) + else: + # No identical (filename, sha256) entry: this is a new registration, + # which for a reissue sits alongside the pin it supersedes. + entries.append(entry) + + payload["files"][year] = sorted(entries, key=_entry_sort_key) + output.mkdir(parents=True, exist_ok=True) + manifest_path.write_text( + yaml.safe_dump(payload, sort_keys=False, allow_unicode=True), + encoding="utf-8", + ) + + return ArtifactRegistrationReport( + manifest_path=str(manifest_path), + source_id=str(payload["source_id"]), + package_id=str(payload["package_id"]), + year=year, + filename=artifact_name, + sha256=checksum, + size_bytes=size_bytes, + vintage=str(vintage), + licence=str(licence), + access=access_class, + registration=registration_id( + source_id=str(payload["source_id"]), + package_id=str(payload["package_id"]), + year=year, + sha256=checksum, + filename=artifact_name, + ), + replaced=replaced, + ) + + +def _assert_manifest_identity( + payload: Mapping[str, Any], + manifest_path: Path, + key: str, + value: str, +) -> None: + """Refuse to register into a manifest that identifies a different source.""" + existing = _text(payload.get(key)) + if existing is not None and existing != value: + raise HashOnlyRegistrationError( + f"{manifest_path} declares {key}={existing!r}; refusing to register " + f"{key}={value!r} into it." + ) + + +def _entry_sort_key(entry: Any) -> tuple[str, str]: + """Return a deterministic sort key for a registration entry.""" + if not isinstance(entry, Mapping): + return ("", "") + return (_text(entry.get("filename")) or "", _text(entry.get("sha256")) or "") + + +def _registration_entry(**values: Any) -> dict[str, Any]: + """Build a deterministic, field-ordered registration entry.""" + entry: dict[str, Any] = {} + for key in _REGISTRATION_FIELD_ORDER: + value = values.get(key) + if value is None: + continue + if isinstance(value, str) and not value.strip(): + continue + entry[key] = value + return entry + + +def _existing_entries(files: Any, year: Any) -> list[Any]: + """Return the existing file entries for a year as a mutable list.""" + if not isinstance(files, dict): + return [] + spec = files.get(year) + if spec is None: + return [] + if isinstance(spec, list): + return list(spec) + return [spec] + + +def _load_manifest(manifest_path: Path) -> dict[str, Any]: + """Load a manifest mapping, or an empty mapping when absent.""" + if not manifest_path.exists(): + return {} + payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} + if not isinstance(payload, dict): + raise HashOnlyRegistrationError(f"Manifest must be a mapping: {manifest_path}") + return payload + + +def _text(value: Any) -> str | None: + """Return a non-empty stripped string, or None.""" + if value is None: + return None + text = str(value).strip() + return text or None + + +def _dedupe(values: Iterable[str]) -> list[str]: + """Return values with duplicates removed, preserving order.""" + seen: set[str] = set() + ordered: list[str] = [] + for value in values: + if value not in seen: + seen.add(value) + ordered.append(value) + return ordered + + +__all__ = [ + "ACCESS_CLASSES", + "ACCESS_LICENSED", + "ACCESS_PUBLIC", + "ACCESS_RESTRICTED", + "ArtifactRegistrationReport", + "DEFAULT_ACCESS", + "DEFAULT_MANIFEST_KIND", + "HashOnlyRegistrationError", + "ListSpecRejected", + "MANIFEST_KINDS", + "MICRODATA_RELEASE_KIND", + "ManifestAccessError", + "MicrodataReleaseNotParseableError", + "PUBLISHER_TABLE_KIND", + "entry_access", + "is_hash_only", + "is_microdata_release", + "iter_file_specs", + "manifest_kind", + "normalize_access", + "recorded_r2", + "register_hash_only_artifact", + "registration_id", + "safe_entry_access", + "safe_manifest_kind", + "stores_bytes", + "validate_file_entry", +] diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 4ddc1807..40f2f028 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -33,6 +33,10 @@ ) from chronicle.env import env_flag, env_value from chronicle.epoch import SCHEMA_IDS, schema_id +from chronicle.registration import ( + MicrodataReleaseNotParseableError, + is_microdata_release, +) from chronicle.sources.cells import ( SourceArtifactMetadata, SourceCell, @@ -849,16 +853,38 @@ def _source_artifact_metadata( raw_r2_uri=raw_r2.get("uri"), ) - def _artifact_content( - self, - year: int, - ) -> tuple[bytes, str, str, dict[str, str]]: + def manifest_payload(self) -> dict[str, Any]: + """Load the artifact manifest this package spec points at.""" manifest_path = files(self.resource_package).joinpath( self.resource_directory, self.manifest, ) with manifest_path.open("r", encoding="utf-8") as file: - manifest = yaml.safe_load(file) + return yaml.safe_load(file) or {} + + def assert_parseable_manifest(self) -> None: + """Refuse to parse a manifest that registers a microdata release. + + Microdata registration is manifest-level identity: no source package + parses a release, and no microdata row, cell, or fact enters Chronicle + (``docs/adr-chronicle-raw-microdata-identity.md``). + """ + manifest = self.manifest_payload() + if is_microdata_release(manifest): + raise MicrodataReleaseNotParseableError( + f"{self.resource_directory}/{self.manifest} registers a " + "microdata release. Registration is identity only: no source " + "package parses a microdata release and no microdata rows, " + "cells, or facts enter Chronicle." + ) + + def _artifact_content( + self, + year: int, + ) -> tuple[bytes, str, str, dict[str, str]]: + manifest = self.manifest_payload() + if is_microdata_release(manifest): + self.assert_parseable_manifest() spec = _year_mapping(manifest["files"], self.artifact_year or year) artifact_path = files(self.resource_package).joinpath( self.resource_directory, @@ -1211,6 +1237,31 @@ def validate_source_package( errors=tuple(errors), ) + try: + package.artifact.assert_parseable_manifest() + except MicrodataReleaseNotParseableError as exc: + errors.append( + SourcePackageIssue( + code="microdata_release_not_parseable", + message=str(exc), + ) + ) + return SourcePackageValidationReport( + package_id=package.package_id, + package_path=str(package.package_path), + year=year, + counts=counts, + errors=tuple(errors), + warnings=tuple(warnings), + ) + except (FileNotFoundError, OSError, ValueError) as exc: + errors.append( + SourcePackageIssue( + code="source_artifact_manifest_unreadable", + message=str(exc), + ) + ) + try: package.artifact._artifact_content(year) except (FileNotFoundError, KeyError, OSError, ValueError) as exc: From b774dbef4dfe78695526d2b69645902f82e85436 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:32:50 -0400 Subject: [PATCH 27/42] Register FRS 2023-24 and SPI 2022-23 hash-only, add the emit/plan script scripts/register_microdata_releases.py resolves a declarative catalogue of releases against a read-only PolicyEngine/microcosm checkout. Every sha256, size, filename, and vintage is read verbatim from Microcosm's reviewed pins; the script authors only the Chronicle identity and the publisher terms. `emit` writes hash-only kind: microdata_release manifests for licensed and restricted releases: 14 distinct DWP FRS 2023-24 tabs (19 references across five Microcosm stages resolve to 14 distinct files) and the HMRC SPI Public Use Tape 2022-23. No bytes are fetched or stored and no R2 key is recorded. BE-SILC 2023 is reported as a blocker rather than registered: Microcosm pins it with no sha256, size, or per-file filename, and no hash is invented. `plan` prints the exact fetch-artifact commands for public releases, copying publisher URLs verbatim and printing TODO where Microcosm records none. Also updates the two artifact-count assertions for the new hash_only_count and hash_only_refused_count report keys. Co-Authored-By: Claude Fable 5.1 --- db/data/dwp/frs_2023_24/manifest.yaml | 288 ++++++ .../spi_public_use_tape_2022_23/manifest.yaml | 29 + scripts/register_microdata_releases.py | 863 ++++++++++++++++++ tests/test_chronicle_artifacts.py | 2 + 4 files changed, 1182 insertions(+) create mode 100644 db/data/dwp/frs_2023_24/manifest.yaml create mode 100644 db/data/hmrc/spi_public_use_tape_2022_23/manifest.yaml create mode 100644 scripts/register_microdata_releases.py diff --git a/db/data/dwp/frs_2023_24/manifest.yaml b/db/data/dwp/frs_2023_24/manifest.yaml new file mode 100644 index 00000000..fae85af5 --- /dev/null +++ b/db/data/dwp/frs_2023_24/manifest.yaml @@ -0,0 +1,288 @@ +source_id: dwp +package_id: dwp-frs-2023-24 +kind: microdata_release +dataset: dwp_dwp-frs-2023-24 +publisher: Department for Work and Pensions +table: Family Resources Survey 2023-24 +files: + 2023: + - filename: accounts.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: c5e31932bfd06087f835d2c83c0984c85a93409bf5ef85b699cb0958abcba1ea + size_bytes: 1807921 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: adult.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: e09f9647d03585c81a528636028b2ed495f8f1fbcf64c5e7b4fe521b67367e06 + size_bytes: 35323384 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: benefits.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: ff30d054cc659bcf23b44c492d98cfd701c0bfdb63e8e9aa9769b490ba9d636b + size_bytes: 4460292 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: benunit.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: 88946815eace8561516d5cbb442c27e319c1e90abc381fb2338f0126e3b9e05b + size_bytes: 21213867 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: child.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: b5dc84fe8b002ee925e61fae23fed27b11537af9fb174f1d07d9cc1748b9702e + size_bytes: 2913156 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: chldcare.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: 566e0ebca1d5e2f3e424e556c91f4cb583d17dadfdfa59feb3841eda7e5976a3 + size_bytes: 273837 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: extchild.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: 8d358d7ee66ee4a7ceab87b4f24fbbf21ac86dc038dc7831e51fb271f96a57ec + size_bytes: 18677 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: househol.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: 5fd26b8b675f33b3b30c9ac789a18da17de734790f77e00ded287d1c3a187b30 + size_bytes: 12387117 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: job.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: 88b77ffe06865f029f713bb1d55ff12bdea8a1234de5bc293e72458fe64f3a74 + size_bytes: 10934873 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: maint.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: f2dc924eb5a51b0c357791693d15b431327dc39c6421011efb313d88bf839695 + size_bytes: 15440 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: mortgage.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: ce36b477d67837c469608a0d68f7ef269ac04758974235f1157d2f6b92cdbfdc + size_bytes: 631783 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: oddjob.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: b4ba3dd3151f73a01422983c60514a3e38458ddfa4fb33ae4ed0326873406305 + size_bytes: 5165 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: penprov.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: ee001461c40306ec24b38b2881e1774121114266a2ee449d606cd0a811c37731 + size_bytes: 522313 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. + - filename: pension.tab + access: licensed + licence: UK Data Service End User Licence + vintage: '2023_24' + sha256: 150d6fad1fce81254fb7aea1526fbb00b63d4027d6e2ac4c26bb90aea3127eb7 + size_bytes: 1225838 + access_route: UK Data Service study SN 9367 under its End User Licence. Bytes + stay in the licensed environment the consumer already operates; no Chronicle + credential grants access to them. + doi: 10.5255/UKDA-SN-9367-2 + study: UK Data Service SN 9367 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json + (frs_spine) + notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; + its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages + cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across + all five stages (identical SHA-256), so the registration carries the study reference + that also carries a DOI. SHA-256 transcribed verbatim from the consumer's reviewed + pin; Chronicle holds no bytes for this release and did not recompute the checksum. diff --git a/db/data/hmrc/spi_public_use_tape_2022_23/manifest.yaml b/db/data/hmrc/spi_public_use_tape_2022_23/manifest.yaml new file mode 100644 index 00000000..30eb6b24 --- /dev/null +++ b/db/data/hmrc/spi_public_use_tape_2022_23/manifest.yaml @@ -0,0 +1,29 @@ +source_id: hmrc +package_id: hmrc-spi-public-use-tape-2022-23 +kind: microdata_release +dataset: hmrc_hmrc-spi-public-use-tape-2022-23 +publisher: HM Revenue and Customs +table: Survey of Personal Incomes Public Use Tape 2022-23 +files: + 2022: + - filename: put2223uk.tab + access: restricted + licence: UK Data Service End User Licence (study SN 9422) + vintage: 2022-23 + sha256: 5ef829461060c91a2a47be59ad541d9b519fc3976d66ca80d4920f711bb96f66 + size_bytes: 141323762 + access_route: UK Data Service study SN 9422. Microcosm reaches the bytes through + PolicyEngine's licensed copy in the private policyengine/policyengine-uk-data-private + Hugging Face repository (spi_2022_23.zip); no Chronicle credential grants access + to them. + doi: 10.5255/UKDA-SN-9422-1 + study: UK Data Service SN 9422 + verified_at: '2026-09-02' + hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json + (hmrc_spi_income) + notes: 'Microcosm classes this artifact kind: private_microdata with access: private_local_input. + Chronicle registers it restricted because the bytes are held only in a private + mirror; if the UKDS terms for SN 9422 are confirmed as End User Licence it can + be reclassified licensed, which changes nothing about storage — both classes + are hash-only. SHA-256 transcribed verbatim from the consumer''s reviewed pin; + Chronicle holds no bytes for this release and did not recompute the checksum.' diff --git a/scripts/register_microdata_releases.py b/scripts/register_microdata_releases.py new file mode 100644 index 00000000..efc50609 --- /dev/null +++ b/scripts/register_microdata_releases.py @@ -0,0 +1,863 @@ +"""Register raw microdata releases from a Microcosm source-stages manifest. + +Chronicle registers every raw microdata release its consumers build from and +stores the bytes of only those a publisher permits it to redistribute +(``docs/adr-chronicle-raw-microdata-identity.md``). This script drives both +halves of that from the pins Microcosm already reviewed: + +``emit`` + Write hash-only ``kind: microdata_release`` manifests for ``licensed`` and + ``restricted`` releases. Every checksum, size, filename, and vintage is read + verbatim from the Microcosm source-stages JSON; nothing is recomputed and + nothing is invented. A release Microcosm pins without a checksum is reported + as a blocker and never registered. + +``plan`` + Print the exact ``chronicle fetch-artifact ... --upload-r2`` commands to run + from a networked machine for ``public`` releases, whose bytes Chronicle does + archive. Publisher URLs are copied verbatim from the Microcosm manifest; a + release whose manifest carries no URL prints a ``TODO`` instead of a guess. + +The catalogue below is the only authored content: it maps a Microcosm artifact +onto Chronicle's ``{source_id, package_id, year, sha256, filename}`` identity +and records the publisher's terms. Run it read-only against a Microcosm +checkout; this script never writes to that repository. + +Usage:: + + python scripts/register_microdata_releases.py emit \\ + --microcosm-root ~/PolicyEngine/microcosm \\ + --root db/data --verified-at 2026-09-02 + + python scripts/register_microdata_releases.py plan \\ + --microcosm-root ~/PolicyEngine/microcosm --root db/data +""" + +from __future__ import annotations + +import argparse +from collections.abc import Iterator, Mapping, Sequence +from dataclasses import dataclass, field +import json +from pathlib import Path +import shlex +import sys +from typing import Any + +# Allow `python scripts/register_microdata_releases.py` from a checkout. +sys.path.insert(0, str(Path(__file__).resolve().parent.parent)) + +from chronicle.registration import ( # noqa: E402 + ACCESS_PUBLIC, + HashOnlyRegistrationError, + register_hash_only_artifact, +) + +#: Microcosm's per-artifact ``kind`` mapped onto Chronicle's access class. +#: +#: ``private_microdata`` maps to ``restricted`` rather than ``licensed``: the +#: bytes are held only in a private mirror, and over-classifying is the safe +#: direction because both classes are registered hash-only and neither ever +#: places bytes in a Chronicle store. +ACCESS_BY_MICROCOSM_KIND: dict[str, str] = { + "public_microdata": "public", + "licensed_microdata": "licensed", + "private_microdata": "restricted", + "restricted_microdata": "restricted", +} + +#: Provenance sentence written onto every registration this script emits. +HASH_PROVENANCE = ( + "SHA-256 transcribed verbatim from the consumer's reviewed pin; Chronicle " + "holds no bytes for this release and did not recompute the checksum." +) + + +@dataclass(frozen=True) +class ArtifactSelector: + """Locate one artifact inside a Microcosm source-stages JSON file. + + ``stage`` names the build stage; ``match`` is a set of artifact fields that + must equal the given values. A selector that matches nothing, or matches + inconsistent bytes across stages, is a hard error rather than a guess. + """ + + stage: str | None = None + match: Mapping[str, Any] = field(default_factory=dict) + kind: str | None = None + + +@dataclass(frozen=True) +class Release: + """One Chronicle registration drawn from a Microcosm pin.""" + + release_id: str + manifest: str + selector: ArtifactSelector + source_id: str + package_id: str + package_dir: str + year: int + table: str + publisher: str + licence: str + access: str + #: Artifact field holding the publisher filename, when not ``filename``. + filename_field: str = "filename" + #: Artifact field holding the publisher URL, for ``public`` releases. + url_field: str = "locator" + study: str | None = None + doi: str | None = None + source_page: str | None = None + access_route: str | None = None + vintage: str | None = None + notes: str | None = None + #: Whether Microcosm's pinned ``sha256`` is of the publisher artifact + #: itself. It is not when the pin covers a derived file or an extracted + #: archive member, and such a hash must never be presented as the checksum + #: a fetch should reproduce. + pinned_sha_is_publisher_bytes: bool = True + #: Set when Microcosm pins the release without a checksum. + blocker: str | None = None + + +UK_STAGES = "packages/microcosm-build/src/microcosm/build/uk/source_stages.json" +UK_HMRC_STAGES = ( + "packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json" +) +BE_STAGES = "packages/microcosm-build/src/microcosm/build/be/source_stages.json" +US_STAGES = "packages/microcosm-build/src/microcosm/build/us/source_stages.json" +US_ACS_2024 = ( + "packages/microcosm-build/src/microcosm/build/us_runtime/acs_2024_1yr_sources.json" +) + +#: FRS 2023-24 tabs, in the order Microcosm's ``frs_spine`` stage lists them. +FRS_TABS: tuple[str, ...] = ( + "accounts", + "adult", + "benefits", + "benunit", + "child", + "chldcare", + "extchild", + "househol", + "job", + "maint", + "mortgage", + "oddjob", + "penprov", + "pension", +) + +FRS_LICENCE = "UK Data Service End User Licence" +FRS_STUDY = "UK Data Service SN 9367" +FRS_DOI = "10.5255/UKDA-SN-9367-2" +FRS_ACCESS_ROUTE = ( + "UK Data Service study SN 9367 under its End User Licence. Bytes stay in " + "the licensed environment the consumer already operates; no Chronicle " + "credential grants access to them." +) +FRS_NOTES = ( + "Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI " + "10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, " + "frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same " + "2023_24 tabs. The tabs are the same bytes across all five stages (identical " + "SHA-256), so the registration carries the study reference that also carries " + "a DOI. " + HASH_PROVENANCE +) + +CATALOGUE: tuple[Release, ...] = ( + *( + Release( + release_id=f"dwp-frs-2023-24:{tab}", + manifest=UK_STAGES, + selector=ArtifactSelector( + stage="frs_spine", + kind="licensed_microdata", + match={"table": tab}, + ), + source_id="dwp", + package_id="dwp-frs-2023-24", + package_dir="dwp/frs_2023_24", + year=2023, + table="Family Resources Survey 2023-24", + publisher="Department for Work and Pensions", + licence=FRS_LICENCE, + access="licensed", + filename_field="locator", + study=FRS_STUDY, + doi=FRS_DOI, + access_route=FRS_ACCESS_ROUTE, + notes=FRS_NOTES, + ) + for tab in FRS_TABS + ), + Release( + release_id="hmrc-spi-public-use-tape-2022-23:put2223uk", + manifest=UK_HMRC_STAGES, + selector=ArtifactSelector( + stage="hmrc_spi_income", + kind="private_microdata", + match={"filename": "put2223uk.tab"}, + ), + source_id="hmrc", + package_id="hmrc-spi-public-use-tape-2022-23", + package_dir="hmrc/spi_public_use_tape_2022_23", + year=2022, + table="Survey of Personal Incomes Public Use Tape 2022-23", + publisher="HM Revenue and Customs", + licence="UK Data Service End User Licence (study SN 9422)", + access="restricted", + study="UK Data Service SN 9422", + doi="10.5255/UKDA-SN-9422-1", + access_route=( + "UK Data Service study SN 9422. Microcosm reaches the bytes through " + "PolicyEngine's licensed copy in the private " + "policyengine/policyengine-uk-data-private Hugging Face repository " + "(spi_2022_23.zip); no Chronicle credential grants access to them." + ), + notes=( + "Microcosm classes this artifact kind: private_microdata with " + "access: private_local_input. Chronicle registers it restricted " + "because the bytes are held only in a private mirror; if the UKDS " + "terms for SN 9422 are confirmed as End User Licence it can be " + "reclassified licensed, which changes nothing about storage — both " + "classes are hash-only. " + HASH_PROVENANCE + ), + ), + Release( + release_id="statbel-be-silc-2023", + manifest=BE_STAGES, + selector=ArtifactSelector( + stage="silc_load", + kind="restricted_microdata", + ), + source_id="statbel", + package_id="statbel-be-silc-2023", + package_dir="statbel/be_silc_2023", + year=2023, + table="BE-SILC 2023 scientific-use files (D, R, H, P)", + publisher="Statbel", + licence="Statbel/Eurostat scientific-use", + access="restricted", + source_page="https://statbel.fgov.be/en/themes/households/poverty-and-living-conditions", + access_route=( + "Statbel BE-SILC scientific-use files: D (household register), " + "R (personal register), H (household data), P (personal data)." + ), + blocker=( + "Microcosm's be/source_stages.json pins the BE-SILC scientific-use " + "files with no sha256, no size_bytes, and no per-file filename — it " + "names only the four file roles. A registration is identified by " + "{source_id, package_id, year, sha256, filename}, so this release " + "cannot be registered until the consumer publishes a reviewed " + "checksum per file. No hash is invented here." + ), + ), + # ---- public releases: bytes are archived, so these are fetch-and-upload ---- + Release( + release_id="census-cps-asec-2023", + manifest=US_STAGES, + selector=ArtifactSelector( + stage="weeks_unemployed_input", + kind="public_microdata", + match={"member": "pppub23.csv"}, + ), + source_id="census_cps", + package_id="census-cps-asec-2023", + package_dir="census/cps_asec_2023", + year=2023, + table="CPS Annual Social and Economic Supplement 2023 public-use files", + publisher="U.S. Census Bureau", + licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + source_page="https://www.census.gov/programs-surveys/cps/data/datasets.html", + ), + Release( + release_id="census-cps-basic-monthly-2024", + manifest=US_STAGES, + selector=ArtifactSelector( + stage="org_wages", + kind="public_microdata", + ), + source_id="census_cps", + package_id="census-cps-basic-monthly-2024", + package_dir="census/cps_basic_monthly_2024", + year=2024, + table="CPS basic monthly public-use files, January-December 2024", + publisher="U.S. Census Bureau", + licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + source_page="https://www2.census.gov/programs-surveys/cps/datasets/2024/basic/", + notes=( + "Microcosm pins the twelve monthly files as the locator string " + "'jan24pub through dec24pub' with no per-file URL, filename " + "extension, or checksum. The publisher directory is the stage " + "source; the twelve filenames must be read off that directory " + "before the fetch commands can be completed." + ), + ), + Release( + release_id="census-acs-pums-2022-household", + manifest=US_STAGES, + selector=ArtifactSelector( + stage="acs_rent", + kind="versioned_derived_microdata", + ), + source_id="census_acs", + package_id="census-acs-pums-2022-1yr", + package_dir="census/acs_pums_2022_1yr", + year=2022, + table="ACS 2022 1-Year PUMS household file", + publisher="U.S. Census Bureau", + licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + url_field="official_household_source", + source_page="https://www.census.gov/programs-surveys/acs", + pinned_sha_is_publisher_bytes=False, + notes=( + "The Microcosm artifact's own sha256 belongs to the derived " + "acs_2022.h5, not to the publisher zip; only the publisher URL is " + "reused here. The fetch computes the release checksum." + ), + ), + Release( + release_id="census-acs-pums-2022-person", + manifest=US_STAGES, + selector=ArtifactSelector( + stage="acs_rent", + kind="versioned_derived_microdata", + ), + source_id="census_acs", + package_id="census-acs-pums-2022-1yr", + package_dir="census/acs_pums_2022_1yr", + year=2022, + table="ACS 2022 1-Year PUMS person file", + publisher="U.S. Census Bureau", + licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + url_field="official_person_source", + source_page="https://www.census.gov/programs-surveys/acs", + pinned_sha_is_publisher_bytes=False, + notes=( + "The Microcosm artifact's own sha256 belongs to the derived " + "acs_2022.h5, not to the publisher zip; only the publisher URL is " + "reused here. The fetch computes the release checksum." + ), + ), + Release( + release_id="census-acs-pums-2024-household", + manifest=US_ACS_2024, + selector=ArtifactSelector(match={"role": "household"}), + source_id="census_acs", + package_id="census-acs-pums-2024-1yr", + package_dir="census/acs_pums_2024_1yr", + year=2024, + table="ACS 2024 1-Year PUMS household file", + publisher="U.S. Census Bureau", + licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + url_field="url", + source_page="https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/", + ), + Release( + release_id="census-acs-pums-2024-person", + manifest=US_ACS_2024, + selector=ArtifactSelector(match={"role": "person"}), + source_id="census_acs", + package_id="census-acs-pums-2024-1yr", + package_dir="census/acs_pums_2024_1yr", + year=2024, + table="ACS 2024 1-Year PUMS person file", + publisher="U.S. Census Bureau", + licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + url_field="url", + source_page="https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/", + ), + Release( + release_id="federal-reserve-scf-2022-summary", + manifest=US_STAGES, + selector=ArtifactSelector( + stage="scf_wealth", + kind="public_microdata", + match={"member": "rscfp2022.dta"}, + ), + source_id="federal_reserve", + package_id="federal-reserve-scf-2022", + package_dir="federal_reserve/scf_2022", + year=2022, + table="Survey of Consumer Finances 2022 summary extract", + publisher="Board of Governors of the Federal Reserve System", + licence="Federal Reserve Board public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + source_page="https://www.federalreserve.gov/econres/scfindex.htm", + ), + Release( + release_id="federal-reserve-scf-2022-full", + manifest=US_STAGES, + selector=ArtifactSelector( + stage="scf_wealth", + kind="public_microdata", + match={"member": "p22i6.dta"}, + ), + source_id="federal_reserve", + package_id="federal-reserve-scf-2022", + package_dir="federal_reserve/scf_2022", + year=2022, + table="Survey of Consumer Finances 2022 full public data set", + publisher="Board of Governors of the Federal Reserve System", + licence="Federal Reserve Board public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + source_page="https://www.federalreserve.gov/econres/scfindex.htm", + notes=( + "Microcosm records no checksum for this zip: 'Full-file SHA-256 " + "pending one network-enabled provisioning fetch'. The fetch below " + "computes and registers it." + ), + ), + Release( + release_id="census-sipp-2023", + manifest=US_STAGES, + selector=ArtifactSelector( + stage="scf_wealth", + kind="public_microdata", + match={"member": "pu2023.csv"}, + ), + source_id="census_sipp", + package_id="census-sipp-2023", + package_dir="census/sipp_2023", + year=2023, + table="Survey of Income and Program Participation 2023 public-use file", + publisher="U.S. Census Bureau", + licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + access=ACCESS_PUBLIC, + source_page="https://www.census.gov/programs-surveys/sipp.html", + pinned_sha_is_publisher_bytes=False, + notes=( + "Microcosm reaches this file through an immutable Hugging Face " + "mirror (revision 21280dca5995e978d706740a8a4b9b7860cfd7b6) and " + "records no Census URL, so the publisher URL must be read off the " + "SIPP dataset page before the fetch. Microcosm's pinned sha256 " + "5c30439e365fc26483318ef61d1d8f4bb2f0e9d6bb47c22c06756a7698733ee2 " + "and size 3726010471 are for the mirrored pu2023.csv member, not " + "for whatever archive the Census page serves, so they are not the " + "checksum this fetch should be expected to reproduce." + ), + ), +) + + +class CatalogueError(RuntimeError): + """Raised when the catalogue cannot be resolved against Microcosm.""" + + +def load_manifest(microcosm_root: Path, relative: str) -> dict[str, Any]: + """Load one Microcosm JSON manifest, read-only.""" + path = microcosm_root / relative + if not path.exists(): + raise CatalogueError(f"Microcosm manifest not found: {path}") + payload = json.loads(path.read_text(encoding="utf-8")) + if not isinstance(payload, dict): + raise CatalogueError(f"Microcosm manifest must be an object: {path}") + return payload + + +def iter_manifest_artifacts( + payload: Mapping[str, Any], +) -> Iterator[tuple[Mapping[str, Any], Mapping[str, Any]]]: + """Yield ``(stage, artifact)`` pairs from either Microcosm manifest shape. + + Source-stages manifests nest artifacts under ``stages``; the ACS runtime + manifest carries a flat top-level ``artifacts`` list with no stages. + """ + stages = payload.get("stages") + if isinstance(stages, list): + for stage in stages: + if not isinstance(stage, Mapping): + continue + for artifact in stage.get("artifacts") or (): + if isinstance(artifact, Mapping): + yield stage, artifact + return + for artifact in payload.get("artifacts") or (): + if isinstance(artifact, Mapping): + yield payload, artifact + + +def select_artifact( + payload: Mapping[str, Any], + selector: ArtifactSelector, + *, + release_id: str, +) -> tuple[Mapping[str, Any], Mapping[str, Any]]: + """Return the single ``(stage, artifact)`` a selector identifies. + + Several Microcosm stages reference the same bytes. Duplicates are accepted + only when they agree on every field the registration reads; disagreement is + an error, never a silent first-match. + """ + matches: list[tuple[Mapping[str, Any], Mapping[str, Any]]] = [] + for stage, artifact in iter_manifest_artifacts(payload): + if selector.stage is not None and stage.get("stage") != selector.stage: + continue + if selector.kind is not None and artifact.get("kind") != selector.kind: + continue + if any(artifact.get(key) != value for key, value in selector.match.items()): + continue + matches.append((stage, artifact)) + if not matches: + raise CatalogueError( + f"{release_id}: no Microcosm artifact matches {selector}. The " + "consumer manifest changed; re-derive the catalogue rather than " + "hand-editing a registration." + ) + first_stage, first = matches[0] + for _stage, artifact in matches[1:]: + if dict(artifact) != dict(first): + raise CatalogueError( + f"{release_id}: Microcosm pins conflicting values for this " + "artifact across stages; refusing to choose between them." + ) + return first_stage, first + + +@dataclass(frozen=True) +class ResolvedRelease: + """A catalogue entry resolved against Microcosm's pinned artifact.""" + + release: Release + stage: Mapping[str, Any] + artifact: Mapping[str, Any] + + @property + def filename(self) -> str | None: + """Publisher filename, from the release's declared artifact field.""" + value = self.artifact.get(self.release.filename_field) + if not value: + return None + # A locator may be a URL or a bare filename; take the last path segment. + return str(value).rstrip("/").rsplit("/", 1)[-1] + + @property + def url(self) -> str | None: + """Publisher URL, when the Microcosm artifact records one.""" + value = self.artifact.get(self.release.url_field) + text = str(value).strip() if value else "" + return text if text.startswith(("http://", "https://")) else None + + @property + def sha256(self) -> str | None: + """Checksum Microcosm pins for these bytes, if any.""" + value = self.artifact.get("sha256") + return str(value) if value else None + + @property + def size_bytes(self) -> int | None: + """Size Microcosm pins for these bytes, if any.""" + value = self.artifact.get("size_bytes") + return int(value) if isinstance(value, int) else None + + @property + def vintage(self) -> str: + """Publisher vintage label, from the release or the artifact.""" + return str(self.release.vintage or self.artifact.get("vintage") or "") + + +def resolve( + microcosm_root: Path, + releases: Sequence[Release], +) -> list[ResolvedRelease]: + """Resolve every catalogue entry against the Microcosm checkout.""" + payloads: dict[str, dict[str, Any]] = {} + resolved: list[ResolvedRelease] = [] + for release in releases: + if release.manifest not in payloads: + payloads[release.manifest] = load_manifest( + microcosm_root, release.manifest + ) + stage, artifact = select_artifact( + payloads[release.manifest], + release.selector, + release_id=release.release_id, + ) + resolved.append(ResolvedRelease(release=release, stage=stage, artifact=artifact)) + return resolved + + +def hash_source(release: Release) -> str: + """Return the provenance pointer recorded on a registration.""" + return f"PolicyEngine/microcosm {release.manifest} ({release.selector.stage})" + + +def emit( + resolved: Sequence[ResolvedRelease], + *, + root: Path, + verified_at: str, + allow_reissue: bool = False, +) -> tuple[list[dict[str, Any]], list[dict[str, str]]]: + """Write hash-only manifests for every registrable non-public release. + + Returns ``(registrations, blockers)``. A release Microcosm pins without a + checksum is a blocker, not a registration: no hash is ever invented. + """ + registrations: list[dict[str, Any]] = [] + blockers: list[dict[str, str]] = [] + for item in resolved: + release = item.release + if release.access == ACCESS_PUBLIC: + continue + if release.blocker: + blockers.append({"release": release.release_id, "reason": release.blocker}) + continue + filename = item.filename + checksum = item.sha256 + if not filename or not checksum: + blockers.append( + { + "release": release.release_id, + "reason": ( + "Microcosm pins this release without a " + f"{'filename' if not filename else 'sha256'}; a " + "registration needs both. No value is invented." + ), + } + ) + continue + report = register_hash_only_artifact( + source_id=release.source_id, + package_id=release.package_id, + year=release.year, + output_dir=root / release.package_dir, + filename=filename, + sha256=checksum, + licence=release.licence, + access=release.access, + vintage=item.vintage, + size_bytes=item.size_bytes, + source_page=release.source_page, + access_route=release.access_route, + doi=release.doi, + study=release.study, + table=release.table, + publisher=release.publisher, + verified_at=verified_at, + hash_source=hash_source(release), + notes=release.notes or HASH_PROVENANCE, + allow_reissue=allow_reissue, + ) + registrations.append(report.to_dict()) + return registrations, blockers + + +def fetch_command( + item: ResolvedRelease, + *, + root: Path, + r2_bucket: str, +) -> tuple[str, list[str]]: + """Return ``(command, todos)`` for one public release. + + The command is the exact ``chronicle fetch-artifact`` invocation to run from + a networked machine. Anything Microcosm does not pin becomes a TODO rather + than a fabricated value. + """ + release = item.release + todos: list[str] = [] + url = item.url + if url is None: + url = "TODO_PUBLISHER_URL" + todos.append( + f"{release.release_id}: Microcosm records no publisher URL " + f"(field {release.url_field!r}); read it off {release.source_page}." + ) + argv = [ + "uv", + "run", + "chronicle", + "fetch-artifact", + "--source-id", + release.source_id, + "--package-id", + release.package_id, + "--year", + str(release.year), + "--out-dir", + str(root / release.package_dir), + "--source-url", + url, + ] + if release.source_page: + argv += ["--source-page", release.source_page] + argv += [ + "--table", + release.table, + "--access", + ACCESS_PUBLIC, + "--licence", + release.licence, + "--upload-r2", + "--r2-bucket", + r2_bucket, + ] + if item.sha256 and release.pinned_sha_is_publisher_bytes: + todos.append( + f"{release.release_id}: expect sha256 {item.sha256}" + + (f" and size {item.size_bytes}" if item.size_bytes else "") + + " — fail the registration if the fetched bytes differ." + ) + elif item.sha256: + todos.append( + f"{release.release_id}: Microcosm's pinned sha256 {item.sha256} is " + "NOT the publisher artifact's checksum — do not use it to verify " + "this fetch. See the note below." + ) + if release.notes: + todos.append(f"{release.release_id}: {release.notes}") + return shlex.join(argv), todos + + +def plan( + resolved: Sequence[ResolvedRelease], + *, + root: Path, + r2_bucket: str, +) -> tuple[list[str], list[str]]: + """Return the fetch commands and TODOs for every public release.""" + commands: list[str] = [] + todos: list[str] = [] + for item in resolved: + if item.release.access != ACCESS_PUBLIC: + continue + command, item_todos = fetch_command(item, root=root, r2_bucket=r2_bucket) + commands.append(command) + todos.extend(item_todos) + return commands, todos + + +def build_parser() -> argparse.ArgumentParser: + """Build the command-line parser.""" + parser = argparse.ArgumentParser( + description=__doc__, + formatter_class=argparse.RawDescriptionHelpFormatter, + ) + parser.add_argument( + "--microcosm-root", + type=Path, + default=Path.home() / "PolicyEngine" / "microcosm", + help="Read-only path to a PolicyEngine/microcosm checkout.", + ) + parser.add_argument( + "--root", + type=Path, + default=Path("db/data"), + help="Chronicle data root that holds the package directories.", + ) + parser.add_argument( + "--release", + action="append", + default=None, + help="Limit to these release IDs. Repeatable.", + ) + parser.add_argument( + "--json", + action="store_true", + help="Emit machine-readable JSON instead of prose.", + ) + subparsers = parser.add_subparsers(dest="command", required=True) + + emit_parser = subparsers.add_parser( + "emit", + help="Write hash-only manifests for licensed and restricted releases", + ) + emit_parser.add_argument( + "--verified-at", + required=True, + help=( + "Date the pins were verified against Microcosm, as YYYY-MM-DD. " + "Required so repeated runs are byte-stable." + ), + ) + emit_parser.add_argument( + "--allow-reissue", + action="store_true", + help="Register different bytes alongside an existing pin for a filename.", + ) + + plan_parser = subparsers.add_parser( + "plan", + help="Print the fetch commands to run for public releases", + ) + plan_parser.add_argument( + "--r2-bucket", + default="ledger-raw", + help="Raw bucket the fetch should upload to.", + ) + return parser + + +def main(argv: list[str] | None = None) -> int: + """Run the registration script.""" + args = build_parser().parse_args(argv) + releases = CATALOGUE + if args.release: + wanted = set(args.release) + releases = tuple(r for r in CATALOGUE if r.release_id in wanted) + missing = wanted - {r.release_id for r in releases} + if missing: + print(f"Unknown release IDs: {sorted(missing)}", file=sys.stderr) + return 2 + + microcosm_root = args.microcosm_root.expanduser() + try: + resolved = resolve(microcosm_root, releases) + except CatalogueError as exc: + print(str(exc), file=sys.stderr) + return 1 + + if args.command == "emit": + try: + registrations, blockers = emit( + resolved, + root=args.root, + verified_at=args.verified_at, + allow_reissue=args.allow_reissue, + ) + except HashOnlyRegistrationError as exc: + print(str(exc), file=sys.stderr) + return 1 + if args.json: + print( + json.dumps( + {"registrations": registrations, "blockers": blockers}, + indent=2, + sort_keys=True, + ) + ) + else: + for registration in registrations: + print(f"registered {registration['registration']}") + for blocker in blockers: + print(f"BLOCKED {blocker['release']}: {blocker['reason']}") + print( + f"\n{len(registrations)} registration(s), {len(blockers)} blocker(s)." + ) + return 0 + + commands, todos = plan(resolved, root=args.root, r2_bucket=args.r2_bucket) + if args.json: + print(json.dumps({"commands": commands, "todos": todos}, indent=2)) + return 0 + print("# Run from a networked machine with R2 credentials.\n") + for command in commands: + print(command + "\n") + if todos: + print("# TODO") + for todo in todos: + print(f"# {todo}") + return 0 + + +if __name__ == "__main__": + raise SystemExit(main()) diff --git a/tests/test_chronicle_artifacts.py b/tests/test_chronicle_artifacts.py index f41f3938..5658c32a 100644 --- a/tests/test_chronicle_artifacts.py +++ b/tests/test_chronicle_artifacts.py @@ -179,6 +179,7 @@ def test_fetch_source_artifact_writes_manifest_and_inventory(tmp_path): assert inventory.counts == { "artifact_count": 1, "checksum_mismatch_count": 0, + "hash_only_count": 0, "manifest_count": 1, "missing_count": 0, "r2_link_count": 0, @@ -245,6 +246,7 @@ def test_publish_source_artifacts_uploads_manifest_entries(tmp_path): assert report.counts == { "artifact_count": 1, "failed_count": 0, + "hash_only_refused_count": 0, "manifest_count": 1, "r2_link_count": 1, "skipped_count": 0, From 58298fc84fecd231201ed4e49ef007cc75b04700 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:40:44 -0400 Subject: [PATCH 28/42] Test the registration refusal surface end to end 59 tests covering the access and kind vocabularies, every manifest-entry validation code, and each command's behaviour: register-artifact refuses public access, a missing or malformed checksum, a missing licence, vintage, access route, or verification timestamp, bytes on disk, a publisher_table manifest, a foreign source_id, and an unrequested reissue; fetch-artifact refuses a hash-only access class and refuses to pull bytes over an existing registration without reading the source; publish-raw refuses hash-only entries without touching bytes; inventory-artifacts accepts them with no local file and flags them when bytes appear. Also pins the committed FRS and SPI registrations as identity-only: no storage block, no R2 key, no bytes beside the manifest. Reorders the fetch refusals so overwriting a hash-only registration reports the byte-boundary violation rather than prompting for --licence. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 5 +- scripts/register_microdata_releases.py | 8 +- .../test_chronicle_microdata_registration.py | 797 ++++++++++++++++++ 3 files changed, 805 insertions(+), 5 deletions(-) create mode 100644 tests/test_chronicle_microdata_registration.py diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 1015c367..893ff6e5 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -706,6 +706,10 @@ def fetch_source_artifact( year=year, ) licence_text = licence.strip() if isinstance(licence, str) else None + # The byte boundary is checked first: overwriting a hash-only registration + # with bytes is the more serious refusal, and its message is the one the + # caller needs, not a prompt to supply a licence. + _assert_no_hash_only_entry(existing_manifest, manifest_path, year, filename) if ( safe_manifest_kind(existing_manifest)[0] == MICRODATA_RELEASE_KIND and not licence_text @@ -714,7 +718,6 @@ def fetch_source_artifact( f"{manifest_path} registers a microdata release, so every entry " "must record its publisher licence; pass --licence." ) - _assert_no_hash_only_entry(existing_manifest, manifest_path, year, filename) fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() content, inferred_filename = _read_artifact(source_url) diff --git a/scripts/register_microdata_releases.py b/scripts/register_microdata_releases.py index efc50609..b93343a4 100644 --- a/scripts/register_microdata_releases.py +++ b/scripts/register_microdata_releases.py @@ -573,15 +573,15 @@ def resolve( resolved: list[ResolvedRelease] = [] for release in releases: if release.manifest not in payloads: - payloads[release.manifest] = load_manifest( - microcosm_root, release.manifest - ) + payloads[release.manifest] = load_manifest(microcosm_root, release.manifest) stage, artifact = select_artifact( payloads[release.manifest], release.selector, release_id=release.release_id, ) - resolved.append(ResolvedRelease(release=release, stage=stage, artifact=artifact)) + resolved.append( + ResolvedRelease(release=release, stage=stage, artifact=artifact) + ) return resolved diff --git a/tests/test_chronicle_microdata_registration.py b/tests/test_chronicle_microdata_registration.py new file mode 100644 index 00000000..b80721e9 --- /dev/null +++ b/tests/test_chronicle_microdata_registration.py @@ -0,0 +1,797 @@ +"""Tests for microdata-release registration: identity without content. + +Chronicle registers every raw microdata release its consumers build from and +stores the bytes of only the ones a publisher permits it to redistribute +(``docs/adr-chronicle-raw-microdata-identity.md``). These tests pin the whole +refusal surface: which access classes may carry bytes, which commands refuse +them, and that a hash-only registration is a complete, valid artifact record +with no local file and no R2 key. +""" + +from __future__ import annotations + +import hashlib +import json +from pathlib import Path + +import pytest +import yaml + +from chronicle.artifacts import ( + fetch_source_artifact, + inventory_source_artifacts, + publish_source_artifacts, +) +from chronicle.harness import main as harness_main +from chronicle.registration import ( + ACCESS_CLASSES, + HashOnlyRegistrationError, + ListSpecRejected, + ManifestAccessError, + MicrodataReleaseNotParseableError, + entry_access, + is_hash_only, + is_microdata_release, + iter_file_specs, + manifest_kind, + normalize_access, + register_hash_only_artifact, + registration_id, + safe_entry_access, + stores_bytes, + validate_file_entry, +) +from chronicle.source_package import SourceArtifactSpec, validate_source_package + + +REPO_ROOT = Path(__file__).resolve().parents[1] +FRS_PACKAGE = REPO_ROOT / "db" / "data" / "dwp" / "frs_2023_24" +SPI_PACKAGE = REPO_ROOT / "db" / "data" / "hmrc" / "spi_public_use_tape_2022_23" + +# A syntactically valid checksum that identifies no real publisher bytes. +FIXTURE_SHA = "a" * 64 + + +def _register(output_dir: Path, **overrides: object) -> object: + """Register a fixture licensed artifact, with per-test overrides.""" + kwargs: dict[str, object] = { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "year": 2023, + "output_dir": output_dir, + "filename": "adult.tab", + "sha256": FIXTURE_SHA, + "licence": "UK Data Service End User Licence", + "access": "licensed", + "vintage": "2023_24", + "size_bytes": 35323384, + "doi": "10.5255/UKDA-SN-9367-2", + "verified_at": "2026-09-02", + } + kwargs.update(overrides) + return register_hash_only_artifact(**kwargs) # type: ignore[arg-type] + + +# -------------------------------------------------------------------------- +# Access and kind vocabularies +# -------------------------------------------------------------------------- + + +def test_access_classes_are_the_closed_contract_set(): + assert ACCESS_CLASSES == ("public", "licensed", "restricted") + + +@pytest.mark.parametrize("access", ACCESS_CLASSES) +def test_only_public_access_may_carry_bytes(access): + assert stores_bytes(access) is (access == "public") + assert is_hash_only(access) is (access != "public") + + +def test_absent_access_is_inferred_public(): + assert normalize_access(None) == "public" + assert entry_access({"filename": "table.xlsx"}) == "public" + + +def test_unknown_access_class_is_refused(): + with pytest.raises(ManifestAccessError, match="Unknown access class"): + normalize_access("internal") + + +def test_unknown_access_class_falls_back_to_restricted_not_public(): + # Never upload bytes because a class failed to parse. + assert safe_entry_access({"access": "internal"}) == "restricted" + + +def test_manifest_kind_defaults_to_publisher_table_and_rejects_unknown(): + assert manifest_kind(None) == "publisher_table" + assert manifest_kind({}) == "publisher_table" + assert is_microdata_release({"kind": "microdata_release"}) is True + with pytest.raises(ManifestAccessError, match="Unknown manifest kind"): + manifest_kind({"kind": "microdata_rows"}) + + +def test_registration_identity_is_the_contract_tuple(): + assert ( + registration_id( + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + sha256=FIXTURE_SHA, + filename="adult.tab", + ) + == f"dwp/dwp-frs-2023-24/2023/{FIXTURE_SHA}/adult.tab" + ) + + +# -------------------------------------------------------------------------- +# Manifest entry validation +# -------------------------------------------------------------------------- + + +def test_microdata_release_entry_requires_access_and_licence(): + errors = validate_file_entry( + {"filename": "adult.tab", "sha256": FIXTURE_SHA}, + kind="microdata_release", + manifest={}, + local_file_exists=False, + ) + + assert "missing_access" in errors + assert "missing_licence" in errors + + +@pytest.mark.parametrize( + ("mutation", "expected_code"), + [ + ({"licence": None}, "missing_licence"), + ({"sha256": None}, "missing_sha256"), + ({"sha256": "not-a-checksum"}, "malformed_sha256"), + ({"sha256": FIXTURE_SHA.upper()}, "malformed_sha256"), + ({"vintage": None}, "missing_vintage"), + ({"doi": None}, "missing_access_route"), + ({"verified_at": None}, "missing_verification_timestamp"), + ], +) +def test_hash_only_entry_reports_each_missing_field(mutation, expected_code): + entry = { + "filename": "adult.tab", + "access": "licensed", + "licence": "UK Data Service End User Licence", + "vintage": "2023_24", + "sha256": FIXTURE_SHA, + "doi": "10.5255/UKDA-SN-9367-2", + "verified_at": "2026-09-02", + } + entry.update(mutation) + entry = {key: value for key, value in entry.items() if value is not None} + + errors = validate_file_entry( + entry, + kind="microdata_release", + manifest={}, + local_file_exists=False, + ) + + assert expected_code in errors + + +def test_hash_only_entry_flags_bytes_and_r2_locations(): + entry = { + "filename": "adult.tab", + "access": "restricted", + "licence": "Statbel/Eurostat scientific-use", + "vintage": "2023", + "sha256": FIXTURE_SHA, + "doi": "10.5255/UKDA-SN-9422-1", + "verified_at": "2026-09-02", + "storage": {"r2": {"bucket": "ledger-raw", "key": "raw/x/y/z"}}, + } + + errors = validate_file_entry( + entry, + kind="microdata_release", + manifest={}, + local_file_exists=True, + ) + + assert "bytes_present_for_hash_only_entry" in errors + assert "r2_location_for_hash_only_entry" in errors + + +def test_public_entry_needs_no_licence_or_access_route(): + assert ( + validate_file_entry( + {"filename": "table.xlsx", "sha256": FIXTURE_SHA}, + kind="publisher_table", + manifest={}, + local_file_exists=True, + ) + == () + ) + + +def test_list_file_spec_expands_only_for_a_microdata_release(): + specs = [{"filename": "adult.tab"}, {"filename": "child.tab"}] + + assert iter_file_specs(specs, kind="microdata_release") == tuple(specs) + + rejected = iter_file_specs(specs, kind="publisher_table") + assert len(rejected) == 1 + assert isinstance(rejected[0], ListSpecRejected) + assert validate_file_entry( + rejected[0], + kind="publisher_table", + manifest={}, + local_file_exists=False, + ) == ("list_file_spec_requires_microdata_release_kind",) + + +# -------------------------------------------------------------------------- +# register-artifact +# -------------------------------------------------------------------------- + + +def test_register_writes_identity_without_bytes_or_an_r2_key(tmp_path): + output_dir = tmp_path / "dwp" / "frs_2023_24" + + report = _register(output_dir) + + manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) + entry = manifest["files"][2023][0] + + assert report.valid + assert report.registration == (f"dwp/dwp-frs-2023-24/2023/{FIXTURE_SHA}/adult.tab") + assert report.to_dict()["r2_location"] is None + assert manifest["kind"] == "microdata_release" + assert entry["access"] == "licensed" + assert entry["licence"] == "UK Data Service End User Licence" + assert entry["sha256"] == FIXTURE_SHA + assert entry["size_bytes"] == 35323384 + assert entry["vintage"] == "2023_24" + assert entry["verified_at"] == "2026-09-02" + assert "storage" not in entry + # The only file the registration creates is the manifest itself. + assert sorted(p.name for p in output_dir.iterdir()) == ["manifest.yaml"] + + +def test_register_refuses_public_access(tmp_path): + with pytest.raises(HashOnlyRegistrationError, match="refuses access='public'"): + _register(tmp_path / "pkg", access="public") + + +@pytest.mark.parametrize("sha256", ["", "abc123", FIXTURE_SHA.upper(), "z" * 64]) +def test_register_never_invents_a_hash(tmp_path, sha256): + with pytest.raises(HashOnlyRegistrationError, match="Never invent a hash"): + _register(tmp_path / "pkg", sha256=sha256) + assert not (tmp_path / "pkg").exists() + + +@pytest.mark.parametrize( + ("overrides", "expected"), + [ + ({"licence": ""}, "must record the publisher licence"), + ({"vintage": ""}, "must record the artifact vintage"), + ({"verified_at": None}, "must record when the checksum was"), + ({"filename": "../adult.tab"}, "must be a bare filename"), + ], +) +def test_register_refuses_an_incomplete_registration(tmp_path, overrides, expected): + with pytest.raises(HashOnlyRegistrationError, match=expected): + _register(tmp_path / "pkg", **overrides) + + +def test_register_refuses_without_an_access_route(tmp_path): + with pytest.raises(HashOnlyRegistrationError, match="how the bytes are reached"): + _register(tmp_path / "pkg", doi=None) + + +def test_register_refuses_while_the_bytes_are_present(tmp_path): + output_dir = tmp_path / "pkg" + output_dir.mkdir() + (output_dir / "adult.tab").write_bytes(b"licensed microdata must not live here") + + with pytest.raises(HashOnlyRegistrationError, match="while its bytes"): + _register(output_dir) + + +def test_register_refuses_a_publisher_table_manifest(tmp_path): + output_dir = tmp_path / "pkg" + output_dir.mkdir() + (output_dir / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "files": {2023: {"filename": "table.ods", "sha256": FIXTURE_SHA}}, + } + ) + ) + + with pytest.raises(HashOnlyRegistrationError, match="is a publisher_table"): + _register(output_dir) + + +def test_register_refuses_a_manifest_for_a_different_source(tmp_path): + output_dir = tmp_path / "pkg" + _register(output_dir) + + with pytest.raises(HashOnlyRegistrationError, match="declares source_id='dwp'"): + _register(output_dir, source_id="hmrc") + + +def test_register_is_idempotent_and_byte_stable(tmp_path): + output_dir = tmp_path / "pkg" + + _register(output_dir) + first = (output_dir / "manifest.yaml").read_bytes() + report = _register(output_dir) + + assert report.replaced is True + assert (output_dir / "manifest.yaml").read_bytes() == first + + +def test_register_refuses_a_reissue_unless_it_is_asked_for(tmp_path): + output_dir = tmp_path / "pkg" + _register(output_dir) + + with pytest.raises(HashOnlyRegistrationError, match="already registers"): + _register(output_dir, sha256="b" * 64) + + _register(output_dir, sha256="b" * 64, allow_reissue=True) + manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) + entries = manifest["files"][2023] + + # A reissue is a new publisher release, so both registrations survive. + assert [entry["sha256"] for entry in entries] == [FIXTURE_SHA, "b" * 64] + + +def test_register_keeps_distinct_files_under_one_vintage(tmp_path): + output_dir = tmp_path / "pkg" + + _register(output_dir, filename="adult.tab", sha256=FIXTURE_SHA) + _register(output_dir, filename="child.tab", sha256="c" * 64) + + manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) + + assert [entry["filename"] for entry in manifest["files"][2023]] == [ + "adult.tab", + "child.tab", + ] + + +# -------------------------------------------------------------------------- +# fetch-artifact +# -------------------------------------------------------------------------- + + +def test_fetch_refuses_a_hash_only_access_class(tmp_path, monkeypatch): + source = tmp_path / "adult.tab" + source.write_bytes(b"licensed microdata") + + def unexpected_read(_source_url): + raise AssertionError("A refused access class must not read the artifact") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(ManifestAccessError, match="fetch-artifact stores bytes"): + fetch_source_artifact( + str(source), + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=tmp_path / "pkg", + access="licensed", + ) + + +def test_fetch_refuses_to_pull_bytes_over_a_hash_only_registration( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + _register(output_dir) + original_manifest = (output_dir / "manifest.yaml").read_bytes() + source = tmp_path / "adult.tab" + source.write_bytes(b"licensed microdata") + + def unexpected_read(_source_url): + raise AssertionError("A refused fetch must not read the artifact") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + + with pytest.raises(ManifestAccessError, match="Its bytes must not enter"): + fetch_source_artifact( + str(source), + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=output_dir, + filename="adult.tab", + access="public", + ) + + assert not (output_dir / "adult.tab").exists() + assert (output_dir / "manifest.yaml").read_bytes() == original_manifest + + +def test_fetch_writes_the_access_class_explicitly(tmp_path): + source = tmp_path / "table.xlsx" + source.write_bytes(b"publisher table") + output_dir = tmp_path / "pkg" + + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-2", + year=2023, + output_dir=output_dir, + ) + + manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) + + assert manifest["files"][2023]["access"] == "public" + + +def test_fetch_into_a_microdata_release_manifest_requires_a_licence( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + _register(output_dir) + source = tmp_path / "codebook.pdf" + source.write_bytes(b"public codebook") + + monkeypatch.setattr( + "chronicle.artifacts._read_artifact", + lambda _url: (_ for _ in ()).throw( + AssertionError("A refused fetch must not read the artifact") + ), + ) + + with pytest.raises(ManifestAccessError, match="must record its publisher licence"): + fetch_source_artifact( + str(source), + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=output_dir, + filename="codebook.pdf", + ) + + +# -------------------------------------------------------------------------- +# publish-raw +# -------------------------------------------------------------------------- + + +def _hash_only_tree(tmp_path: Path) -> Path: + """Build a data root holding one hash-only registration.""" + root = tmp_path / "data" + _register(root / "dwp" / "frs_2023_24") + return root + + +def test_publish_raw_refuses_hash_only_entries_without_reading_bytes( + tmp_path, monkeypatch +): + root = _hash_only_tree(tmp_path) + + def unexpected_upload(*args, **kwargs): + raise AssertionError("A hash-only registration must never be uploaded") + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", unexpected_upload) + + report = publish_source_artifacts(root) + + assert not report.valid + assert report.counts["hash_only_refused_count"] == 1 + assert report.counts["uploaded_count"] == 0 + entry = report.entries[0] + assert entry.skipped == "hash_only_access:licensed" + assert entry.errors == ("hash_only_access_refuses_bytes:licensed",) + assert entry.r2_location is None + assert entry.upload is None + + +def test_publish_raw_skip_hash_only_reports_the_skip_without_failing( + tmp_path, monkeypatch +): + root = _hash_only_tree(tmp_path) + monkeypatch.setattr( + "chronicle.artifacts._upload_r2_object", + lambda *args, **kwargs: (_ for _ in ()).throw( + AssertionError("A hash-only registration must never be uploaded") + ), + ) + + report = publish_source_artifacts(root, skip_hash_only=True) + + assert report.valid + assert report.counts["hash_only_refused_count"] == 1 + assert report.counts["uploaded_count"] == 0 + assert report.entries[0].errors == () + assert report.entries[0].r2_location is None + + +def test_publish_raw_leaves_the_manifest_untouched(tmp_path, monkeypatch): + root = _hash_only_tree(tmp_path) + manifest_path = root / "dwp" / "frs_2023_24" / "manifest.yaml" + original = manifest_path.read_bytes() + monkeypatch.setattr( + "chronicle.artifacts._upload_r2_object", + lambda *args, **kwargs: (_ for _ in ()).throw(AssertionError("no upload")), + ) + + publish_source_artifacts(root, skip_hash_only=True) + + assert manifest_path.read_bytes() == original + + +# -------------------------------------------------------------------------- +# inventory-artifacts +# -------------------------------------------------------------------------- + + +def test_inventory_accepts_a_hash_only_entry_with_no_local_file(tmp_path): + root = _hash_only_tree(tmp_path) + + report = inventory_source_artifacts(root) + + assert report.valid + assert report.counts["hash_only_count"] == 1 + assert report.counts["missing_count"] == 0 + entry = report.entries[0] + assert entry.valid + assert entry.exists is False + assert entry.hash_only is True + assert entry.access == "licensed" + assert entry.licence == "UK Data Service End User Licence" + assert entry.sha256_expected == FIXTURE_SHA + assert entry.sha256_actual is None + # The size is the publisher's, recorded rather than measured. + assert entry.size_bytes == 35323384 + + +def test_inventory_flags_a_hash_only_entry_whose_bytes_are_present(tmp_path): + root = _hash_only_tree(tmp_path) + (root / "dwp" / "frs_2023_24" / "adult.tab").write_bytes(b"leaked bytes") + + report = inventory_source_artifacts(root) + + assert not report.valid + assert "bytes_present_for_hash_only_entry" in report.entries[0].errors + + +def test_inventory_rejects_a_list_entry_outside_a_microdata_release(tmp_path): + package = tmp_path / "data" / "dwp" / "tables" + package.mkdir(parents=True) + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-tables", + "files": {2023: [{"filename": "a.ods", "sha256": FIXTURE_SHA}]}, + } + ) + ) + + report = inventory_source_artifacts(tmp_path / "data") + + assert not report.valid + assert report.entries[0].errors == ( + "list_file_spec_requires_microdata_release_kind", + ) + + +# -------------------------------------------------------------------------- +# Source packages never parse a microdata release +# -------------------------------------------------------------------------- + + +def test_source_artifact_spec_refuses_to_parse_a_registered_release(): + spec = SourceArtifactSpec( + source_name="dwp", + source_table="Family Resources Survey 2023-24", + resource_package="db", + resource_directory="data/dwp/frs_2023_24", + manifest="manifest.yaml", + vintage="2023_24", + extracted_at="2026-09-02", + extraction_method="none", + parser="delimited_text_full_rows", + artifact_year=2023, + ) + + with pytest.raises(MicrodataReleaseNotParseableError, match="registers a"): + spec.assert_parseable_manifest() + with pytest.raises(MicrodataReleaseNotParseableError): + spec._artifact_content(2023) + + +def test_validate_package_reports_a_microdata_release_carve_out(tmp_path): + package_dir = tmp_path / "frs_2023_24" + package_dir.mkdir() + (package_dir / "source_package.yaml").write_text( + yaml.safe_dump( + { + "schema_version": "ledger.source_package.v1", + "package_id": "dwp-frs-2023-24-parse-attempt", + "label": "Attempt to parse a registered microdata release", + "artifact": { + "source_name": "dwp", + "source_table": "Family Resources Survey 2023-24", + "resource_package": "db", + "resource_directory": "data/dwp/frs_2023_24", + "manifest": "manifest.yaml", + "vintage": "2023_24", + "extracted_at": "2026-09-02", + "extraction_method": "none", + "parser": "delimited_text_full_rows", + "artifact_year": 2023, + }, + "record_sets": [], + }, + sort_keys=False, + ) + ) + + report = validate_source_package(package_dir, year=2023) + + assert not report.valid + assert "microdata_release_not_parseable" in {issue.code for issue in report.errors} + + +# -------------------------------------------------------------------------- +# CLI +# -------------------------------------------------------------------------- + + +def test_cli_register_artifact_round_trips(tmp_path, capsys): + output_dir = tmp_path / "pkg" + + exit_code = harness_main( + [ + "register-artifact", + "--source-id", + "hmrc", + "--package-id", + "hmrc-spi-public-use-tape-2022-23", + "--year", + "2022", + "--out-dir", + str(output_dir), + "--filename", + "put2223uk.tab", + "--sha256", + FIXTURE_SHA, + "--size-bytes", + "141323762", + "--vintage", + "2022-23", + "--licence", + "UK Data Service End User Licence (study SN 9422)", + "--access", + "restricted", + "--doi", + "10.5255/UKDA-SN-9422-1", + "--verified-at", + "2026-09-02", + ] + ) + payload = json.loads(capsys.readouterr().out) + + assert exit_code == 0 + assert payload["valid"] is True + assert payload["access"] == "restricted" + assert payload["r2_location"] is None + assert payload["registration"] == ( + f"hmrc/hmrc-spi-public-use-tape-2022-23/2022/{FIXTURE_SHA}/put2223uk.tab" + ) + assert not (output_dir / "put2223uk.tab").exists() + + +def test_cli_register_artifact_rejects_public_access(tmp_path): + with pytest.raises(SystemExit): + harness_main( + [ + "register-artifact", + "--source-id", + "census_cps", + "--package-id", + "census-cps-asec-2023", + "--year", + "2023", + "--out-dir", + str(tmp_path / "pkg"), + "--filename", + "asecpub23csv.zip", + "--sha256", + FIXTURE_SHA, + "--vintage", + "2023", + "--licence", + "Public domain", + "--access", + "public", + ] + ) + + +# -------------------------------------------------------------------------- +# The committed registrations +# -------------------------------------------------------------------------- + + +def _committed_entries(package: Path) -> list[dict]: + manifest = yaml.safe_load((package / "manifest.yaml").read_text()) + assert manifest["kind"] == "microdata_release" + return [entry for entries in manifest["files"].values() for entry in entries] + + +def test_committed_frs_registration_covers_every_pinned_tab(): + entries = _committed_entries(FRS_PACKAGE) + + assert [entry["filename"] for entry in entries] == [ + "accounts.tab", + "adult.tab", + "benefits.tab", + "benunit.tab", + "child.tab", + "chldcare.tab", + "extchild.tab", + "househol.tab", + "job.tab", + "maint.tab", + "mortgage.tab", + "oddjob.tab", + "penprov.tab", + "pension.tab", + ] + + +@pytest.mark.parametrize("package", [FRS_PACKAGE, SPI_PACKAGE]) +def test_committed_registrations_are_identity_only(package): + entries = _committed_entries(package) + + assert entries + for entry in entries: + assert entry["access"] in {"licensed", "restricted"} + assert entry["licence"] + assert entry["vintage"] + assert entry["verified_at"] + assert len(entry["sha256"]) == 64 + assert int(entry["size_bytes"]) > 0 + assert "storage" not in entry + # No bytes accompany a hash-only registration. + assert not (package / entry["filename"]).exists() + + +@pytest.mark.parametrize("package", [FRS_PACKAGE, SPI_PACKAGE]) +def test_committed_registrations_hold_no_microdata_bytes(package): + assert sorted(path.name for path in package.iterdir()) == ["manifest.yaml"] + + +def test_committed_registrations_pass_inventory(): + report = inventory_source_artifacts(REPO_ROOT / "db" / "data") + hash_only = [entry for entry in report.entries if entry.hash_only] + + assert report.valid + assert len(hash_only) == 15 + assert all(entry.valid and not entry.exists for entry in hash_only) + assert all(entry.r2 is None for entry in hash_only) + + +def test_no_committed_registration_computes_an_r2_key(): + for package in (FRS_PACKAGE, SPI_PACKAGE): + text = (package / "manifest.yaml").read_text() + assert "storage:" not in text + assert "ledger-raw" not in text + assert "r2://" not in text + + +def test_registration_hashes_are_not_hashes_of_anything_chronicle_holds(): + # A registered checksum identifies publisher bytes Chronicle never sees; + # it must never coincide with the hash of the manifest that records it. + for package in (FRS_PACKAGE, SPI_PACKAGE): + manifest_hash = hashlib.sha256( + (package / "manifest.yaml").read_bytes() + ).hexdigest() + assert manifest_hash not in (package / "manifest.yaml").read_text() From 8a4acf5d2102112a48c62666a9bb66caaceb7d39 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:42:51 -0400 Subject: [PATCH 29/42] Document hash-only registrations and the BE-SILC blocker Adds the microdata-release example to the storage doc's Object Key Conventions (public releases keep the ordinary content-addressed key; licensed and restricted ones have no key at all), a Hash-Only Registrations section to the harness doc, a register-artifact entry to the README artifact command block, and a data-sources note recording why BE-SILC 2023 cannot be registered yet. The governance, architecture, AGENTS, and non-goal amendments are left to the ADR branch (PR #222), which already makes them. Co-Authored-By: Claude Fable 5.1 --- README.md | 16 ++++ docs/agent-source-package-harness.md | 58 ++++++++++++++ .../be-silc-2023-registration-blocker.md | 75 +++++++++++++++++++ docs/storage-architecture.md | 14 ++++ 4 files changed, 163 insertions(+) create mode 100644 docs/data-sources/be-silc-2023-registration-blocker.md diff --git a/README.md b/README.md index 7c526a97..08dc05d0 100644 --- a/README.md +++ b/README.md @@ -384,6 +384,22 @@ uv run chronicle inventory-artifacts --root db/data # Upload all existing manifest-declared local artifacts to the raw archive and # write storage.r2 metadata back into the manifests: uv run chronicle publish-raw --root db/data + +# Register a licensed or restricted release by identity alone. No bytes are +# fetched, stored, or uploaded, and no R2 key is recorded: +uv run chronicle register-artifact \ + --source-id dwp \ + --package-id dwp-frs-2023-24 \ + --year 2023 \ + --out-dir db/data/dwp/frs_2023_24 \ + --filename adult.tab \ + --sha256 e09f9647d03585c81a528636028b2ed495f8f1fbcf64c5e7b4fe521b67367e06 \ + --size-bytes 35323384 \ + --vintage 2023_24 \ + --licence "UK Data Service End User Licence" \ + --access licensed \ + --doi 10.5255/UKDA-SN-9367-2 \ + --verified-at 2026-09-02 ``` To coordinate broad PE source migration without jumping straight to semantic diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 56d65bb7..09dfbaed 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -46,6 +46,64 @@ emit a one-time deprecation warning naming the `CHRONICLE_` variable to set instead; see "Environment Variable Rename Window" in [`docs/storage-architecture.md`](storage-architecture.md#environment-variable-rename-window). +## Hash-Only Registrations + +Not every raw artifact a build starts from may be redistributed. Every manifest +file entry carries an `access` class from a closed set — `public`, `licensed`, +or `restricted` — and a `licence` naming the publisher's terms as an identifier +or URL. `public` is inferred when an entry omits `access`, and `fetch-artifact` +now writes the class explicitly onto every entry it touches. Both fields are +required on a `kind: microdata_release` manifest. + +Only `public` bytes enter a Chronicle store. A `licensed` or `restricted` +artifact is registered *hash-only*: the manifest records the checksum, size, +vintage, licence, access route, and verification timestamp, and no bytes are +fetched, written, or uploaded. Agents should register one with: + +```bash +uv run chronicle register-artifact \ + --source-id dwp \ + --package-id dwp-frs-2023-24 \ + --year 2023 \ + --out-dir db/data/dwp/frs_2023_24 \ + --filename adult.tab \ + --sha256 e09f9647d03585c81a528636028b2ed495f8f1fbcf64c5e7b4fe521b67367e06 \ + --size-bytes 35323384 \ + --vintage 2023_24 \ + --licence "UK Data Service End User Licence" \ + --access licensed \ + --doi 10.5255/UKDA-SN-9367-2 \ + --verified-at 2026-09-02 +``` + +Agents should never invent a checksum to satisfy the command: `--sha256` must be +a lowercase 64-character digest taken from a reviewed pin, and a release whose +checksum nobody has published is a blocker to record, not a value to guess. The +command refuses `--access public`, refuses bytes sitting beside the manifest, +and refuses to write into a `publisher_table` manifest. + +The other commands enforce the same boundary from their side. `fetch-artifact` +refuses a `licensed` or `restricted` access class before reading anything, and +refuses to pull bytes over an entry already registered hash-only. `publish-raw` +refuses such an entry without reading or uploading its bytes; pass +`--skip-hash-only` to publish a tree that deliberately mixes both kinds. +`inventory-artifacts` treats a hash-only entry with no local file as valid — the +absent bytes are the correct state — and reports an error if the bytes appear. + +Because several files can share one vintage, a `kind: microdata_release` +manifest may give `files[year]` as a list of entries rather than a single +mapping. A list under any other manifest kind is an error. + +No source package parses a microdata release. `validate-package` fails with +`microdata_release_not_parseable` if a package spec points at one, and no +microdata row, cell, or fact ever enters Chronicle. Registration is +manifest-level identity; see `docs/adr-chronicle-raw-microdata-identity.md`. + +`scripts/register_microdata_releases.py` drives both halves from a read-only +PolicyEngine/microcosm checkout: `emit` writes the hash-only manifests from +Microcosm's reviewed pins, and `plan` prints the `fetch-artifact` commands to +run for public releases from a networked machine. + For broad PE source migration, generate the agent queue from the manifest before assigning work: diff --git a/docs/data-sources/be-silc-2023-registration-blocker.md b/docs/data-sources/be-silc-2023-registration-blocker.md new file mode 100644 index 00000000..56519546 --- /dev/null +++ b/docs/data-sources/be-silc-2023-registration-blocker.md @@ -0,0 +1,75 @@ +# BE-SILC 2023: Registration Blocker + +The Belgian SILC scientific-use files are a `restricted_microdata` root of +Microcosm's Belgian build, so they belong in Chronicle's microdata registry +under `docs/adr-chronicle-raw-microdata-identity.md`. They are **not registered +yet**, and this document records why. + +## Blocker + +A registration is identified by `{source_id, package_id, year, sha256, +filename}`. Microcosm's pin carries none of the last two. The whole artifact +entry in `packages/microcosm-build/src/microcosm/build/be/source_stages.json` +(stage `silc_load`) is: + +```json +{ + "format": "csv_or_spss", + "kind": "restricted_microdata", + "licence": "Statbel/Eurostat scientific-use; restricted — private artifacts only", + "locator": "Statbel BE-SILC scientific-use files: D (household register), R (personal register), H (household data), P (personal data)", + "vintage": "2023" +} +``` + +There is no `sha256`, no `size_bytes`, and no per-file `filename` — the locator +names four file *roles*, not four files. Every other pinned microdata artifact +in Microcosm carries a reviewed checksum; this one does not. + +`chronicle register-artifact` refuses the release rather than accepting a +placeholder, and `scripts/register_microdata_releases.py` reports it as a +blocker instead of emitting a manifest. No checksum is invented for a file +Chronicle has never seen and may never hold. + +## What the registration will record once unblocked + +Everything except the identity is already known and is held in the script's +catalogue entry `statbel-be-silc-2023`: + +| Field | Value | +|-------|-------| +| `source_id` | `statbel` | +| `package_id` | `statbel-be-silc-2023` | +| `year` | 2023 | +| `access` | `restricted` | +| `licence` | Statbel/Eurostat scientific-use | +| `vintage` | 2023 | +| `source_page` | | +| `access_route` | Statbel BE-SILC scientific-use files: D, R, H, P | + +The access class is `restricted`, so the registration is hash-only whatever the +checksums turn out to be: no BE-SILC bytes enter any Chronicle store, and no +`ledger-raw` key exists for them. + +## To unblock + +1. Microcosm publishes a reviewed SHA-256, size, and exact filename for each of + the four scientific-use files, in `be/source_stages.json` (tracked on the + consumer side in PolicyEngine/microcosm#848). +2. Re-run the emitter, which will pick the pins up with no catalogue change: + + ```bash + uv run python scripts/register_microdata_releases.py \ + --microcosm-root ~/PolicyEngine/microcosm \ + --root db/data \ + --release statbel-be-silc-2023 \ + emit --verified-at + ``` + +3. Delete the `blocker` field from the `statbel-be-silc-2023` catalogue entry, + and delete this document. + +Until then `uv run chronicle inventory-artifacts --root db/data` reports 15 +hash-only registrations — the 14 DWP Family Resources Survey 2023-24 tabs and +the HMRC Survey of Personal Incomes Public Use Tape 2022-23 — and BE-SILC is +absent by design. diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index b91d06fa..22cc00d2 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -88,6 +88,20 @@ derived/nz/ird/ird-working-for-families-statistics-sept-2025/2024/{build_id}/chr Legacy US derived keys likewise remain `derived/{source_id}/...`. +A registered microdata release uses the same content-addressed key as any other +raw artifact, and it exists only when the release's `access` class is `public`. +A Census public-use file is a US publisher, so it keeps the legacy shape: + +```text +raw/census_cps/census-cps-asec-2023/2023/{sha256}/asecpub23csv.zip +``` + +A `licensed` or `restricted` release has no key at all. Its registration is +`{source_id, package_id, year, sha256, filename}` recorded in `manifest.yaml`, +and no object is written to either bucket — so the DWP Family Resources Survey +tabs, which would otherwise route to `raw/uk/dwp/...`, occupy no key. See +`docs/adr-chronicle-raw-microdata-identity.md`. + Derived artifacts are reproducible and may be replaced by a new build, but a specific `{build_id}` path should be immutable once published. From cf8b2184d5bfe8cd58a23ad96827c38eedd7fabc Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:49:07 -0400 Subject: [PATCH 30/42] Refuse a multi-file vintage in the source-package artifact path Only a kind: microdata_release manifest may list several files under one vintage. _year_mapping previously returned the list and let the caller raise TypeError on spec["filename"], which validate_source_package does not catch. It now raises a ValueError naming the shape, so a malformed publisher-table manifest becomes a validation error instead of a crash. Co-Authored-By: Claude Fable 5.1 --- chronicle/source_package.py | 20 +++++++++++++++++-- .../test_chronicle_microdata_registration.py | 9 +++++++++ 2 files changed, 27 insertions(+), 2 deletions(-) diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 40f2f028..0fcee5e9 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -2269,12 +2269,28 @@ def _required(payload: dict[str, Any], key: str, context: str) -> Any: def _year_mapping(files_by_year: dict[Any, Any], year: int) -> dict[str, str]: if year in files_by_year: - return files_by_year[year] + return _single_year_spec(files_by_year[year], year) if str(year) in files_by_year: - return files_by_year[str(year)] + return _single_year_spec(files_by_year[str(year)], year) raise ValueError(f"No source artifact for year {year}") +def _single_year_spec(spec: Any, year: int) -> dict[str, str]: + """Return one file spec, refusing the multi-file microdata-release shape. + + Only a ``kind: microdata_release`` manifest may list several files under one + vintage, and no source package parses one of those, so a list here is a + malformed publisher-table manifest rather than something to index into. + """ + if isinstance(spec, list): + raise ValueError( + f"Source artifact for year {year} is a list of {len(spec)} entries. " + "Only a kind: microdata_release manifest may list several files " + "under one vintage, and no source package parses one." + ) + return spec + + def _read_source_artifact_content( artifact_path: Any, spec: dict[str, Any], diff --git a/tests/test_chronicle_microdata_registration.py b/tests/test_chronicle_microdata_registration.py index b80721e9..75f68717 100644 --- a/tests/test_chronicle_microdata_registration.py +++ b/tests/test_chronicle_microdata_registration.py @@ -606,6 +606,15 @@ def test_source_artifact_spec_refuses_to_parse_a_registered_release(): spec._artifact_content(2023) +def test_year_mapping_refuses_a_multi_file_vintage(): + from chronicle.source_package import _year_mapping + + with pytest.raises(ValueError, match="list of 2 entries"): + _year_mapping( + {2023: [{"filename": "adult.tab"}, {"filename": "child.tab"}]}, 2023 + ) + + def test_validate_package_reports_a_microdata_release_carve_out(tmp_path): package_dir = tmp_path / "frs_2023_24" package_dir.mkdir() From 67ca2ce43e6c3b081fe21abf513705bab6716d15 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:54:30 -0400 Subject: [PATCH 31/42] Close six escapes found auditing the registration boundary The byte boundary was escapable and the fetch plan did not run: - _assert_no_hash_only_entry only inspected files[year] for the requested year and only entries that survived the kind-gated list expansion, while the write target and R2 key are directory-scoped. A licensed release registered under one vintage could be fetched into the tree under another, and a list under a manifest with an absent or misspelled kind was invisible to the guard. Both wrote licensed bytes into the package tree and would have uploaded them with --upload-r2. The guard now walks every vintage and expands lists unconditionally; validating the shape stays the validator's job. - _upsert_manifest replaced the whole files[year] value, so fetching a different filename into a package silently deleted an existing hash-only registration. It now replaces only the entry for its own filename. - register_hash_only_artifact raised on the first filename match with a different checksum, so re-registering a current pin failed once a reissue existed and `emit` stopped being re-runnable. Now two passes: an exact (filename, sha256) match is replaced wherever it sits. - publish-raw returned before validating a hash-only entry, so --skip-hash-only reported bytes on disk or a recorded R2 key as valid, disagreeing with inventory-artifacts about the same tree. The entry is now always validated; the flag turns off the refusal, not the contract check. - Nothing could register a public microdata release as kind: microdata_release, though the ADR puts redistributable public-use files in scope. fetch-artifact gains --kind, and a microdata-release manifest keeps the list shape so several files can share one vintage. - Every command `plan` printed used --source-url, which fetch-artifact does not accept; all nine died at argparse. They now use --url and pass --kind, and a test parses each generated command against the real CLI parser. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 101 +++++-- chronicle/harness.py | 14 + chronicle/registration.py | 19 +- scripts/register_microdata_releases.py | 7 +- .../test_chronicle_microdata_registration.py | 279 ++++++++++++++++++ 5 files changed, 389 insertions(+), 31 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 893ff6e5..6e450821 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -5,6 +5,7 @@ import hashlib import json import mimetypes +from collections.abc import Iterator from dataclasses import dataclass from datetime import UTC, datetime from pathlib import Path @@ -29,6 +30,7 @@ from chronicle.registration import ( ACCESS_PUBLIC, MICRODATA_RELEASE_KIND, + manifest_kind as normalize_manifest_kind, ListSpecRejected, ManifestAccessError, is_hash_only, @@ -653,6 +655,7 @@ def fetch_source_artifact( manifest_filename: str = DEFAULT_MANIFEST_FILENAME, access: str = ACCESS_PUBLIC, licence: str | None = None, + kind: str | None = None, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -706,14 +709,19 @@ def fetch_source_artifact( year=year, ) licence_text = licence.strip() if isinstance(licence, str) else None + # A public microdata release is archived like any other public artifact, + # but its manifest still declares the release kind so a source package is + # refused and several files may share one vintage. + manifest_kind_value = ( + normalize_manifest_kind({"kind": kind}) + if kind is not None + else safe_manifest_kind(existing_manifest)[0] + ) # The byte boundary is checked first: overwriting a hash-only registration # with bytes is the more serious refusal, and its message is the one the # caller needs, not a prompt to supply a licence. _assert_no_hash_only_entry(existing_manifest, manifest_path, year, filename) - if ( - safe_manifest_kind(existing_manifest)[0] == MICRODATA_RELEASE_KIND - and not licence_text - ): + if manifest_kind_value == MICRODATA_RELEASE_KIND and not licence_text: raise ManifestAccessError( f"{manifest_path} registers a microdata release, so every entry " "must record its publisher licence; pass --licence." @@ -790,6 +798,7 @@ def fetch_source_artifact( fetched_at=fetched_at, access=access_class, licence=licence_text, + kind=manifest_kind_value, r2_location=(r2_location if upload_r2 and r2_upload and r2_upload.ok else None), record_revision=record_revision, ) @@ -1778,6 +1787,7 @@ def _upsert_manifest( fetched_at: str, access: str, licence: str | None, + kind: str, r2_location: ArtifactStorageLocation | None, record_revision: bool = False, ) -> None: @@ -1807,6 +1817,8 @@ def _upsert_manifest( "fetched_at": fetched_at, } ) + if kind == MICRODATA_RELEASE_KIND: + payload["kind"] = kind existing = ( payload["files"].get(year) if isinstance(payload["files"], dict) else None ) @@ -1851,21 +1863,33 @@ def _upsert_manifest( # revision over a never-published entry supersedes nothing. if storage: file_entry["storage"] = storage - if isinstance(existing, list): - replaced = [ - entry - for entry in existing - if not (isinstance(entry, dict) and entry.get("filename") == filename) - ] - payload["files"][year] = [*replaced, file_entry] - else: + entries = list(existing) if isinstance(existing, list) else _as_entry_list(existing) + # A registration is a durable statement, so a fetch replaces only the entry + # for its own filename and never silently drops another one. Overwriting a + # hash-only registration is refused outright by _assert_no_hash_only_entry. + kept = [ + entry + for entry in entries + if not (isinstance(entry, dict) and entry.get("filename") == filename) + ] + if not kept and not isinstance(existing, list) and kind != MICRODATA_RELEASE_KIND: + # A single-file publisher table keeps the historical mapping shape. payload["files"][year] = file_entry + else: + payload["files"][year] = [*kept, file_entry] manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False, allow_unicode=True), encoding="utf-8", ) +def _as_entry_list(existing: Any) -> list[Any]: + """Return an existing ``files[year]`` value as a list of entries.""" + if existing is None: + return [] + return [existing] + + def _load_manifest_payload(manifest_path: Path) -> dict[str, Any]: """Load a manifest mapping, or an empty mapping when absent or unreadable.""" if not manifest_path.exists(): @@ -1876,20 +1900,37 @@ def _load_manifest_payload(manifest_path: Path) -> dict[str, Any]: return payload +def _iter_manifest_entries(manifest: dict[str, Any]) -> Iterator[tuple[Any, Any]]: + """Yield every ``(year, entry)`` a manifest declares, whatever its shape. + + Deliberately not gated on the manifest ``kind``: a guard must see the + entries a manifest actually holds, including a list under a manifest whose + kind is absent or misspelled. Reporting that shape as malformed is the + validator's job, not the guard's. + """ + files = manifest.get("files") + if not isinstance(files, dict): + return + for year, spec in files.items(): + for entry in spec if isinstance(spec, list) else (spec,): + yield year, entry + + def _assert_no_hash_only_entry( manifest: dict[str, Any], manifest_path: Path, year: Any, filename: str | None, ) -> None: - """Refuse to fetch bytes over an existing hash-only registration.""" + """Refuse to fetch bytes over an existing hash-only registration. + + The write target is a path in the package directory, so the search spans + every vintage rather than the requested one: a licensed release registered + under one year must not be fetched into the tree under another. + """ if not filename: return - files = manifest.get("files") - if not isinstance(files, dict): - return - kind, _kind_error = safe_manifest_kind(manifest) - for spec in iter_file_specs(files.get(year), kind=kind): + for entry_year, spec in _iter_manifest_entries(manifest): if not isinstance(spec, dict): continue if spec.get("filename") != filename: @@ -1897,7 +1938,7 @@ def _assert_no_hash_only_entry( access = safe_entry_access(spec) if is_hash_only(access): raise ManifestAccessError( - f"{manifest_path} registers {filename!r} for {year} as " + f"{manifest_path} registers {filename!r} for {entry_year} as " f"access={access!r}. Its bytes must not enter a Chronicle " "store; keep the hash-only registration." ) @@ -1964,7 +2005,21 @@ def _publish_raw_manifest_entry( access = safe_entry_access(spec) if is_hash_only(access): # Refuse before touching bytes: no Chronicle store holds a licensed or - # restricted artifact, so there is nothing here to upload. + # restricted artifact, so there is nothing here to upload. The entry is + # still validated, because bytes on disk or a recorded R2 key are + # contract violations that --skip-hash-only must not hide. + hash_only_errors = list( + validate_file_entry( + spec, + kind=kind or safe_manifest_kind(manifest)[0], + manifest=manifest, + local_file_exists=(manifest_path.parent / filename).exists() + if filename + else False, + ) + ) + if not skip_hash_only: + hash_only_errors.insert(0, f"hash_only_access_refuses_bytes:{access}") return ( RawArtifactPublishEntry( manifest_path=str(manifest_path), @@ -1977,11 +2032,7 @@ def _publish_raw_manifest_entry( size_bytes=spec.get("size_bytes"), r2_location=None, upload=None, - errors=( - () - if skip_hash_only - else (f"hash_only_access_refuses_bytes:{access}",) - ), + errors=tuple(dict.fromkeys(hash_only_errors)), skipped=f"{HASH_ONLY_SKIP_PREFIX}{access}", ), None, diff --git a/chronicle/harness.py b/chronicle/harness.py index acbbc4dc..8ce68471 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -53,6 +53,7 @@ ) from chronicle.registration import ( ACCESS_CLASSES, + MANIFEST_KINDS, ArtifactRegistrationReport, register_hash_only_artifact, ) @@ -348,6 +349,7 @@ def fetch_artifact_file( manifest_filename: str = DEFAULT_MANIFEST_FILENAME, access: str = "public", licence: str | None = None, + kind: str | None = None, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -374,6 +376,7 @@ def fetch_artifact_file( manifest_filename=manifest_filename, access=access, licence=licence, + kind=kind, upload_r2=upload_r2, record_revision=record_revision, r2_bucket=r2_bucket, @@ -993,6 +996,16 @@ def main(argv: list[str] | None = None) -> int: "microdata-release manifest." ), ) + artifact_parser.add_argument( + "--kind", + choices=list(MANIFEST_KINDS), + help=( + "Manifest kind to declare. Pass microdata_release to archive a " + "public-use microdata release, which may hold several files under " + "one vintage and is never parsed by a source package. Defaults to " + "the existing manifest's kind, or publisher_table." + ), + ) artifact_parser.add_argument( "--upload-r2", action="store_true", @@ -1589,6 +1602,7 @@ def main(argv: list[str] | None = None) -> int: manifest_filename=args.manifest, access=args.access, licence=args.licence, + kind=args.kind, upload_r2=args.upload_r2, record_revision=args.record_revision, r2_bucket=args.r2_bucket, diff --git a/chronicle/registration.py b/chronicle/registration.py index 6044af00..0194a831 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -458,6 +458,10 @@ def register_hash_only_artifact( payload.setdefault("files", {}) entries = _existing_entries(payload["files"], year) + # Two passes, so re-registering an existing pin stays idempotent even after + # a reissue has added a second entry for the same filename. A single pass + # would raise on the first filename match with a different checksum before + # it could reach the exact match further down the list. replaced = False for index, existing in enumerate(entries): if not isinstance(existing, Mapping): @@ -468,16 +472,21 @@ def register_hash_only_artifact( entries[index] = entry replaced = True break - if not allow_reissue: + if not replaced: + superseded = [ + existing + for existing in entries + if isinstance(existing, Mapping) + and _text(existing.get("filename")) == artifact_name + ] + if superseded and not allow_reissue: raise HashOnlyRegistrationError( f"{manifest_path} already registers {artifact_name!r} for " - f"{year} with sha256={existing.get('sha256')!r}. Different " + f"{year} with sha256={superseded[0].get('sha256')!r}. Different " "bytes are a new publisher release, not a pin replacement; " "pass --allow-reissue to register both." ) - else: - # No identical (filename, sha256) entry: this is a new registration, - # which for a reissue sits alongside the pin it supersedes. + # A reissue sits alongside the pin it supersedes. entries.append(entry) payload["files"][year] = sorted(entries, key=_entry_sort_key) diff --git a/scripts/register_microdata_releases.py b/scripts/register_microdata_releases.py index b93343a4..6384878a 100644 --- a/scripts/register_microdata_releases.py +++ b/scripts/register_microdata_releases.py @@ -49,6 +49,7 @@ from chronicle.registration import ( # noqa: E402 ACCESS_PUBLIC, + MICRODATA_RELEASE_KIND, HashOnlyRegistrationError, register_hash_only_artifact, ) @@ -685,7 +686,7 @@ def fetch_command( str(release.year), "--out-dir", str(root / release.package_dir), - "--source-url", + "--url", url, ] if release.source_page: @@ -697,6 +698,10 @@ def fetch_command( ACCESS_PUBLIC, "--licence", release.licence, + # A public microdata release is archived, but it is still a release: + # several files share one vintage and no source package parses it. + "--kind", + MICRODATA_RELEASE_KIND, "--upload-r2", "--r2-bucket", r2_bucket, diff --git a/tests/test_chronicle_microdata_registration.py b/tests/test_chronicle_microdata_registration.py index 75f68717..f608b1bc 100644 --- a/tests/test_chronicle_microdata_registration.py +++ b/tests/test_chronicle_microdata_registration.py @@ -47,6 +47,8 @@ REPO_ROOT = Path(__file__).resolve().parents[1] FRS_PACKAGE = REPO_ROOT / "db" / "data" / "dwp" / "frs_2023_24" SPI_PACKAGE = REPO_ROOT / "db" / "data" / "hmrc" / "spi_public_use_tape_2022_23" +# Read-only consumer checkout the emit/plan script resolves pins against. +MICROCOSM_ROOT = Path.home() / "PolicyEngine" / "populace" # A syntactically valid checksum that identifies no real publisher bytes. FIXTURE_SHA = "a" * 64 @@ -724,6 +726,283 @@ def test_cli_register_artifact_rejects_public_access(tmp_path): ) +# -------------------------------------------------------------------------- +# Regressions: the byte boundary must not be escapable +# -------------------------------------------------------------------------- + + +def _licensed_manifest(package: Path, payload: dict) -> Path: + package.mkdir(parents=True, exist_ok=True) + (package / "manifest.yaml").write_text(yaml.safe_dump(payload)) + return package + + +def test_fetch_refuses_a_registration_recorded_under_another_vintage(tmp_path): + # The write target is a path in the package directory, not a year, so a + # registration under 2022 must still block a fetch requested for 2023. + package = _licensed_manifest( + tmp_path / "pkg", + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "microdata_release", + "files": { + 2022: [ + { + "filename": "adult.tab", + "access": "licensed", + "licence": "UK Data Service End User Licence", + "vintage": "2022_23", + "sha256": FIXTURE_SHA, + "doi": "10.5255/UKDA-SN-9367-2", + "verified_at": "2026-09-02", + } + ] + }, + }, + ) + source = tmp_path / "adult.tab" + source.write_bytes(b"licensed microdata") + + with pytest.raises(ManifestAccessError, match="for 2022 as access='licensed'"): + fetch_source_artifact( + str(source), + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=package, + filename="adult.tab", + licence="UK Data Service End User Licence", + access="public", + ) + + assert not (package / "adult.tab").exists() + + +def test_fetch_refuses_a_list_entry_in_a_manifest_without_a_kind(tmp_path): + # A missing or misspelled kind must not make the guard blind to the + # entries the manifest actually holds. + package = _licensed_manifest( + tmp_path / "pkg", + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "files": { + 2023: [ + { + "filename": "adult.tab", + "access": "licensed", + "licence": "UK Data Service End User Licence", + "sha256": FIXTURE_SHA, + } + ] + }, + }, + ) + source = tmp_path / "adult.tab" + source.write_bytes(b"licensed microdata") + + with pytest.raises(ManifestAccessError, match="Its bytes must not enter"): + fetch_source_artifact( + str(source), + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=package, + filename="adult.tab", + access="public", + ) + + assert not (package / "adult.tab").exists() + + +def test_fetch_never_drops_an_existing_registration(tmp_path): + output_dir = tmp_path / "pkg" + _register(output_dir) + source = tmp_path / "other.csv" + source.write_bytes(b"a public table in the same package") + + fetch_source_artifact( + str(source), + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=output_dir, + filename="other.csv", + licence="Open Government Licence", + access="public", + ) + + manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) + entries = manifest["files"][2023] + + assert [entry["filename"] for entry in entries] == ["adult.tab", "other.csv"] + assert entries[0]["access"] == "licensed" + assert entries[0]["sha256"] == FIXTURE_SHA + + +def test_publish_raw_reports_a_violation_even_when_skipping(tmp_path, monkeypatch): + root = _hash_only_tree(tmp_path) + (root / "dwp" / "frs_2023_24" / "adult.tab").write_bytes(b"leaked bytes") + monkeypatch.setattr( + "chronicle.artifacts._upload_r2_object", + lambda *args, **kwargs: (_ for _ in ()).throw(AssertionError("no upload")), + ) + + report = publish_source_artifacts(root, skip_hash_only=True) + + # --skip-hash-only turns off the refusal, not the contract check. + assert not report.valid + assert "bytes_present_for_hash_only_entry" in report.entries[0].errors + + +def test_reregistering_the_current_pin_stays_idempotent_after_a_reissue(tmp_path): + output_dir = tmp_path / "pkg" + _register(output_dir, sha256=FIXTURE_SHA) + _register(output_dir, sha256="b" * 64, allow_reissue=True) + + report = _register(output_dir, sha256="b" * 64) + + assert report.replaced is True + entries = yaml.safe_load((output_dir / "manifest.yaml").read_text())["files"][2023] + assert [entry["sha256"] for entry in entries] == [FIXTURE_SHA, "b" * 64] + + +# -------------------------------------------------------------------------- +# Public microdata releases +# -------------------------------------------------------------------------- + + +def test_fetch_can_declare_a_public_microdata_release(tmp_path): + output_dir = tmp_path / "pkg" + household = tmp_path / "csv_hus.zip" + household.write_bytes(b"public household pums") + person = tmp_path / "csv_pus.zip" + person.write_bytes(b"public person pums") + + for source in (household, person): + fetch_source_artifact( + str(source), + source_id="census_acs", + package_id="census-acs-pums-2022-1yr", + year=2022, + output_dir=output_dir, + licence="U.S. Census Bureau public-use file", + access="public", + kind="microdata_release", + ) + + manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) + entries = manifest["files"][2022] + + # A release may hold several files under one vintage, and both are kept. + assert manifest["kind"] == "microdata_release" + assert [entry["filename"] for entry in entries] == ["csv_hus.zip", "csv_pus.zip"] + assert all(entry["access"] == "public" for entry in entries) + assert all(entry["licence"] for entry in entries) + + report = inventory_source_artifacts(output_dir) + assert report.valid + assert report.counts["hash_only_count"] == 0 + + +def test_fetch_rejects_an_unknown_manifest_kind(tmp_path): + source = tmp_path / "table.xlsx" + source.write_bytes(b"publisher table") + + with pytest.raises(ManifestAccessError, match="Unknown manifest kind"): + fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-2", + year=2023, + output_dir=tmp_path / "pkg", + kind="microdata_rows", + ) + + +# -------------------------------------------------------------------------- +# The generated fetch plan +# -------------------------------------------------------------------------- + + +def test_planned_fetch_commands_parse_against_the_real_cli(): + """Every command `plan` prints must be runnable as printed.""" + import argparse + import contextlib + import shlex + import subprocess + import sys + + result = subprocess.run( + [ + sys.executable, + str(REPO_ROOT / "scripts" / "register_microdata_releases.py"), + "--microcosm-root", + str(MICROCOSM_ROOT), + "plan", + ], + capture_output=True, + text=True, + cwd=REPO_ROOT, + ) + if result.returncode != 0: + pytest.skip(f"No readable microcosm checkout: {result.stderr.strip()[:120]}") + + commands = [ + line + for line in result.stdout.splitlines() + if line.startswith("uv run chronicle") + ] + assert commands + + captured: dict[str, argparse.ArgumentParser] = {} + real_parse = argparse.ArgumentParser.parse_args + + def capture(self, args=None, namespace=None): + captured["parser"] = self + raise SystemExit(0) + + argparse.ArgumentParser.parse_args = capture + try: + with contextlib.suppress(SystemExit): + harness_main(["--help"]) + finally: + argparse.ArgumentParser.parse_args = real_parse + + parser = captured["parser"] + for command in commands: + argv = shlex.split(command)[3:] + namespace = parser.parse_args(argv) + assert namespace.command == "fetch-artifact" + assert namespace.access == "public" + assert namespace.licence + assert namespace.kind == "microdata_release" + + +def test_plan_never_prints_a_fabricated_url(): + """A release Microcosm does not pin a URL for prints TODO, not a guess.""" + import subprocess + import sys + + result = subprocess.run( + [ + sys.executable, + str(REPO_ROOT / "scripts" / "register_microdata_releases.py"), + "--microcosm-root", + str(MICROCOSM_ROOT), + "plan", + ], + capture_output=True, + text=True, + cwd=REPO_ROOT, + ) + if result.returncode != 0: + pytest.skip("No readable microcosm checkout") + + assert "TODO_PUBLISHER_URL" in result.stdout + + # -------------------------------------------------------------------------- # The committed registrations # -------------------------------------------------------------------------- From c9f53e5061cd2754036abf999dd0fe67b367104a Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 11:55:02 -0400 Subject: [PATCH 32/42] Document --kind for public microdata releases Co-Authored-By: Claude Fable 5.1 --- docs/agent-source-package-harness.md | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-) diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 09dfbaed..09422f5f 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -90,9 +90,29 @@ refuses such an entry without reading or uploading its bytes; pass `inventory-artifacts` treats a hash-only entry with no local file as valid — the absent bytes are the correct state — and reports an error if the bytes appear. +A `public` microdata release is different: its bytes are redistributable, so it +is acquired with `fetch-artifact` like any other public artifact, and it does +get an R2 key. Pass `--kind microdata_release` so the manifest still declares +what it is: + +```bash +uv run chronicle fetch-artifact \ + --url https://www2.census.gov/programs-surveys/acs/data/pums/2022/1-Year/csv_hus.zip \ + --source-id census_acs \ + --package-id census-acs-pums-2022-1yr \ + --year 2022 \ + --out-dir db/data/census/acs_pums_2022_1yr \ + --access public \ + --licence "U.S. Census Bureau public-use file" \ + --kind microdata_release \ + --upload-r2 +``` + Because several files can share one vintage, a `kind: microdata_release` manifest may give `files[year]` as a list of entries rather than a single -mapping. A list under any other manifest kind is an error. +mapping — the ACS household and person files above land side by side. A list +under any other manifest kind is an error. A fetch replaces only the entry for +its own filename, so acquiring a second file never drops the first. No source package parses a microdata release. `validate-package` fails with `microdata_release_not_parseable` if a package spec points at one, and no From 1ea2a561584616bb9a18575489fa40f74fa0f9aa Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Wed, 2 Sep 2026 12:20:50 -0400 Subject: [PATCH 33/42] Match house doc style in the new sections Title Case headings and prose hard-wrapped at 78 characters, matching every other heading and paragraph in these files. Only lines this branch added are touched; the diff stays a pure insertion. Co-Authored-By: Claude Fable 5.1 --- docs/agent-source-package-harness.md | 43 ++++++++++--------- .../be-silc-2023-registration-blocker.md | 20 ++++----- docs/storage-architecture.md | 7 +-- 3 files changed, 36 insertions(+), 34 deletions(-) diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 09422f5f..2cdd1579 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -48,12 +48,12 @@ instead; see "Environment Variable Rename Window" in ## Hash-Only Registrations -Not every raw artifact a build starts from may be redistributed. Every manifest -file entry carries an `access` class from a closed set — `public`, `licensed`, -or `restricted` — and a `licence` naming the publisher's terms as an identifier -or URL. `public` is inferred when an entry omits `access`, and `fetch-artifact` -now writes the class explicitly onto every entry it touches. Both fields are -required on a `kind: microdata_release` manifest. +Not every raw artifact a build starts from may be redistributed. Every +manifest file entry carries an `access` class from a closed set — `public`, +`licensed`, or `restricted` — and a `licence` naming the publisher's terms as +an identifier or URL. `public` is inferred when an entry omits `access`, and +`fetch-artifact` now writes the class explicitly onto every entry it touches. +Both fields are required on a `kind: microdata_release` manifest. Only `public` bytes enter a Chronicle store. A `licensed` or `restricted` artifact is registered *hash-only*: the manifest records the checksum, size, @@ -76,24 +76,25 @@ uv run chronicle register-artifact \ --verified-at 2026-09-02 ``` -Agents should never invent a checksum to satisfy the command: `--sha256` must be -a lowercase 64-character digest taken from a reviewed pin, and a release whose -checksum nobody has published is a blocker to record, not a value to guess. The -command refuses `--access public`, refuses bytes sitting beside the manifest, -and refuses to write into a `publisher_table` manifest. +Agents should never invent a checksum to satisfy the command: `--sha256` must +be a lowercase 64-character digest taken from a reviewed pin, and a release +whose checksum nobody has published is a blocker to record, not a value to +guess. The command refuses `--access public`, refuses bytes sitting beside the +manifest, and refuses to write into a `publisher_table` manifest. The other commands enforce the same boundary from their side. `fetch-artifact` refuses a `licensed` or `restricted` access class before reading anything, and -refuses to pull bytes over an entry already registered hash-only. `publish-raw` -refuses such an entry without reading or uploading its bytes; pass -`--skip-hash-only` to publish a tree that deliberately mixes both kinds. -`inventory-artifacts` treats a hash-only entry with no local file as valid — the -absent bytes are the correct state — and reports an error if the bytes appear. - -A `public` microdata release is different: its bytes are redistributable, so it -is acquired with `fetch-artifact` like any other public artifact, and it does -get an R2 key. Pass `--kind microdata_release` so the manifest still declares -what it is: +refuses to pull bytes over an entry already registered hash-only. +`publish-raw` refuses such an entry without reading or uploading its bytes; +pass `--skip-hash-only` to publish a tree that deliberately mixes both kinds. +`inventory-artifacts` treats a hash-only entry with no local file as valid — +the absent bytes are the correct state — and reports an error if the bytes +appear. + +A `public` microdata release is different: its bytes are redistributable, so +it is acquired with `fetch-artifact` like any other public artifact, and it +does get an R2 key. Pass `--kind microdata_release` so the manifest still +declares what it is: ```bash uv run chronicle fetch-artifact \ diff --git a/docs/data-sources/be-silc-2023-registration-blocker.md b/docs/data-sources/be-silc-2023-registration-blocker.md index 56519546..0353713b 100644 --- a/docs/data-sources/be-silc-2023-registration-blocker.md +++ b/docs/data-sources/be-silc-2023-registration-blocker.md @@ -2,8 +2,8 @@ The Belgian SILC scientific-use files are a `restricted_microdata` root of Microcosm's Belgian build, so they belong in Chronicle's microdata registry -under `docs/adr-chronicle-raw-microdata-identity.md`. They are **not registered -yet**, and this document records why. +under `docs/adr-chronicle-raw-microdata-identity.md`. They are **not +registered yet**, and this document records why. ## Blocker @@ -22,16 +22,16 @@ entry in `packages/microcosm-build/src/microcosm/build/be/source_stages.json` } ``` -There is no `sha256`, no `size_bytes`, and no per-file `filename` — the locator -names four file *roles*, not four files. Every other pinned microdata artifact -in Microcosm carries a reviewed checksum; this one does not. +There is no `sha256`, no `size_bytes`, and no per-file `filename` — the +locator names four file *roles*, not four files. Every other pinned microdata +artifact in Microcosm carries a reviewed checksum; this one does not. `chronicle register-artifact` refuses the release rather than accepting a placeholder, and `scripts/register_microdata_releases.py` reports it as a blocker instead of emitting a manifest. No checksum is invented for a file Chronicle has never seen and may never hold. -## What the registration will record once unblocked +## What The Registration Will Record Once Unblocked Everything except the identity is already known and is held in the script's catalogue entry `statbel-be-silc-2023`: @@ -47,11 +47,11 @@ catalogue entry `statbel-be-silc-2023`: | `source_page` | | | `access_route` | Statbel BE-SILC scientific-use files: D, R, H, P | -The access class is `restricted`, so the registration is hash-only whatever the -checksums turn out to be: no BE-SILC bytes enter any Chronicle store, and no -`ledger-raw` key exists for them. +The access class is `restricted`, so the registration is hash-only whatever +the checksums turn out to be: no BE-SILC bytes enter any Chronicle store, and +no `ledger-raw` key exists for them. -## To unblock +## To Unblock 1. Microcosm publishes a reviewed SHA-256, size, and exact filename for each of the four scientific-use files, in `be/source_stages.json` (tracked on the diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index 22cc00d2..07f50420 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -88,9 +88,10 @@ derived/nz/ird/ird-working-for-families-statistics-sept-2025/2024/{build_id}/chr Legacy US derived keys likewise remain `derived/{source_id}/...`. -A registered microdata release uses the same content-addressed key as any other -raw artifact, and it exists only when the release's `access` class is `public`. -A Census public-use file is a US publisher, so it keeps the legacy shape: +A registered microdata release uses the same content-addressed key as any +other raw artifact, and it exists only when the release's `access` class is +`public`. A Census public-use file is a US publisher, so it keeps the legacy +shape: ```text raw/census_cps/census-cps-asec-2023/2023/{sha256}/asecpub23csv.zip From 17e91fcb26f67d2439953cf65742b84687c4ca70 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 01:59:13 -0400 Subject: [PATCH 34/42] Make classification, redistribution evidence and attestation explicit The ADR on PR #222 says every manifest declares its kind, that bytes are archived only under an allowlisted licence with artifact-bound evidence, and that each registration names who asserts its checksum. None of that was in the registration vocabulary: - manifest_kind() no longer defaults a manifest with content to publisher_table. Manifests that predate the rule are frozen byte for byte in chronicle/grandfathered_manifests.py; any other kindless manifest is a ManifestKindError. - chronicle/licences.py is the allowlist of redistributable terms (a U.S. Government work, OGL v3, CC0, CC BY 4.0), each with its evidence URL, and the validator for a licence_evidence block binding one artifact to one term. - hash_source is the closed set chronicle_fetch / consumer_attested / consumer_pin, each with its attester fields; register_hash_only_artifact requires one and refuses a consumer_pin with a verified_at. - Filenames are bare names compared case-folded (bare_filename, filename_key), 2023 and '2023' are one vintage key (resolve_vintage_key), and validate_manifest_files reports duplicate keys, non-canonical names and filename collisions across a whole manifest. - register_hash_only_artifact refuses to reclassify an entry that is public or records an R2 object, resolves the vintage key the manifest already uses, and validates the manifest before touching it. Co-Authored-By: Claude Fable 5.1 --- chronicle/grandfathered_manifests.py | 554 +++++++++++++++++++++ chronicle/licences.py | 137 ++++++ chronicle/registration.py | 699 +++++++++++++++++++++++++-- 3 files changed, 1338 insertions(+), 52 deletions(-) create mode 100644 chronicle/grandfathered_manifests.py create mode 100644 chronicle/licences.py diff --git a/chronicle/grandfathered_manifests.py b/chronicle/grandfathered_manifests.py new file mode 100644 index 00000000..1e9de9ee --- /dev/null +++ b/chronicle/grandfathered_manifests.py @@ -0,0 +1,554 @@ +"""Manifests that predate the explicit-kind rule, frozen at their pre-rule bytes. + +Every manifest created or modified after ``docs/adr-chronicle-raw-microdata- +identity.md`` declares ``kind``: ``publisher_table`` or ``microdata_release``. +The manifests listed here existed before that rule and declare none. They read +as ``publisher_table`` only while their bytes still match the digest frozen +here: a grandfathered manifest that is modified in any way -- by +``fetch-artifact``, which always writes ``kind``, or by hand -- leaves the +freeze and must declare its kind. A kindless manifest that is not on this list +is an error, never a publisher table by default. + +The list is frozen at the freeze commit: entries are removed once a manifest +declares its kind, and never added. ``tests/test_chronicle_manifest_kind.py`` +checks that every kindless manifest in the tree is listed here with its frozen +digest, so a new kindless manifest cannot land. +""" + +from __future__ import annotations + +import hashlib +from collections.abc import Mapping +from pathlib import PurePosixPath +from types import MappingProxyType +from typing import Any + +__all__ = [ + "GRANDFATHERED_KINDLESS_MANIFESTS", + "grandfathered_manifest_key", + "is_grandfathered_manifest", + "manifest_digest", +] + +#: Repository-relative manifest path -> SHA-256 of the manifest bytes at the +#: freeze. +GRANDFATHERED_KINDLESS_MANIFESTS: Mapping[str, str] = MappingProxyType( + { + "db/data/bea/nipa_total_wages_salaries/manifest.yaml": ( + "765f391487a698506647a2bb60823f6742046abae0b8bed62874c6f98bd50d93" + ), + "db/data/bea/regional_personal_income_state/manifest.yaml": ( + "a1283e1ccc574fbeb9710abd409d41c4b735b75c674a34bcc58e6df755278e56" + ), + "db/data/bfp/economic_outlook_2026_06/manifest.yaml": ( + "0a46d70e52cb8ea3a51af5e104e378e257ee1bc8e7c03aabf5a2286b09024e7c" + ), + "db/data/cbo/individual_income_tax_receipts_2026_02/manifest.yaml": ( + "c24431df9284fef70334e51b0d9eaa03a47553f777de0f78d6430db6d7e79954" + ), + "db/data/cbo/revenue_projections_income_by_source_2026_02/manifest.yaml": ( + "acbd1ff5577b638cea9dcbceba6a3fce7911169ffce661410b4e72adcc0d5b76" + ), + "db/data/census/acs_s0101_district_2024/manifest.yaml": ( + "41f1df4f9e5377993ef8bc6be27f26a690d6d9bee2bb73a4c0b5212bd9e9fdb1" + ), + "db/data/census/acs_s0101_national_2024/manifest.yaml": ( + "a5bce23e75ec58ca86247a52ac03ff793924a66513b1c8f04cec45da001e1037" + ), + "db/data/census/acs_s0101_state_2024/manifest.yaml": ( + "d38f64b5ad62b05ed6c3e3c6b4d803f03a4543611d0a5f9bb4459f17785713e7" + ), + "db/data/census/acs_s2201_district_2024/manifest.yaml": ( + "ce5062818e45345a95a5d87da5415f426a3e114b128cab5c62cd00f8f0c999f6" + ), + "db/data/census/b01001_female_15_44_2023/manifest.yaml": ( + "d5bf52beefdeea84998391c6ea2905b6fd89bcd46dcbd9e7e7d539183d45f8d4" + ), + "db/data/census/pep_2024_age_sex/manifest_national_source_package.yaml": ( + "4392052b88202e9799712b46fa291ec3f396c3b0c84f25e1f6f8bbc6029b4678" + ), + "db/data/census/pep_2024_age_sex/manifest_state_source_package.yaml": ( + "2e0e173301250dfd7b339b2d6a6b4a367c1536c42a84567acb705dfa98a6bb5f" + ), + "db/data/census/pep_county_2024/manifest.yaml": ( + "7622c3dbaab7a1eafce5682c072cce9a79d0d60edc80936ce0f76b127c8f9d55" + ), + "db/data/census/population_projections_2023/manifest.yaml": ( + "fbc68ea1bd58126e5122de51ce1470eaec1001b69328fb959a1d84d859d71ca1" + ), + "db/data/census/stc_individual_income_tax/manifest.yaml": ( + "b6146e6ce891dc26306f0ad550e608e63118ae33029285974158ee44fdd5085f" + ), + "db/data/cms_aca/effectuated_enrollment_2022/manifest.yaml": ( + "d6fb650034b2d1ef4b783f92fae9f31422534ed2c391add978652d9f4455e322" + ), + "db/data/cms_aca/oep_state_level/manifest.yaml": ( + "671bb0ae03eec9317ba17fd2adcdba6b289cb12f1f2fdc776bb3274adb234a3c" + ), + "db/data/cms_aca/oep_state_level_2022/manifest.yaml": ( + "0b0a9c41400b77504a60ab8e07d1559360f93b1af946e1fbcf4c6418646aab1a" + ), + "db/data/cms_aca/oep_state_level_2025/manifest.yaml": ( + "9de293a350360e0d1804ad1d1fe9fca73f71ca613fd9094db50530e7e3f19d4b" + ), + "db/data/cms_medicaid/chip_monthly_enrollment_dataset/manifest.yaml": ( + "8fa93cf69e7c4466d7c73680ca0a287245550c9ffdfbcfde631aa98fdf38455a" + ), + "db/data/cms_medicare/medicare_trustees_report_2025/manifest.yaml": ( + "8fa057bcf1bcb630bb4e7ce8c32a199378d57886ea7be85f0846beb1cebac1e3" + ), + "db/data/cms_nhe/historical_service_source/manifest.yaml": ( + "2fe570abe9079b013c03045b5714cb81af68c4d4b747494ba457f5097bc368d5" + ), + "db/data/cms_nhe/historical_service_source/manifest_source_package.yaml": ( + "319e4e352ebbfd7af32b6c9f67811854eeb7615c23efa02c14cd2d284538face" + ), + "db/data/cms_nhe/table_24/manifest.yaml": ( + "d6c70a307040d5fd4e99f4ecec6a97a167ac71d17bc2d42cf3ce27bd8ae648d8" + ), + "db/data/dfc_ni/uc_statistics_may_2026/manifest.yaml": ( + "b8ed0c6a85339195940f75148152f0f924f9a9500bbcb2b26b35c90d712d1591" + ), + "db/data/dfe/funded_early_education_childcare_2026/manifest.yaml": ( + "6cc43ea3d3e16ec4845e42c2972b415ae2db7c07d035ed93b1db3ef4562eb67c" + ), + "db/data/dft/bus0415_fares_index_2026/manifest.yaml": ( + "6453b9a65651239eddc97a0b41114667aafb39c2e9c183773261a870a42dccf0" + ), + "db/data/dft/bus05i_revenue_support_2025/manifest.yaml": ( + "4cea963c6ed0177ea4846279bf32242cc96c0e67530a3bdee4a4970ff6826d96" + ), + "db/data/dft/nts0705_local_bus_trips_2024/manifest.yaml": ( + "6bd4ecba81b1e0482356a1bc9e79842743d5f0a07e55476822f3ff946615a7fb" + ), + "db/data/dft/nts_vehicle_ownership_2024/manifest.yaml": ( + "46c35b4ac7117f41fc1524a70a7712709904f265e5c27e1ab6778b3b18923d26" + ), + "db/data/dwp/benefit_cap_november_2025/manifest.yaml": ( + "92e48e42d94d2dc7da18c39cb061c17c0a15f935de8f9f552f56cee4aac54323" + ), + "db/data/dwp/benefit_statistics_february_2026/manifest.yaml": ( + "dd07920ba85e98726da226366d6d93854ea0b06338f587ed3962db353fc78a8b" + ), + "db/data/dwp/pip_daily_living_foi_2025/manifest.yaml": ( + "4ee8292dbc7c9d539c9f86e4c1926c7cc044b071006252622f8716f79418839a" + ), + "db/data/dwp/uc_childcare_element_march_2021_august_2025/manifest.yaml": ( + "eac3f1eeb946b68b9bcd2460fc621c4f60956808b332b04743c78071beed8132" + ), + "db/data/dwp/uc_deductions_march_2025_february_2026/manifest.yaml": ( + "2c148e2981347334c537f12a8e6caea11ccae543678271f76308af70594b2081" + ), + "db/data/dwp/uc_households_by_constituency_children_may_2025/manifest.yaml": ( + "073f66903927591d0e3ad5f36eb5cc9ccc573a28b379c600c9d74ac1a189a15d" + ), + "db/data/dwp/uc_households_by_constituency_may_2025/manifest.yaml": ( + "a846fc14a57e82b3eb71d2d72afa21e57ed05e9d922bb4dd04fe73022859fa4a" + ), + "db/data/dwp/uc_households_by_local_authority_may_2025/manifest.yaml": ( + "ff95bda95cc1e56f4993b75c3f218373c41c77097f5e804da28f1e16d5cfbe81" + ), + "db/data/dwp/uc_households_carer_entitlement_april_december_2025/manifest.yaml": ( + "67729d45580419b84b85dc679e45869dbc2f3dafda46cec53d5a531543f0a4d3" + ), + "db/data/dwp/uc_households_children_april_december_2025/manifest.yaml": ( + "9b44c346294f84fcfb5e199f8b555d052dae828cecb7de2a633a46c34d215b62" + ), + "db/data/dwp/uc_households_family_type_april_december_2025/manifest.yaml": ( + "6a55f6c3219c1f8bbea56800bfa9a324f61bec446008bc412949c4f66f0ea812" + ), + "db/data/dwp/uc_households_housing_entitlement_april_december_2025/manifest.yaml": ( + "867b596a4224ab1e73a03c6951106f49e0eb1d5821aa397a292e8bafaf9c3dba" + ), + "db/data/dwp/uc_households_lcwra_entitlement_april_december_2025/manifest.yaml": ( + "6a0164ef778d320dbbe61e414a68066bc1b64fe42ebf5bbb6571a4d47fd7f1af" + ), + "db/data/dwp/uc_payment_distribution_may_2025/manifest.yaml": ( + "fb54f10c9829ab08b1304cf50644dff46d5241cc156edb91b081becec34558cb" + ), + "db/data/dwp/uc_scotland_youngest_child_may_2025/manifest.yaml": ( + "0c0d3dcb13bdb4ad391dd7198ee3b75f4e185b0000f149092ea92b767c5cb110" + ), + "db/data/dwp/uc_two_child_limit_2025/manifest.yaml": ( + "068959db08a1a970caaafed86caa1cb48e5aafc1cbd800af98fed09c309a03d4" + ), + "db/data/eurostat/gov_10a_taxag/manifest.yaml": ( + "d023d782a0d34ab0823fdd75394b234f9d23754060f334726fb936c333170293" + ), + "db/data/eurostat/ilc_di01/manifest.yaml": ( + "7da09c7bcb936fc34672f88dffc6ce4e16b67254823b8dc2ccf76956858d0992" + ), + "db/data/eurostat/ilc_li02/manifest.yaml": ( + "2cbcdb9face89a8e66716c4aaa6167091a07fadf8aba1f12f3975e92c00d4e7c" + ), + "db/data/eurostat/nasa_10_nf_tr/manifest.yaml": ( + "627cb9b76778e8cd85062c7afe3ce4d0581497d15f0fdd048a1e2aec7910f62f" + ), + "db/data/eurostat/spr_exp_func/manifest.yaml": ( + "618e21e482f571fc882d49b848f0c57c01f01d6b9c70faa2c39baa0402750496" + ), + "db/data/federal_reserve/z1_household_net_worth_2026/manifest.yaml": ( + "ba43416d74a2d30b727625e9ce80825511b955b7a3507a5e8b6d5f786b437c5f" + ), + "db/data/fpb/economic_outlook_2026_2031_june_2026/manifest.yaml": ( + "2251eeb3287c72e558f94811122962b56a30a8071d8d10508e025ef2fb98c27e" + ), + "db/data/hhs_acf/tanf_caseload_2024/manifest.yaml": ( + "013fad141428e15fc4d29d0f61501522e20e21f278765716aec5635559bf8108" + ), + "db/data/hhs_acf/tanf_financial_2024/manifest.yaml": ( + "c8d9a2129b177ddfc33ffb7b0f4cd745b83bb18b3d195ef4327f0224dd604d54" + ), + "db/data/hhs_acf_liheap/fy2023_national_profile/manifest.yaml": ( + "53eca5fd455593231bfe15aa79584b665e413665f51ce4bbc46701a1815fea87" + ), + "db/data/hhs_acf_liheap/fy2024_national_profile/manifest.yaml": ( + "5d44f82f41f911cc213cbc545b70627612fa328ee03f7552d774af644e95d75e" + ), + "db/data/hmrc/cgt_age_2026/manifest.yaml": ( + "19ad2f017bc99c2d3dd48f638055119bc0d77621baf184ca269d045f5f1a00cd" + ), + "db/data/hmrc/cgt_country_region_2026/manifest.yaml": ( + "fe9354f7751797eefcc2a8c35d123e77b936e861db88a9d8205deec7f87076e5" + ), + "db/data/hmrc/cgt_gain_by_income_2026/manifest.yaml": ( + "10a802ead7dd1f81fd3740babc4adf292d0389ac8efed0d83bf78bc1270f47e0" + ), + "db/data/hmrc/cgt_size_of_gain_2026/manifest.yaml": ( + "e77c9f68e40ef7878918605e6348eb8f70472bdb3ac30322d8fc0380e4629b5b" + ), + "db/data/hmrc/cgt_statistics_2026/manifest.yaml": ( + "939a93e7e9f7a437ea58331e0f804408b1a4481e6229bc91f41e5685d09fa4b0" + ), + "db/data/hmrc/salary_sacrifice_reform_2029_headcounts/manifest.yaml": ( + "00e2c4d343532b946ce0c6959468554e0a9bd4059a1c7f171d0a9888f2516464" + ), + "db/data/hmrc/salary_sacrifice_relief_2024_25/manifest.yaml": ( + "ae65160f1da3bbb56681d334091c5a137cb8cd9a3066dc9e20a64fb01de3847b" + ), + "db/data/hmrc/spi_income_bands_2023_24/manifest.yaml": ( + "972496558bc9cadd1ea4607c635a75279460aab160578cff63cd2f11740c1774" + ), + "db/data/hmrc/spi_income_by_area_2023_24/manifest.yaml": ( + "55a57302bf60a2d654a2dd9511416d7654e1dd9d99fa8c9541479799f5eb4c3f" + ), + "db/data/hmrc/tax_free_childcare_march_2026/manifest.yaml": ( + "49ff7bf6aa28f21275f263c378a0619f2026bf2e3356274fe7362ad3f0a997b9" + ), + "db/data/hmrc/vat_firm_sector_targets_2024_25/manifest.yaml": ( + "a775ff5af9834b5deb0f80cf9941efd63db3f97c319969ff246fc32cac6eb881" + ), + "db/data/hmrc/vat_firm_targets_2024_25/manifest.yaml": ( + "b557e65210cfd8bd55180075af09e15a1a8b6cd8fcca693d96805e653af8936b" + ), + "db/data/ici/fact_book_table_30/manifest.yaml": ( + "7fd06be33ec2af5658f8ffea95b8d9aa8276f2cf2d569d15ff7dade754f5f307" + ), + "db/data/irs_soi/congressional_district_2022/manifest.yaml": ( + "62e1d4c8d00b0f3e3d2c57ae6accd064d8b4118788e2524369604e7cbb1a39a9" + ), + "db/data/irs_soi/county_2022/manifest.yaml": ( + "84dd9f95b7a17478cdf852ca85202be68484181b855e5c6d2a39ce0abd67cf0b" + ), + "db/data/irs_soi/filing_season_week47_2024/manifest.yaml": ( + "a58b5e8e998a0273d37c7b073ffce30e2d56723c30048a4bfb52b360462a2f12" + ), + "db/data/irs_soi/historic_table_2/manifest.yaml": ( + "941ce7849e7834e188436bec8c880f0917e0509a6fccf2565d7d6dba69ccb5bc" + ), + "db/data/irs_soi/ira_contributions/manifest_roth_source_package.yaml": ( + "2c2f5031e81358222e35ed1ccdf0d285fd49aac7f8ba90d5ef40e38856d453fa" + ), + "db/data/irs_soi/ira_contributions/manifest_traditional_source_package.yaml": ( + "e7f447dc67770be73a00c88ded5dff2b2e8569b876f244c7db3667b44ee7df61" + ), + "db/data/irs_soi/state_2022/manifest.yaml": ( + "1c8ab10d77167adce93cdc695fb2a41afb997b81001962462c96cacaa37e6575" + ), + "db/data/irs_soi/table_1_1/manifest.yaml": ( + "3877ef9629ad30fcaf19aa0b2cf75ce7797aea7598d02a7cc3af7ae945c43a51" + ), + "db/data/irs_soi/table_1_2/manifest.yaml": ( + "6e3db0b2809cced841f0acfa073dfa0304389f08753c8bdfae21c55ca9845b80" + ), + "db/data/irs_soi/table_1_4/manifest.yaml": ( + "d9b365c0e9fa17874af0eb5aa4014fe8d7edca21fa54889ad3c7a5d533a9d469" + ), + "db/data/irs_soi/table_2_1/manifest.yaml": ( + "02cff4531ddd4ab456e6b0cefd95d711f2370d13c48b1c5b1736a51bc0f5f6f3" + ), + "db/data/irs_soi/table_2_5/manifest.yaml": ( + "51246f286233be241d0fb3243efe8dfc86d3390233d56c1e49931dab063b6a10" + ), + "db/data/irs_soi/table_4_3/manifest.yaml": ( + "f0a0836f9856b2a767309577ee0f61a9c8adcb2b31a4ca9f4f45d2436818f9f6" + ), + "db/data/irs_soi/w2_statistics/manifest_2020_source_package.yaml": ( + "9f61439825fa7ea82325310e0f04d5292842c468b471632385ecd1a60133c3d3" + ), + "db/data/isc/annual_census_2023/manifest.yaml": ( + "178582171c6636a1f8b2bff88213ea1583db6849abb19556a7b088cbced8c26f" + ), + "db/data/isc/annual_census_2024/manifest.yaml": ( + "a9c03be443bc0c684a6f8be08c9cc3ae51e6d379db17a04ecccf050c34dced88" + ), + "db/data/jct/obbba_revenue_estimates_2025/manifest.yaml": ( + "12e1ae416138e6171f3f043328859de9aef4ebcfd5a1af822198859ad823b667" + ), + "db/data/jct/tax_expenditures_2024/manifest.yaml": ( + "e3216c3e1b782691ed9da6c45b9a853149a7a863a5dd4a0efe72845f9a7f1320" + ), + "db/data/jrc/euromod_be_baseline_statistics_2025/manifest.yaml": ( + "7f02b5f98607f7b69fca12e28e92e8c8cf2a091712e8b34f81a55164740b4efb" + ), + "db/data/kff/marketplace_effectuated_enrollment/manifest.yaml": ( + "0227988dea8c69843fed2d3c4c179de5e6f2c242ccb1e9eb8b6574ed12420748" + ), + "db/data/mhclg/council_tax_collection_england_2025_26/manifest.yaml": ( + "ceed04b0de8fc684be2f8856868def92d3c98ad953aa79ad7e3c44ec64ce0c6e" + ), + "db/data/mhclg/council_tax_levels_england_2026_27/manifest.yaml": ( + "f0af6b945807a6da7cefcba4eee33840ecf5ffaa189a6e6d0df24ed7d3e4d68e" + ), + "db/data/mhclg/council_tax_levels_england_summary_2025_26/manifest.yaml": ( + "0fc78b37cac572e48a1453226f012596e61a3aa690b1f6bad3c0270e4875ab44" + ), + "db/data/mhclg/ehs_weekly_housing_costs_2023_24/manifest.yaml": ( + "9420c57fec384921fea8f948f0fb84a20f7c634072aac92bae23a32982c7cd1f" + ), + "db/data/nbb/national_accounts_household_disposable_income_2024/manifest.yaml": ( + "699eb899809e669017d7cd6fdf0e19eb5ca1561228ff01145981883fcae02684" + ), + "db/data/nisra/census2021_households_lgd/manifest.yaml": ( + "6457db483b85c04378432f4f56311e5fe77852c27c35155f59435cbd5a3c2bc6" + ), + "db/data/nisra/census2021_households_pcon24/manifest.yaml": ( + "15fc9bf9ddbda812a0a961244316f601f918d66c4d0876e4de264b57f97edbd1" + ), + "db/data/nisra/census2021_tenure_lgd/manifest.yaml": ( + "c696267fdb598b43c713eb626eae3bbd20b32c23522ce87aff65ab7e10cb40da" + ), + "db/data/nisra/pcon24_population_by_age_2024/manifest.yaml": ( + "973db94fa8a7200a98eb78d642e311b2090dd1d5d1ac54c273ee8e081e4d41a5" + ), + "db/data/nrs/census2022_households_ukpc24/manifest.yaml": ( + "c6bbbc5e23dcd3e3fe61152844debd416f91111e7c8c4e9cea4b5fe9fc54c7e9" + ), + "db/data/nrs/census2022_uv404_tenure_council_area/manifest.yaml": ( + "cbe62456a64cd1b2404fd79b97a730b7400fdf7ee1eef31077cc5d1fbd7bdfea" + ), + "db/data/nrs/pcon24_population_by_age_2024/manifest.yaml": ( + "69e3a274aa93623eff741cb62a718c08c507f78cf4d67357f87d8596ede18ef0" + ), + "db/data/obr/efo_aggregates_march_2026/manifest.yaml": ( + "64dab83a0d63684a771fa05624931d6fff1b71370021e24aeeaa4db8c4a1bb87" + ), + "db/data/obr/efo_economy_march_2026/manifest.yaml": ( + "33e3f30422ea8170ab13b8828e0c7b7f9dbc26165c7dd216abffece53ae90eae" + ), + "db/data/obr/efo_expenditure_march_2026/manifest.yaml": ( + "9ae17f2ed29fab5091bfe6eea3bc3eda16f4614fd7572206041e7a4f39785d86" + ), + "db/data/obr/efo_receipts_march_2026/manifest.yaml": ( + "a0794755ea8de46cf98456b965127087e9512cd636a50ea882397f946abbfd47" + ), + "db/data/onem_rva/unemployment_2024/manifest.yaml": ( + "bcc298a74823509123de5cb4ace444c22f67516f009b2d9ade4adbae9b74fe2e" + ), + "db/data/ons/census2021_ts041_households_lad/manifest.yaml": ( + "08b60f70f9548def4f821b0326bf1a4ae1a05fafcc703900a8221cb9689d4e7b" + ), + "db/data/ons/census2021_ts041_households_pcon24/manifest.yaml": ( + "9393cf5f5c6e2ce29d40938fff52c843236a958115cba465b8d83c8d414cfa91" + ), + "db/data/ons/census2021_ts054_tenure_lad/manifest.yaml": ( + "67be94cefa57786109e29793d76bb21af374aa8325ac98c9fdbd493bd2cb7c26" + ), + "db/data/ons/families_households_2025/manifest.yaml": ( + "d9ca199ec60bcf584757065e918d20d25baa010499b7ba779814a7a79c4c2d3b" + ), + "db/data/ons/households_by_type_country_2025/manifest.yaml": ( + "11cb033bd19846de333c20f025773019db105fd3ca2b9bbc9cfffb9cb6def7c6" + ), + "db/data/ons/lad_population_by_age_2024/manifest.yaml": ( + "96d3dd78b53cf1c3b9de6e1556f535cfa491a2728ee214b6ea89ebd0aa0f73b7" + ), + "db/data/ons/mye_2023_england_regions/manifest.yaml": ( + "f7bfe957544c7b498a005c825f9448f253935ae641a3bdac604319260609e949" + ), + "db/data/ons/mye_2023_uk_countries/manifest.yaml": ( + "9a553c91a801bc1544c538e22761f4eb226eb2209a770667e9d1c902f74751b9" + ), + "db/data/ons/mye_2024_uk/manifest.yaml": ( + "d23afe667223bad9c9fa1f954ccab65d21a8af1e221a9eb50fb9e2debf0c097c" + ), + "db/data/ons/national_balance_sheet_land_2025/manifest.yaml": ( + "091412976df9d01cac8de486b380f2e7ae129a3209a09d59acf2471ad81909e5" + ), + "db/data/ons/npp_2024_uk/manifest.yaml": ( + "36bc156c49939ad023d3e57489f4d6ac94d5cc8b8b6cc81ca782ed4197ad1d62" + ), + "db/data/ons/pcon24_population_by_age_2024/manifest.yaml": ( + "f3042a9f4764d34d17ecbb6b539a2f0b736ac36e64e3328a9c980060112a84bf" + ), + "db/data/ons/pipr_private_rent_march_2026/manifest.yaml": ( + "9c8548f433803ea081fcc3529905f6a8fcd6499fe7e5887e5f674c39a90ac593" + ), + "db/data/ons/pipr_rents_by_area_june_2026/manifest.yaml": ( + "3d6d8c202aa8c40f79d79cef735aa53167f2e6383a78f4d07227ea039dbe87f5" + ), + "db/data/ons/public_sector_employment_2026/manifest.yaml": ( + "a126604b78b10e5dc6bf2019fdb118862bc23402a3f6c5b49d37ec5f4e3c4a2f" + ), + "db/data/ons/savings_interest_income/manifest.yaml": ( + "f85200aaafa452dc3f2b7e7940253c552d47f481a068f6adff63ef9a511e8de7" + ), + "db/data/ons/small_area_income_msoa_fye2023/manifest.yaml": ( + "f5b0813fdbd8ab84ed651e3442c93b034c7b1f5995aaf65b0f4b9215c53c2800" + ), + "db/data/ons/subnational_dwellings_by_tenure_2024/manifest.yaml": ( + "66926aa574b44a9b2db66fdd65f1a1d4b5a36cc976ac04c3568c7baa0bf6e423" + ), + "db/data/ons/uk_business_firm_sector_targets_2025/manifest.yaml": ( + "b5c1101e89b6c47835fb22addfae764edc2c728d35def77c701ca95343ea9f6b" + ), + "db/data/ons/uk_business_firm_targets_2025/manifest.yaml": ( + "407e08fb557132d8a9f5d7f0ce2552b1023c2b86c9da1bbb9dc1f7932a6a3c34" + ), + "db/data/onss/contributions_2024/manifest.yaml": ( + "cd8dac264b4aae0e9257f4d3f1dae8cbc61d42cb58ee4449439ef01e0c1f3c66" + ), + "db/data/opgroeien/groeipakket_caseload_2025/manifest.yaml": ( + "d74a0b9c70eee26dc349985cc231ef81e5f2684bb5e7d63987cde2abd62fd8eb" + ), + "db/data/scotgov/band_d_council_tax_rates_2026_27/manifest.yaml": ( + "290d477aa64e951e34fb560a7892f1a4c77aebf2ce71a6e1652d95669b2d57eb" + ), + "db/data/scotgov/band_d_equivalents_2025/manifest.yaml": ( + "22e6ac2a13339d50f7ee2c86d886b32d5242330290687f2c8875ad11e0495e3f" + ), + "db/data/scotgov/council_tax_bands_2025/manifest.yaml": ( + "67918754625b035094c79f30d62cc638e96bc4925faf6bc78e981ae6275759cc" + ), + "db/data/scotgov/council_tax_collection_2024_25/manifest.yaml": ( + "3cc00a7c7762ed68ea7c9145b429a26360f061ac89518d510dacb123f846cd20" + ), + "db/data/scotgov/council_tax_collection_2025_26/manifest.yaml": ( + "bde33b4e442769aa93b215667ce4c87e5747f86b056306ada61af8862b5f5220" + ), + "db/data/scotgov/scottish_budget_social_security_assistance_2026/manifest.yaml": ( + "a46059c0c5cde32d90a7291ec1b497d65d3f11e93cccf532f04b7684b52fa0e9" + ), + "db/data/scotgov/slgfs_council_tax_2024_25/manifest.yaml": ( + "13bed33835cf6829cb0a34f78f9a59c03ee1fd5c8f18c7589c2b3549b8b15317" + ), + "db/data/sfpd/legal_pension_caseload_2025/manifest.yaml": ( + "6aee6347bdbd98534cfadf0620a63bf875eb61722845a2efb902a605e0292885" + ), + "db/data/slc/student_loan_borrower_forecasts_england_2025/manifest.yaml": ( + "5bcb4476cd10707cb12c5583207f96858b7f99af58cf02bb230719e96083da24" + ), + "db/data/slc/student_loan_repayments_england_2025/manifest.yaml": ( + "cdabb4ef47f3e31d7bdc6c03cc5104e7311a2944ef3fef889f6848e65ce49f4e" + ), + "db/data/slc/student_loan_repayments_northern_ireland_2025/manifest.yaml": ( + "e6c9012ee553b3c2b745b3519b8eef57bc265009841b27fa9624330e6ad67cc1" + ), + "db/data/slc/student_loan_repayments_scotland_2025/manifest.yaml": ( + "1653c9411a3aa209c4246d06411a41a0c9ca7c35335e22ba14edeef6773f04df" + ), + "db/data/slc/student_loan_repayments_wales_2025/manifest.yaml": ( + "d78d6ecbd553c88f6fdb37f5230e8ce1a659e028c912eb41ad037779d2565187" + ), + "db/data/slc/student_support_england_2025/manifest.yaml": ( + "be20a9f99f9b670e1066256a055e9bd36cd3f3438ec1ff5b84ebd34fb1e970ca" + ), + "db/data/spf_finances/pit_2023/manifest.yaml": ( + "3fb455eece095ec8178067c560eb8f125bcc3e849af569494142d1770914fedb" + ), + "db/data/ssa/annual_statistical_supplement_2025/manifest.yaml": ( + "04028ca0dd26e94acdfddda08a36998f440ba226403b85e54d0cfa5a23c9b43f" + ), + "db/data/ssa/ssi_monthly_statistics_2024_12/manifest.yaml": ( + "8dc1e91c4f49b7db3ecb30eaa7e9ade885e4c4c96125373551163286fcb05903" + ), + "db/data/ssa/ssi_table_7b1_2024/manifest.yaml": ( + "11e11f4ba5ed569cae2a0a4494c149ea9fc2919e0cb8a255cd591c6f7e1891e1" + ), + "db/data/statbel/fiscal_income_commune_2023_nis_2025/manifest.yaml": ( + "c0d93d91b73201deb31d8234bf5ff6b79b67e6600cab897214ce5b135a57839f" + ), + "db/data/statbel/fiscal_income_distribution_2023/manifest.yaml": ( + "612a5debce569635dfabc7df48747c938388ad4e624f32eb02de6f2b0d7ac1e5" + ), + "db/data/statbel/nis_2025_commune_crosswalk/manifest.yaml": ( + "d38cc3252396c9f343b5acca1f1537d8b128a05d8c6acb4846e49a25ae680cef" + ), + "db/data/statbel/population_structure_nuts1_2025/manifest.yaml": ( + "1ee3c24db3c24feb887225ef7565a3e8443d53666fbcbf8328fae57c902cb8fb" + ), + "db/data/statbel/population_structure_nuts1_2026/manifest.yaml": ( + "e21a51518842e591c535145bbbe8e6e2ff7cb33fa53513d8d17a1443a405139a" + ), + "db/data/usda_snap/fy69_to_current/manifest.yaml": ( + "e735895977bfb23c2a2a7d36b4b255be4fcd7b2712b06468de7d936dcf841830" + ), + "db/data/usda_snap/fy69_to_current/manifest_fy2025_monthly_source_package.yaml": ( + "ed366f4d02e86356abce8c30ee54b341154b48deb615819c5e71e538df3851c8" + ), + "db/data/voa/council_tax_bands_2025/manifest.yaml": ( + "1e5d497ad917eb0c0c17003d915eb37b1a8def303c065378e785e59afbaf3ca4" + ), + "db/data/voa/council_tax_stock_by_lad_2025/manifest.yaml": ( + "4b2da68c429bb75ea915f753dc8c66d401d9858a92f4cdd403a9107cc3ececfd" + ), + "db/data/welshgov/council_tax_collection_2024_25/manifest.yaml": ( + "2b6bee77282acc6bee952161d432684e0fde6f66a4dd4d59e0fe8aa8d964fcff" + ), + "db/data/welshgov/council_tax_collection_2025_26/manifest.yaml": ( + "769dd2f8f7178f4df2290316eb2034ffbdb4850fc25e71b4ebf5c319d8be94d2" + ), + "db/data/welshgov/council_tax_levels_2026_27/manifest.yaml": ( + "36021ed92e559233d7442685df4e04331b11fd320fe13f40ac9b1e9afc0f2850" + ), + "db/data/welshgov/ctrs_annual_report_2024_25/manifest.yaml": ( + "e3836e875eaf9656e296e66e7017086161a3f76303697ff0d132640bd99dfd70" + ), + "db/data/welshgov/ctrs_annual_report_2025_26/manifest.yaml": ( + "dcbcd60d7ff2775827205dcbde03a3479cc8c53f030dc1943b9a2ba5e9609455" + ), + } +) + + +def manifest_digest(manifest_path: Any) -> str: + """Return the SHA-256 of a manifest file's bytes.""" + return hashlib.sha256(manifest_path.read_bytes()).hexdigest() + + +def grandfathered_manifest_key(manifest_path: Any) -> str | None: + """Return the frozen-list key ``manifest_path`` addresses, if any. + + Manifests are addressed by their repository-relative path, so the lookup + matches the longest trailing run of path segments that is a key. A path + outside the repository can only match by carrying the same segments, and + then only counts once its bytes match too. + """ + parts = PurePosixPath(str(manifest_path).replace("\\", "/")).parts + for start in range(len(parts)): + candidate = "/".join(parts[start:]) + if candidate in GRANDFATHERED_KINDLESS_MANIFESTS: + return candidate + return None + + +def is_grandfathered_manifest(manifest_path: Any) -> bool: + """Whether the file at ``manifest_path`` is frozen kindless, byte for byte.""" + key = grandfathered_manifest_key(manifest_path) + if key is None: + return False + try: + digest = manifest_digest(manifest_path) + except (OSError, AttributeError): + return False + return digest == GRANDFATHERED_KINDLESS_MANIFESTS[key] diff --git a/chronicle/licences.py b/chronicle/licences.py new file mode 100644 index 00000000..a3679ba7 --- /dev/null +++ b/chronicle/licences.py @@ -0,0 +1,137 @@ +"""Redistributable licence terms Chronicle may archive microdata bytes under. + +Being downloadable is not a licence, and a licence *name* on a manifest entry +is not evidence that a particular file was issued under it. Chronicle archives +a microdata release's bytes only when the entry's ``licence`` is one of the +terms below and the entry carries ``licence_evidence`` binding the artifact to +that term (``docs/adr-chronicle-raw-microdata-identity.md``). This module is +the allowlist, kept in code so every term carries the evidence for why +redistribution is permitted. + +Adding a term is a reviewed code change: give it a stable identifier (SPDX +where one exists), the legal basis, and a durable URL to the terms. +""" + +from __future__ import annotations + +from collections.abc import Mapping +from dataclasses import dataclass +from types import MappingProxyType +from typing import Any + +__all__ = [ + "LICENCE_EVIDENCE_FIELDS", + "REDISTRIBUTABLE_LICENCES", + "RedistributableLicence", + "is_redistributable_licence", + "licence_evidence_errors", +] + + +@dataclass(frozen=True) +class RedistributableLicence: + """One term under which Chronicle may hold and re-serve publisher bytes.""" + + identifier: str + name: str + basis: str + evidence_url: str + + +REDISTRIBUTABLE_LICENCES: Mapping[str, RedistributableLicence] = MappingProxyType( + { + "US-Government-Work": RedistributableLicence( + identifier="US-Government-Work", + name="Work of the United States Government", + basis=( + "17 U.S.C. §105: copyright protection is not available for any " + "work of the United States Government, so a federal statistical " + "agency's public-use file may be copied and redistributed." + ), + evidence_url=( + "https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-" + "title17-section105" + ), + ), + "OGL-UK-3.0": RedistributableLicence( + identifier="OGL-UK-3.0", + name="Open Government Licence v3.0", + basis=( + "The licence grants a worldwide, royalty-free, perpetual, " + "non-exclusive licence to copy, publish, distribute and transmit " + "the information, subject to attribution." + ), + evidence_url=( + "https://www.nationalarchives.gov.uk/doc/open-government-licence/" + "version/3/" + ), + ), + "CC0-1.0": RedistributableLicence( + identifier="CC0-1.0", + name="Creative Commons CC0 1.0 Universal", + basis=( + "The affirmer waives all copyright and related rights, so the " + "work may be copied and redistributed without restriction." + ), + evidence_url="https://creativecommons.org/publicdomain/zero/1.0/legalcode", + ), + "CC-BY-4.0": RedistributableLicence( + identifier="CC-BY-4.0", + name="Creative Commons Attribution 4.0 International", + basis=( + "Section 2(a)(1) grants a worldwide, royalty-free, non-exclusive " + "licence to reproduce and share the licensed material, subject to " + "attribution." + ), + evidence_url="https://creativecommons.org/licenses/by/4.0/legalcode", + ), + } +) + +#: Fields a ``licence_evidence`` block must carry to bind an artifact to a term. +LICENCE_EVIDENCE_FIELDS: tuple[str, ...] = ( + "issuer", + "licence", + "scope", + "url", + "sha256", +) + + +def is_redistributable_licence(licence: Any) -> bool: + """Whether ``licence`` names a term on the allowlist.""" + return isinstance(licence, str) and licence.strip() in REDISTRIBUTABLE_LICENCES + + +def licence_evidence_errors( + evidence: Any, + *, + licence: Any, + sha256: Any, +) -> list[str]: + """Return error codes for a ``licence_evidence`` block. + + The block binds one artifact to one allowlisted term: its ``licence`` must + be the entry's own (allowlisted) licence, its ``sha256`` the entry's own + checksum, and its ``url`` a durable http(s) location of the evidence. + """ + if evidence is None: + return ["missing_licence_evidence"] + if not isinstance(evidence, Mapping): + return ["malformed_licence_evidence"] + errors: list[str] = [] + for field in LICENCE_EVIDENCE_FIELDS: + value = evidence.get(field) + if not isinstance(value, str) or not value.strip(): + errors.append(f"licence_evidence_missing_field:{field}") + if errors: + return errors + if evidence["licence"].strip() != str(licence).strip(): + errors.append("licence_evidence_licence_mismatch") + if not is_redistributable_licence(evidence["licence"]): + errors.append(f"licence_not_redistributable:{evidence['licence'].strip()}") + if evidence["sha256"].strip() != str(sha256 or "").strip(): + errors.append("licence_evidence_sha256_mismatch") + if not evidence["url"].strip().startswith(("http://", "https://")): + errors.append("licence_evidence_url_not_durable") + return errors diff --git a/chronicle/registration.py b/chronicle/registration.py index 0194a831..179d5189 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -1,31 +1,48 @@ """Access classification and hash-only registration for Chronicle artifacts. Chronicle registers every raw artifact its consumers build from and stores the -bytes of only those the publisher permits it to redistribute. Two manifest file -fields carry that split: +bytes of only those the publisher permits it to redistribute. Manifest fields +carry that split: + +``kind`` + Manifest-level: ``publisher_table`` or ``microdata_release``. Every + manifest created or modified after + ``docs/adr-chronicle-raw-microdata-identity.md`` declares it. Manifests + that predate the rule are read as publisher tables only while they match + the frozen list in :mod:`chronicle.grandfathered_manifests`; any other + kindless manifest is an error, never a publisher table by default. ``licence`` - The publisher's terms, as an identifier or a URL. + The publisher's terms. For a public microdata release this is an + identifier from the allowlist in :mod:`chronicle.licences`, and the entry + also carries ``licence_evidence`` binding this artifact to that term. ``access`` A closed class: ``public``, ``licensed``, or ``restricted``. -``public`` artifacts keep the existing fetch/publish path: bytes are archived in -the raw R2 bucket under the content-addressed key +``hash_source`` and its attester + Who asserts the checksum: ``chronicle_fetch`` (``attested_by: chronicle``, + ``verified_at`` = fetch date), ``consumer_attested`` (``attested_by`` = the + consumer, ``attestation_evidence``, ``verified_at``), or ``consumer_pin`` + (``attested_by`` = the consumer, ``pinned_from`` = repository, path and + commit, no ``verified_at``). + +``public`` artifacts keep the fetch/publish path: bytes are archived in the raw +R2 bucket under the content-addressed key ``raw/{source_id}/{package_id}/{year}/{sha256}/{filename}``. ``licensed`` and ``restricted`` artifacts are registered *hash-only*: the manifest records the -checksum, vintage, licence, and access route, and no Chronicle store ever holds -the bytes. That key exists only for ``public`` artifacts. +checksum, vintage, licence, access route and attestation, and no Chronicle +store ever holds the bytes. That key exists only for ``public`` artifacts. A registration is identified by ``{source_id, package_id, year, sha256, -filename}``. Consumers reference a registration by exactly that tuple. - -See ``docs/adr-chronicle-raw-microdata-identity.md``. +filename}``. Consumers reference a registration by exactly that tuple. The +filename is a bare name inside the package directory, compared case-folded, +and ``2023`` and ``'2023'`` are one vintage key. """ from __future__ import annotations -from collections.abc import Iterable, Mapping +from collections.abc import Iterable, Iterator, Mapping from dataclasses import dataclass from pathlib import Path import re @@ -33,28 +50,54 @@ import yaml +from chronicle.grandfathered_manifests import is_grandfathered_manifest +from chronicle.licences import is_redistributable_licence, licence_evidence_errors + ACCESS_PUBLIC = "public" ACCESS_LICENSED = "licensed" ACCESS_RESTRICTED = "restricted" #: The closed set of access classes a manifest file entry may declare. ACCESS_CLASSES: tuple[str, ...] = (ACCESS_PUBLIC, ACCESS_LICENSED, ACCESS_RESTRICTED) -#: Access class inferred for an entry that does not declare one. +#: Access class inferred for a publisher-table entry that does not declare one. DEFAULT_ACCESS = ACCESS_PUBLIC PUBLISHER_TABLE_KIND = "publisher_table" MICRODATA_RELEASE_KIND = "microdata_release" -#: The closed set of manifest kinds. Manifests without ``kind`` are tables. +#: The closed set of manifest kinds. MANIFEST_KINDS: tuple[str, ...] = (PUBLISHER_TABLE_KIND, MICRODATA_RELEASE_KIND) +#: Kind of a manifest that does not exist yet, and of a frozen kindless one. DEFAULT_MANIFEST_KIND = PUBLISHER_TABLE_KIND +HASH_SOURCE_CHRONICLE_FETCH = "chronicle_fetch" +HASH_SOURCE_CONSUMER_ATTESTED = "consumer_attested" +HASH_SOURCE_CONSUMER_PIN = "consumer_pin" +#: The closed set of checksum provenances a registration may declare. +HASH_SOURCES: tuple[str, ...] = ( + HASH_SOURCE_CHRONICLE_FETCH, + HASH_SOURCE_CONSUMER_ATTESTED, + HASH_SOURCE_CONSUMER_PIN, +) +#: Provenances a hash-only registration may declare: Chronicle never fetched +#: the bytes, so the checksum is always the consumer's. +HASH_ONLY_HASH_SOURCES: tuple[str, ...] = ( + HASH_SOURCE_CONSUMER_ATTESTED, + HASH_SOURCE_CONSUMER_PIN, +) +#: The attester of a ``chronicle_fetch`` checksum. +CHRONICLE_ATTESTER = "chronicle" +#: Fields a ``pinned_from`` block carries: where the consumer's pin was read. +PINNED_FROM_FIELDS: tuple[str, ...] = ("repository", "path", "commit") + _SHA256_RE = re.compile(r"^[0-9a-f]{64}$") +_COMMIT_RE = re.compile(r"^[0-9a-f]{40}$") # Registration entry key order, so emitted manifests are byte-stable. _REGISTRATION_FIELD_ORDER: tuple[str, ...] = ( "filename", "access", "licence", + "licence_evidence", "vintage", "sha256", "size_bytes", @@ -65,6 +108,9 @@ "fetched_at", "verified_at", "hash_source", + "attested_by", + "attestation_evidence", + "pinned_from", "notes", ) @@ -73,6 +119,18 @@ class ManifestAccessError(ValueError): """Raised when a manifest declares an unusable access class or kind.""" +class ManifestKindError(ManifestAccessError): + """Raised when a manifest declares no kind and is not frozen kindless.""" + + +class ArtifactFilenameError(ManifestAccessError): + """Raised when a filename is not a bare name inside the package directory.""" + + +class AmbiguousVintageKeyError(ManifestAccessError): + """Raised when a manifest records one vintage under both key spellings.""" + + class HashOnlyRegistrationError(ValueError): """Raised when a hash-only registration is malformed or would store bytes.""" @@ -92,13 +150,37 @@ class ListSpecRejected: spec: Any -def manifest_kind(manifest: Mapping[str, Any] | None) -> str: - """Return a manifest's declared kind, defaulting to ``publisher_table``.""" - if not isinstance(manifest, Mapping): +# -------------------------------------------------------------------------- +# Manifest kind +# -------------------------------------------------------------------------- + + +def manifest_kind( + manifest: Mapping[str, Any] | None, + *, + manifest_path: Any = None, +) -> str: + """Return a manifest's declared kind. + + An absent or empty manifest has the default kind: the command creating it + declares one. A manifest with content must declare ``kind`` itself, unless + ``manifest_path`` names a file frozen kindless before the rule and its + bytes still match the freeze. + """ + if not isinstance(manifest, Mapping) or not manifest: return DEFAULT_MANIFEST_KIND declared = manifest.get("kind") if declared is None: - return DEFAULT_MANIFEST_KIND + if manifest_path is not None and is_grandfathered_manifest(manifest_path): + return PUBLISHER_TABLE_KIND + where = str(manifest_path) if manifest_path is not None else "Manifest" + raise ManifestKindError( + f"{where} declares no kind. Every manifest created or modified " + f"after the microdata-identity ADR declares kind: one of " + f"{list(MANIFEST_KINDS)}; a kindless manifest is read as a " + "publisher table only while it matches the frozen list in " + "chronicle/grandfathered_manifests.py byte for byte." + ) kind = str(declared) if kind not in MANIFEST_KINDS: raise ManifestAccessError( @@ -107,18 +189,41 @@ def manifest_kind(manifest: Mapping[str, Any] | None) -> str: return kind -def safe_manifest_kind(manifest: Mapping[str, Any] | None) -> tuple[str, str | None]: - """Return ``(kind, error_code)`` without raising on an unknown kind.""" +def safe_manifest_kind( + manifest: Mapping[str, Any] | None, + *, + manifest_path: Any = None, +) -> tuple[str, str | None]: + """Return ``(kind, error_code)`` without raising. + + The reporting commands use this so a manifest with a missing or unknown + kind is still walked and reported. The returned kind is only what the + entries are read *as* for that report; the error code says the manifest + itself is invalid. + """ try: - return manifest_kind(manifest), None + return manifest_kind(manifest, manifest_path=manifest_path), None + except ManifestKindError: + return DEFAULT_MANIFEST_KIND, "manifest_kind_missing" except ManifestAccessError: declared = manifest.get("kind") if isinstance(manifest, Mapping) else None return DEFAULT_MANIFEST_KIND, f"unknown_manifest_kind:{declared}" -def is_microdata_release(manifest: Mapping[str, Any] | None) -> bool: +def is_microdata_release( + manifest: Mapping[str, Any] | None, + *, + manifest_path: Any = None, +) -> bool: """Whether a manifest registers a microdata release rather than a table.""" - return manifest_kind(manifest) == MICRODATA_RELEASE_KIND + return manifest_kind(manifest, manifest_path=manifest_path) == ( + MICRODATA_RELEASE_KIND + ) + + +# -------------------------------------------------------------------------- +# Access +# -------------------------------------------------------------------------- def normalize_access(access: str | None) -> str: @@ -155,6 +260,27 @@ def safe_entry_access(spec: Any) -> str: return ACCESS_RESTRICTED +def strict_entry_access(spec: Any, *, kind: str) -> str: + """Return an entry's access class, refusing to infer one for a release. + + A publisher-table entry that omits ``access`` is public; a microdata + release entry must say what it is, and an unknown class is refused with + the value the manifest actually declares. + """ + if not isinstance(spec, Mapping): + return DEFAULT_ACCESS + declared = spec.get("access") + if declared is None: + if kind == MICRODATA_RELEASE_KIND: + raise ManifestAccessError( + f"Entry {spec.get('filename')!r} declares no access class. A " + "microdata release entry must declare access; Chronicle will " + "not infer public for it." + ) + return DEFAULT_ACCESS + return normalize_access(declared) + + def stores_bytes(access: str) -> bool: """Whether Chronicle may hold this access class's bytes.""" return normalize_access(access) == ACCESS_PUBLIC @@ -165,6 +291,105 @@ def is_hash_only(access: str) -> bool: return not stores_bytes(access) +# -------------------------------------------------------------------------- +# Filenames and vintage keys +# -------------------------------------------------------------------------- + + +def is_bare_filename(value: Any) -> bool: + """Whether ``value`` names a file inside a directory, with no path.""" + text = _text(value) + if text is None or text != str(value) or text in (".", ".."): + return False + if "/" in text or "\\" in text or "\x00" in text: + return False + return Path(text).name == text + + +def bare_filename(value: Any, *, what: str = "filename") -> str: + """Return ``value`` as a bare filename, refusing any other spelling. + + ``./adult.tab``, ``sub/../adult.tab``, ``adult.tab/`` and an absolute path + all resolve to the same file as ``adult.tab`` once joined under the package + directory, so the manifest and every guard use one spelling. + """ + if not is_bare_filename(value): + raise ArtifactFilenameError( + f"{what} must be a bare filename inside the package directory, not " + f"{value!r}; it may not carry a directory, '.', '..', a trailing " + "slash, surrounding whitespace, or an absolute path." + ) + return str(value) + + +def filename_key(value: Any) -> str: + """Return the comparison key for a filename. + + Case-folded, because the package directories these commands run in are as + often as not on a case-insensitive filesystem, where ``ADULT.TAB`` and + ``adult.tab`` are one file. Treating them as one artifact path is the safe + rule everywhere. + """ + return Path(str(value)).name.casefold() + + +def vintage_key_forms(year: Any) -> tuple[Any, ...]: + """Return the key spellings that address the same vintage as ``year``. + + ``2023`` and ``'2023'`` are one vintage: every identity Chronicle derives + from a year (registration ids, R2 keys) renders it as text, and the + source-package reader accepts both. Label keys such as ``'A_1'`` have no + other spelling. + """ + if isinstance(year, bool): + return (year,) + if isinstance(year, int): + return (year, str(year)) + text = str(year) + if text.isdecimal() and (text == "0" or not text.startswith("0")): + return (text, int(text)) + return (year,) + + +def resolve_vintage_key(files: Mapping[Any, Any], year: Any) -> Any | None: + """Return the key ``files`` already uses for ``year``'s vintage, or None. + + Refuses a mapping that records the vintage under both spellings: one + vintage has one key, and Chronicle will not choose which entry is the + record. + """ + present = [form for form in vintage_key_forms(year) if form in files] + if len(present) > 1: + raise AmbiguousVintageKeyError( + f"Vintage {year!r} is recorded under both keys {present!r}; one " + "vintage has one key. Merge the entries by hand first." + ) + return present[0] if present else None + + +def iter_manifest_entries( + manifest: Mapping[str, Any] | None, +) -> Iterator[tuple[Any, int | None, Any]]: + """Yield every ``(key, index, entry)`` a manifest declares, whatever shape. + + Deliberately not gated on the manifest ``kind``: a guard must see the + entries a manifest actually holds, including a list under a manifest whose + kind is absent or misspelled. ``index`` is the entry's position in a list + value and None for a single mapping. + """ + if not isinstance(manifest, Mapping): + return + files = manifest.get("files") + if not isinstance(files, Mapping): + return + for key, spec in files.items(): + if isinstance(spec, list): + for index, entry in enumerate(spec): + yield key, index, entry + else: + yield key, None, spec + + def registration_id( *, source_id: str, @@ -193,6 +418,75 @@ def iter_file_specs(spec: Any, *, kind: str) -> tuple[Any, ...]: return (spec,) +# -------------------------------------------------------------------------- +# Validation vocabulary +# -------------------------------------------------------------------------- + + +def validate_manifest_files(manifest: Mapping[str, Any] | None) -> tuple[str, ...]: + """Return manifest-level error codes: shape, key and filename collisions. + + These are properties of the ``files`` mapping as a whole, which no single + entry can see: a vintage recorded under two key spellings, a filename that + is not a bare name, and two entries that resolve to one file in the + package directory while disagreeing about what it holds. + """ + if not isinstance(manifest, Mapping): + return () + files = manifest.get("files") + if files is None: + return () + if not isinstance(files, Mapping): + return ("files_not_a_mapping",) + + errors: list[str] = [] + for key in files: + for other in vintage_key_forms(key): + if other != key and other in files: + errors.append(f"duplicate_vintage_key:{key}") + + # Same file, one package directory: group every entry by its resolved name. + by_name: dict[str, list[tuple[Any, Mapping[str, Any]]]] = {} + for key, _index, entry in iter_manifest_entries(manifest): + if not isinstance(entry, Mapping): + continue + filename = entry.get("filename") + if filename is None: + continue + if not is_bare_filename(filename): + errors.append(f"non_canonical_filename:{filename}") + by_name.setdefault(filename_key(filename), []).append((key, entry)) + + for name, entries in by_name.items(): + if len(entries) < 2: + continue + classes = {is_hash_only(safe_entry_access(entry)) for _key, entry in entries} + if len(classes) > 1: + # One path cannot be both bytes Chronicle holds and bytes it must + # never hold. + errors.append(f"filename_collision:{name}") + continue + hash_only = classes.pop() + seen: dict[Any, set[str]] = {} + for key, entry in entries: + digest = _text(entry.get("sha256")) or "" + vintage = seen.setdefault(key, set()) + if digest in vintage: + errors.append(f"duplicate_filename_in_vintage:{name}") + vintage.add(digest) + if hash_only: + # Several vintages, or an explicit reissue, may register the same + # filename with different bytes: no file exists to collide. + continue + digests = {_text(entry.get("sha256")) or "" for _key, entry in entries} + if len(digests) > 1: + # Public entries share one path in the tree and one current object + # per name; a revision is recorded in storage.previous_r2, not as a + # second entry. + errors.append(f"filename_collision:{name}") + return tuple(_dedupe(errors)) + + def validate_file_entry( spec: Any, *, @@ -203,7 +497,9 @@ def validate_file_entry( """Return stable error codes for one manifest file entry. The codes are the refusal vocabulary shared by ``inventory-artifacts``, - ``publish-raw``, and ``register-artifact``. + ``publish-raw``, ``fetch-artifact`` and ``register-artifact``. + ``local_file_exists`` says whether the entry's filename exists beside the + manifest, in the package directory. """ if isinstance(spec, ListSpecRejected): return ("list_file_spec_requires_microdata_release_kind",) @@ -211,6 +507,10 @@ def validate_file_entry( return () errors: list[str] = [] + filename = spec.get("filename") + if filename is not None and not is_bare_filename(filename): + errors.append(f"non_canonical_filename:{filename}") + declared_access = spec.get("access") if declared_access is None: if kind == MICRODATA_RELEASE_KIND: @@ -233,9 +533,24 @@ def validate_file_entry( local_file_exists=local_file_exists, ) ) + elif kind == MICRODATA_RELEASE_KIND: + errors.extend( + _public_release_entry_errors(spec, local_file_exists=local_file_exists) + ) + if is_hash_only(access) or kind == MICRODATA_RELEASE_KIND: + errors.extend(_attestation_errors(spec)) return tuple(_dedupe(errors)) +def _checksum_errors(spec: Mapping[str, Any]) -> list[str]: + sha256 = _text(spec.get("sha256")) + if not sha256: + return ["missing_sha256"] + if not _SHA256_RE.match(sha256): + return ["malformed_sha256"] + return [] + + def _hash_only_entry_errors( spec: Mapping[str, Any], *, @@ -246,21 +561,92 @@ def _hash_only_entry_errors( errors: list[str] = [] if not _text(spec.get("licence")): errors.append("missing_licence") - sha256 = _text(spec.get("sha256")) - if not sha256: - errors.append("missing_sha256") - elif not _SHA256_RE.match(sha256): - errors.append("malformed_sha256") + errors.extend(_checksum_errors(spec)) if not _text(spec.get("vintage")): errors.append("missing_vintage") if not _access_route(spec, manifest): errors.append("missing_access_route") - if not (_text(spec.get("verified_at")) or _text(spec.get("fetched_at"))): - errors.append("missing_verification_timestamp") if local_file_exists: errors.append("bytes_present_for_hash_only_entry") if recorded_r2(spec): errors.append("r2_location_for_hash_only_entry") + if recorded_previous_r2(spec): + errors.append("r2_history_for_hash_only_entry") + return errors + + +def _public_release_entry_errors( + spec: Mapping[str, Any], + *, + local_file_exists: bool, +) -> list[str]: + """Return refusal codes for a public microdata release entry. + + Bytes are archived only under an allowlisted licence with evidence that + binds this artifact to it, and never inside the package directory: public + microdata is staged outside ``db/data`` and uploaded from there. + """ + errors: list[str] = [] + errors.extend(_checksum_errors(spec)) + if not _text(spec.get("vintage")): + errors.append("missing_vintage") + licence = _text(spec.get("licence")) + if licence and not is_redistributable_licence(licence): + errors.append(f"licence_not_redistributable:{licence}") + errors.extend( + licence_evidence_errors( + spec.get("licence_evidence"), + licence=licence, + sha256=_text(spec.get("sha256")), + ) + ) + if local_file_exists: + errors.append("bytes_present_for_microdata_release_entry") + return errors + + +def _attestation_errors(spec: Mapping[str, Any]) -> list[str]: + """Return refusal codes for an entry's ``hash_source`` and attester.""" + errors: list[str] = [] + hash_source = _text(spec.get("hash_source")) + if not hash_source: + return ["missing_hash_source"] + if hash_source not in HASH_SOURCES: + return [f"unknown_hash_source:{hash_source}"] + attested_by = _text(spec.get("attested_by")) + if not attested_by: + errors.append("missing_attested_by") + verified_at = _text(spec.get("verified_at")) + if hash_source == HASH_SOURCE_CHRONICLE_FETCH: + if attested_by and attested_by != CHRONICLE_ATTESTER: + errors.append("attested_by_not_chronicle") + if not verified_at: + errors.append("missing_verified_at") + elif hash_source == HASH_SOURCE_CONSUMER_ATTESTED: + if not _text(spec.get("attestation_evidence")): + errors.append("missing_attestation_evidence") + if not verified_at: + errors.append("missing_verified_at") + else: + errors.extend(_pinned_from_errors(spec.get("pinned_from"))) + if verified_at: + errors.append("verified_at_forbidden_for_consumer_pin") + return errors + + +def _pinned_from_errors(pinned_from: Any) -> list[str]: + if pinned_from is None: + return ["missing_pinned_from"] + if not isinstance(pinned_from, Mapping): + return ["malformed_pinned_from"] + errors = [ + f"pinned_from_missing_field:{field}" + for field in PINNED_FROM_FIELDS + if not _text(pinned_from.get(field)) + ] + commit = _text(pinned_from.get("commit")) + if commit and not _COMMIT_RE.match(commit): + errors.append("malformed_pinned_from_commit") return errors @@ -292,6 +678,40 @@ def recorded_r2(spec: Any) -> Mapping[str, Any] | None: return r2 if isinstance(r2, Mapping) else None +def recorded_previous_r2(spec: Any) -> tuple[Any, ...]: + """Return the entry's ``storage.previous_r2`` history, if it carries one.""" + if not isinstance(spec, Mapping): + return () + storage = spec.get("storage") + if not isinstance(storage, Mapping): + return () + previous = storage.get("previous_r2") + if isinstance(previous, list): + return tuple(previous) + return (previous,) if previous else () + + +def records_r2_object(spec: Any) -> bool: + """Whether an entry names any object in the raw bucket, current or past.""" + return recorded_r2(spec) is not None or bool(recorded_previous_r2(spec)) + + +def normalize_hash_source(value: Any, *, allowed: Iterable[str] = HASH_SOURCES) -> str: + """Return a validated ``hash_source`` value.""" + text = _text(value) + allowed = tuple(allowed) + if text is None or text not in allowed: + raise ManifestAccessError( + f"Unknown hash_source {value!r}; expected one of {list(allowed)}." + ) + return text + + +# -------------------------------------------------------------------------- +# Hash-only registration +# -------------------------------------------------------------------------- + + @dataclass(frozen=True) class ArtifactRegistrationReport: """Report from registering one hash-only source artifact.""" @@ -308,6 +728,8 @@ class ArtifactRegistrationReport: access: str registration: str replaced: bool + hash_source: str + attested_by: str errors: tuple[str, ...] = () @property @@ -331,6 +753,8 @@ def to_dict(self) -> dict[str, Any]: "access": self.access, "registration": self.registration, "replaced": self.replaced, + "hash_source": self.hash_source, + "attested_by": self.attested_by, "r2_location": None, "errors": list(self.errors), } @@ -347,6 +771,11 @@ def register_hash_only_artifact( licence: str, access: str, vintage: str, + hash_source: str, + attested_by: str, + attestation_evidence: str | None = None, + pinned_from: Mapping[str, Any] | None = None, + verified_at: str | None = None, size_bytes: int | None = None, source_page: str | None = None, source_url: str | None = None, @@ -357,16 +786,16 @@ def register_hash_only_artifact( table: str | None = None, publisher: str | None = None, fetched_at: str | None = None, - verified_at: str | None = None, - hash_source: str | None = None, notes: str | None = None, allow_reissue: bool = False, ) -> ArtifactRegistrationReport: """Register a licensed or restricted artifact by identity, without bytes. Writes (or updates) a ``kind: microdata_release`` manifest entry carrying the - checksum, size, vintage, licence, access route, and verification timestamp. - No bytes are read, written, or uploaded, and no R2 key is recorded. + checksum, size, vintage, licence, access route, and the attestation of who + asserts the checksum. No bytes are read, written, or uploaded, and no R2 + key is recorded. Every refusal below happens before the manifest is + touched. """ access_class = normalize_access(access) if stores_bytes(access_class): @@ -389,16 +818,19 @@ def register_hash_only_artifact( raise HashOnlyRegistrationError( f"A {access_class} registration must record the artifact vintage." ) - artifact_name = _text(filename) - if not artifact_name or Path(artifact_name).name != artifact_name: + if not is_bare_filename(filename): raise HashOnlyRegistrationError( f"Registration filename must be a bare filename; got {filename!r}." ) - if not (_text(verified_at) or _text(fetched_at)): - raise HashOnlyRegistrationError( - "A hash-only registration must record when the checksum was " - "verified; pass --verified-at." - ) + artifact_name = str(filename) + provenance = _hash_only_attestation( + hash_source=hash_source, + attested_by=attested_by, + attestation_evidence=attestation_evidence, + pinned_from=pinned_from, + verified_at=verified_at, + access_class=access_class, + ) output = Path(output_dir) local_path = output / artifact_name @@ -411,7 +843,10 @@ def register_hash_only_artifact( manifest_path = output / "manifest.yaml" payload = _load_manifest(manifest_path) - existing_kind = manifest_kind(payload) + try: + existing_kind = manifest_kind(payload, manifest_path=manifest_path) + except ManifestAccessError as exc: + raise HashOnlyRegistrationError(str(exc)) from exc if payload and existing_kind != MICRODATA_RELEASE_KIND: raise HashOnlyRegistrationError( f"{manifest_path} is a {existing_kind} manifest; hash-only " @@ -430,9 +865,8 @@ def register_hash_only_artifact( doi=doi, study=study, fetched_at=fetched_at, - verified_at=verified_at, - hash_source=hash_source, notes=notes, + **provenance, ) route_context = dict(payload) if source_page: @@ -445,6 +879,28 @@ def register_hash_only_artifact( _assert_manifest_identity(payload, manifest_path, "source_id", source_id) _assert_manifest_identity(payload, manifest_path, "package_id", package_id) + files = payload.get("files") + if files is not None and not isinstance(files, dict): + raise HashOnlyRegistrationError( + f"{manifest_path} files must be a mapping; it is a " + f"{type(files).__name__}. Chronicle will not write into a manifest " + "it cannot read." + ) + manifest_errors = validate_manifest_files(payload) + if manifest_errors: + raise HashOnlyRegistrationError( + f"{manifest_path} is not a valid manifest: " + f"{', '.join(manifest_errors)}. Fix it by hand before registering " + "into it." + ) + _assert_no_archived_identity(payload, manifest_path, artifact_name, access_class) + + try: + vintage_key = resolve_vintage_key(files or {}, year) + except AmbiguousVintageKeyError as exc: + raise HashOnlyRegistrationError(f"{manifest_path}: {exc}") from exc + key = vintage_key if vintage_key is not None else year + payload.setdefault("source_id", source_id) payload.setdefault("package_id", package_id) payload["kind"] = MICRODATA_RELEASE_KIND @@ -457,7 +913,8 @@ def register_hash_only_artifact( payload.setdefault("table", table) payload.setdefault("files", {}) - entries = _existing_entries(payload["files"], year) + entries = _existing_entries(payload["files"], key) + wanted = filename_key(artifact_name) # Two passes, so re-registering an existing pin stays idempotent even after # a reissue has added a second entry for the same filename. A single pass # would raise on the first filename match with a different checksum before @@ -466,7 +923,7 @@ def register_hash_only_artifact( for index, existing in enumerate(entries): if not isinstance(existing, Mapping): continue - if _text(existing.get("filename")) != artifact_name: + if filename_key(existing.get("filename")) != wanted: continue if _text(existing.get("sha256")) == checksum: entries[index] = entry @@ -477,19 +934,19 @@ def register_hash_only_artifact( existing for existing in entries if isinstance(existing, Mapping) - and _text(existing.get("filename")) == artifact_name + and filename_key(existing.get("filename")) == wanted ] if superseded and not allow_reissue: raise HashOnlyRegistrationError( f"{manifest_path} already registers {artifact_name!r} for " - f"{year} with sha256={superseded[0].get('sha256')!r}. Different " + f"{key!r} with sha256={superseded[0].get('sha256')!r}. Different " "bytes are a new publisher release, not a pin replacement; " "pass --allow-reissue to register both." ) # A reissue sits alongside the pin it supersedes. entries.append(entry) - payload["files"][year] = sorted(entries, key=_entry_sort_key) + payload["files"][key] = sorted(entries, key=_entry_sort_key) output.mkdir(parents=True, exist_ok=True) manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False, allow_unicode=True), @@ -515,9 +972,126 @@ def register_hash_only_artifact( filename=artifact_name, ), replaced=replaced, + hash_source=provenance["hash_source"], + attested_by=provenance["attested_by"], ) +def _hash_only_attestation( + *, + hash_source: Any, + attested_by: Any, + attestation_evidence: Any, + pinned_from: Any, + verified_at: Any, + access_class: str, +) -> dict[str, Any]: + """Validate and return the attestation fields of a hash-only registration. + + Chronicle never fetched the bytes, so the checksum is always the + consumer's: attested against bytes it holds (``consumer_attested``, with + evidence and a date) or transcribed from a reviewed pin in its repository + (``consumer_pin``, with the repository, path and commit and no + verification date). + """ + try: + source = normalize_hash_source(hash_source, allowed=HASH_ONLY_HASH_SOURCES) + except ManifestAccessError as exc: + raise HashOnlyRegistrationError( + f"A {access_class} registration must record how its checksum is " + f"known: {exc} Chronicle holds no bytes for it, so " + f"{HASH_SOURCE_CHRONICLE_FETCH!r} does not apply." + ) from exc + attester = _text(attested_by) + if not attester: + raise HashOnlyRegistrationError( + f"A {source} registration must name the consumer that attests the " + "checksum; pass --attested-by." + ) + fields: dict[str, Any] = {"hash_source": source, "attested_by": attester} + if source == HASH_SOURCE_CONSUMER_ATTESTED: + if not _text(attestation_evidence): + raise HashOnlyRegistrationError( + "A consumer_attested registration must record the consumer's " + "attestation evidence; pass --attestation-evidence." + ) + if not _text(verified_at): + raise HashOnlyRegistrationError( + "A consumer_attested registration must record when the checksum " + "was verified against the bytes; pass --verified-at." + ) + fields["attestation_evidence"] = str(attestation_evidence) + fields["verified_at"] = str(verified_at) + return fields + if pinned_from is None or not isinstance(pinned_from, Mapping): + raise HashOnlyRegistrationError( + "A consumer_pin registration must record where the pin was read: " + "pass --pinned-from-repository, --pinned-from-path and " + "--pinned-from-commit." + ) + errors = _pinned_from_errors(pinned_from) + if errors: + raise HashOnlyRegistrationError( + "A consumer_pin registration's pinned_from must carry the " + f"repository, path and a 40-hex commit: {', '.join(errors)}." + ) + if _text(verified_at): + raise HashOnlyRegistrationError( + "A consumer_pin registration carries no verified_at: Chronicle did " + "not verify the checksum against bytes, it transcribed the " + "consumer's pin. Record the pin's commit instead." + ) + fields["pinned_from"] = { + field: str(pinned_from[field]).strip() for field in PINNED_FROM_FIELDS + } + return fields + + +def _assert_no_archived_identity( + payload: Mapping[str, Any], + manifest_path: Path, + artifact_name: str, + access_class: str, +) -> None: + """Refuse to register hash-only a filename the manifest holds as public. + + A public entry may have been archived: its object sits in the raw bucket + under ``storage.r2`` (or in ``storage.previous_r2`` once revised). + Replacing that entry with a hash-only one would leave the bytes in a + Chronicle store with nothing recording them, and inventory would report + the tree clean. The transition is refused until the public entry, and the + object it names, have been explicitly removed. + """ + wanted = filename_key(artifact_name) + for key, _index, existing in iter_manifest_entries(payload): + if not isinstance(existing, Mapping): + continue + if filename_key(existing.get("filename")) != wanted: + continue + recorded = [ + str(block.get("uri") or block.get("key") or block) + for block in (recorded_r2(existing), *recorded_previous_r2(existing)) + if isinstance(block, Mapping) + ] + if recorded: + raise HashOnlyRegistrationError( + f"{manifest_path} records the R2 object(s) {recorded} for " + f"{existing.get('filename')!r} ({key!r}, " + f"access={safe_entry_access(existing)!r}). Registering it " + f"{access_class} would leave those bytes in a Chronicle store " + "with nothing recording them. Remove the object and its " + "storage record explicitly first; Chronicle will not reclassify " + "an archived release in place." + ) + if not is_hash_only(safe_entry_access(existing)): + raise HashOnlyRegistrationError( + f"{manifest_path} already registers {existing.get('filename')!r} " + f"({key!r}) as access={safe_entry_access(existing)!r}. A change " + f"of access class to {access_class!r} is an explicit decision: " + "remove the public entry by hand, then register the release." + ) + + def _assert_manifest_identity( payload: Mapping[str, Any], manifest_path: Path, @@ -553,11 +1127,11 @@ def _registration_entry(**values: Any) -> dict[str, Any]: return entry -def _existing_entries(files: Any, year: Any) -> list[Any]: - """Return the existing file entries for a year as a mutable list.""" +def _existing_entries(files: Any, key: Any) -> list[Any]: + """Return the existing file entries under a vintage key as a mutable list.""" if not isinstance(files, dict): return [] - spec = files.get(year) + spec = files.get(key) if spec is None: return [] if isinstance(spec, list): @@ -599,27 +1173,48 @@ def _dedupe(values: Iterable[str]) -> list[str]: "ACCESS_LICENSED", "ACCESS_PUBLIC", "ACCESS_RESTRICTED", + "AmbiguousVintageKeyError", + "ArtifactFilenameError", "ArtifactRegistrationReport", + "CHRONICLE_ATTESTER", "DEFAULT_ACCESS", "DEFAULT_MANIFEST_KIND", + "HASH_ONLY_HASH_SOURCES", + "HASH_SOURCES", + "HASH_SOURCE_CHRONICLE_FETCH", + "HASH_SOURCE_CONSUMER_ATTESTED", + "HASH_SOURCE_CONSUMER_PIN", "HashOnlyRegistrationError", "ListSpecRejected", "MANIFEST_KINDS", "MICRODATA_RELEASE_KIND", "ManifestAccessError", + "ManifestKindError", "MicrodataReleaseNotParseableError", + "PINNED_FROM_FIELDS", "PUBLISHER_TABLE_KIND", + "bare_filename", "entry_access", + "filename_key", + "is_bare_filename", "is_hash_only", "is_microdata_release", "iter_file_specs", + "iter_manifest_entries", "manifest_kind", "normalize_access", + "normalize_hash_source", + "recorded_previous_r2", "recorded_r2", + "records_r2_object", "register_hash_only_artifact", "registration_id", + "resolve_vintage_key", "safe_entry_access", "safe_manifest_kind", "stores_bytes", + "strict_entry_access", "validate_file_entry", + "validate_manifest_files", + "vintage_key_forms", ] From 44e1f8d21e6768f41072c4b6d0c80de3e39bc33a Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 01:59:13 -0400 Subject: [PATCH 35/42] Validate the manifest and the filename before a fetch touches anything fetch-artifact resolved the destination name only inside the read, compared filenames as raw strings, trusted an explicit --kind over the stored one, looked up only the integer year, and replaced list entries wholesale. Each of those let bytes past the hash-only boundary or lost a recorded identity. Now, before the publisher is read: the filename is a bare name (inferred from the URL without I/O when --filename is omitted), every alias of a hash-only registration is refused, a --kind that conflicts with the manifest is refused, a kindless manifest with content is refused, and the whole manifest is validated with the same codes inventory reports. The vintage entry is selected by key spelling and bare filename, so PR #226's recorded identity guard covers list-shaped release vintages, and _upsert_manifest replaces that entry in place, carries forward the fields a fetch does not own, never turns a publisher-table mapping into a list, and always writes kind. A public microdata release is archived only with --expected-sha256 and licence evidence bound to it: the fetched bytes are checked against the pin before anything is written or uploaded, --record-revision does not override that, and the bytes are staged outside the package tree and uploaded from there. The entry records hash_source chronicle_fetch, attested_by chronicle and the fetch date. publish-raw and inventory-artifacts report manifest-level defects, skip a manifest they cannot classify, and read release bytes from the staging directory only. The CLI gains the matching flags and prints refusals as errors instead of tracebacks. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 963 ++++++++++++++++++++++++++++++++++------- chronicle/harness.py | 279 ++++++++++-- 2 files changed, 1050 insertions(+), 192 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 6e450821..7b73e25e 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -5,7 +5,7 @@ import hashlib import json import mimetypes -from collections.abc import Iterator +from collections.abc import Mapping from dataclasses import dataclass from datetime import UTC, datetime from pathlib import Path @@ -27,19 +27,36 @@ ) from chronicle.env import env_value from chronicle.epoch import EMIT_EPOCH, Epoch, canonicalize_key, hash_domain +from chronicle.licences import ( + REDISTRIBUTABLE_LICENCES, + is_redistributable_licence, + licence_evidence_errors, +) from chronicle.registration import ( + ACCESS_CLASSES, ACCESS_PUBLIC, + CHRONICLE_ATTESTER, + HASH_SOURCE_CHRONICLE_FETCH, + MANIFEST_KINDS, MICRODATA_RELEASE_KIND, - manifest_kind as normalize_manifest_kind, + AmbiguousVintageKeyError, ListSpecRejected, ManifestAccessError, + ManifestKindError, + bare_filename, + filename_key, + is_bare_filename, is_hash_only, iter_file_specs, + iter_manifest_entries, + manifest_kind as normalize_manifest_kind, normalize_access, recorded_r2, + resolve_vintage_key, safe_entry_access, safe_manifest_kind, validate_file_entry, + validate_manifest_files, ) @@ -67,6 +84,46 @@ # name is an input, not a constant, wherever a caller addresses a package. DEFAULT_MANIFEST_FILENAME = "manifest.yaml" +#: Wrangler invocation used for R2 uploads unless a caller overrides it. +DEFAULT_WRANGLER_COMMAND = "npx wrangler" + +MICRODATA_STAGING_DIR_ENV = "CHRONICLE_MICRODATA_STAGING_DIR" +#: Where a public microdata release's bytes are staged before upload. Outside +#: the repository by construction: public microdata is archived in R2, never +#: committed beside its manifest (docs/adr-chronicle-raw-microdata-identity.md). +DEFAULT_MICRODATA_STAGING_DIR = ( + Path.home() / ".cache" / "policyengine-chronicle" / "microdata-staging" +) + + +def default_microdata_staging_dir() -> Path: + """Resolve the staging root: ``$CHRONICLE_MICRODATA_STAGING_DIR`` or default.""" + return Path( + env_value(MICRODATA_STAGING_DIR_ENV, default=DEFAULT_MICRODATA_STAGING_DIR) + ) + + +def microdata_staging_path( + *, + staging_dir: str | Path | None, + source_id: str, + package_id: str, + year: Any, + sha256: str, + filename: str, +) -> Path: + """Return the transient, content-addressed staging path for release bytes. + + Mirrors the raw R2 key shape so a staged file is addressed by the same + identity as the object it becomes. + """ + root = ( + Path(staging_dir) + if staging_dir is not None + else (default_microdata_staging_dir()) + ) + return root / source_id / package_id / str(year) / sha256 / Path(filename).name + def _manifest_path(output: Path, manifest_filename: str) -> Path: """Return the named manifest inside ``output``. @@ -204,6 +261,16 @@ class RecordedR2LocatorError(SourceArtifactManifestError): """ +class ExpectedArtifactIdentityError(SourceArtifactManifestError): + """Fetched bytes are not the bytes a reviewed pin said to expect. + + Distinct from :class:`SourceArtifactRevisionError`: a revision has an + opt-in (``--record-revision`` supersedes what the manifest records), an + expectation has none. The operator re-reviews the release and changes + ``--expected-sha256``; Chronicle never archives an unreviewed reissue. + """ + + # New UK and New Zealand uploads are namespaced by country. US objects predate # the country segment and deliberately keep their legacy ``raw/{source_id}`` # and ``derived/{source_id}`` shapes. Publisher directories are the stable @@ -656,11 +723,17 @@ def fetch_source_artifact( access: str = ACCESS_PUBLIC, licence: str | None = None, kind: str | None = None, + publisher: str | None = None, + vintage: str | None = None, + expected_sha256: str | None = None, + expected_size_bytes: int | None = None, + licence_evidence: Mapping[str, Any] | None = None, + staging_dir: str | Path | None = None, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, r2_prefix: str | None = None, - wrangler_command: str = "npx wrangler", + wrangler_command: str = DEFAULT_WRANGLER_COMMAND, ) -> ArtifactFetchReport: """Fetch/register a source artifact and optionally upload it to R2. @@ -676,10 +749,23 @@ def fetch_source_artifact( that disagree with the entry's recorded identity raise :class:`SourceArtifactRevisionError` before anything is overwritten. - Only ``public`` artifacts travel this path: it writes bytes into the package - tree and can upload them to the raw bucket. Licensed and restricted - artifacts are registered hash-only with + ``expected_sha256`` (and ``expected_size_bytes``) pin the bytes a reviewed + identity says the publisher serves: bytes that hash differently raise + :class:`ExpectedArtifactIdentityError` before anything is written or + uploaded, and ``record_revision`` does not override that. + + Only ``public`` artifacts travel this path: it writes bytes and can upload + them to the raw bucket. A publisher table's bytes land beside its + manifest; a public microdata release (``kind: microdata_release``) is + archived only under an allowlisted ``licence`` with ``licence_evidence`` + bound to ``expected_sha256``, and its bytes are staged in a transient + directory outside the package tree and uploaded from there. Licensed and + restricted artifacts are registered hash-only with :func:`chronicle.registration.register_hash_only_artifact`. + + Every refusal happens before the publisher is read; the checks that need + the bytes (expected identity, recorded identity) run before anything is + written. """ r2_bucket = r2_bucket or default_r2_raw_bucket() output = Path(output_dir) @@ -691,63 +777,125 @@ def fetch_source_artifact( "Register a licensed or restricted artifact by identity with " "`chronicle register-artifact`." ) + # The destination name is a pure function of the arguments, so it is + # resolved -- and any alias of a registered name refused -- before the + # publisher is read. Reading first and refusing afterwards would pull + # gated bytes to decide their name. + artifact_filename = bare_filename( + filename if filename is not None else _infer_artifact_filename(source_url), + what=( + "--filename" + if filename is not None + else "The filename inferred from the URL" + ), + ) + expected = _expected_identity(expected_sha256, expected_size_bytes) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, default_prefix=DEFAULT_R2_PREFIX, source_id=source_id, package_path=output, ) - # Read and validate the entry being written before anything is fetched: a - # manifest Chronicle cannot read, or a recorded block that names two - # different objects, is a refusal that need not touch the publisher. + + # Read and validate the manifest being written before anything is fetched: + # a manifest Chronicle cannot read, a manifest name that would sit beside + # the ones a package keeps, a recorded block that names two different + # objects, or a registration the fetch would overwrite are all refusals + # that need not touch the publisher. _refuse_a_stray_default_manifest(output, manifest_path) existing_manifest = _read_manifest(manifest_path) _manifest_files(existing_manifest, manifest_path) - recorded_identity = _recorded_identity( - _manifest_file_spec(existing_manifest, year), + # The byte boundary is checked first: overwriting a hash-only registration + # with bytes is the more serious refusal, and its message is the one the + # caller needs, not a prompt about the manifest's kind or licence. + _assert_no_hash_only_entry(existing_manifest, manifest_path, artifact_filename) + manifest_kind_value = _resolve_manifest_kind( + existing_manifest, manifest_path=manifest_path, - year=year, + requested_kind=kind, ) - licence_text = licence.strip() if isinstance(licence, str) else None - # A public microdata release is archived like any other public artifact, - # but its manifest still declares the release kind so a source package is - # refused and several files may share one vintage. - manifest_kind_value = ( - normalize_manifest_kind({"kind": kind}) - if kind is not None - else safe_manifest_kind(existing_manifest)[0] + _assert_manifest_valid_for_fetch( + existing_manifest, + manifest_path, + kind=manifest_kind_value, + package_dir=output, ) - # The byte boundary is checked first: overwriting a hash-only registration - # with bytes is the more serious refusal, and its message is the one the - # caller needs, not a prompt to supply a licence. - _assert_no_hash_only_entry(existing_manifest, manifest_path, year, filename) - if manifest_kind_value == MICRODATA_RELEASE_KIND and not licence_text: - raise ManifestAccessError( - f"{manifest_path} registers a microdata release, so every entry " - "must record its publisher licence; pass --licence." + licence_text = licence.strip() if isinstance(licence, str) else None + release = manifest_kind_value == MICRODATA_RELEASE_KIND + evidence: dict[str, str] | None = None + if release: + evidence = _release_fetch_evidence( + manifest_path, + existing_manifest, + filename=artifact_filename, + package_dir=output, + licence=licence_text, + vintage=vintage, + publisher=publisher, + expected=expected, + licence_evidence=licence_evidence, ) - fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() - content, inferred_filename = _read_artifact(source_url) - artifact_filename = filename or inferred_filename - if not artifact_filename: - raise ValueError("Could not infer artifact filename; pass --filename.") - _assert_no_hash_only_entry( + vintage_key, existing_value, selected_spec, _index = _select_vintage_entry( existing_manifest, - manifest_path, - year, - artifact_filename, + manifest_path=manifest_path, + year=year, + filename=artifact_filename, + kind=manifest_kind_value, + ) + recorded_identity = _recorded_identity( + selected_spec, + manifest_path=manifest_path, + year=vintage_key, + ) + _assert_table_vintage_is_revisable( + existing_value, + recorded_identity, + manifest_path=manifest_path, + year=vintage_key, + filename=artifact_filename, + release=release, ) + if ( + expected.sha256 + and recorded_identity is not None + and not record_revision + and not recorded_identity.holds( + sha256=expected.sha256, filename=artifact_filename + ) + ): + # The manifest already identifies other bytes: no download can satisfy + # both the recorded identity and the expectation, so refuse before it. + raise ExpectedArtifactIdentityError( + f"{manifest_path} entry {vintage_key!r} already records " + f"sha256={recorded_identity.sha256} " + f"filename={recorded_identity.filename or 'unknown'}, and the " + f"reviewed pin expects sha256={expected.sha256}. Re-review the pin, " + "or pass --record-revision together with the reviewed " + "--expected-sha256 to register the publisher revision." + ) + + fetched_at = datetime.now(UTC).replace(microsecond=0).isoformat() + content, _inferred_filename = _read_artifact(source_url) sha256 = hashlib.sha256(content).hexdigest() size_bytes = len(content) # Guard before the cached artifact is touched. A rejected fetch must leave # the recorded bytes and their manifest entry exactly as they were. + _assert_expected_identity( + expected, + manifest_path=manifest_path, + year=vintage_key, + filename=artifact_filename, + source_url=source_url, + sha256=sha256, + size_bytes=size_bytes, + ) _assert_recorded_identity_holds_these_bytes( recorded_identity, manifest_path=manifest_path, - year=year, + year=vintage_key, filename=artifact_filename, sha256=sha256, size_bytes=size_bytes, @@ -755,8 +903,21 @@ def fetch_source_artifact( record_revision=record_revision, ) - output.mkdir(parents=True, exist_ok=True) - local_path = output / artifact_filename + if release: + # Public microdata never lands in the package tree: it is staged in an + # untracked, transient directory and uploaded from there. + local_path = microdata_staging_path( + staging_dir=staging_dir, + source_id=source_id, + package_id=package_id, + year=year, + sha256=sha256, + filename=artifact_filename, + ) + local_path.parent.mkdir(parents=True, exist_ok=True) + else: + output.mkdir(parents=True, exist_ok=True) + local_path = output / artifact_filename local_path.write_bytes(content) r2_location = ArtifactStorageLocation( @@ -788,8 +949,9 @@ def fetch_source_artifact( source_id=source_id, package_id=package_id, dataset=dataset or f"{source_id}_{package_id}", - source_page=source_page or source_url, - table=table or package_id, + source_page=source_page, + table=table, + publisher=publisher, year=year, filename=artifact_filename, source_url=source_url, @@ -799,6 +961,9 @@ def fetch_source_artifact( access=access_class, licence=licence_text, kind=manifest_kind_value, + vintage=vintage, + licence_evidence=evidence, + expected=expected, r2_location=(r2_location if upload_r2 and r2_upload and r2_upload.ok else None), record_revision=record_revision, ) @@ -814,7 +979,7 @@ def fetch_source_artifact( sha256=sha256, size_bytes=size_bytes, fetched_at=fetched_at, - r2_location=r2_location if upload_r2 and r2_upload and r2_upload.ok else None, + r2_location=r2_location if upload_r2 else None, r2_upload=r2_upload, errors=tuple(errors), ) @@ -965,15 +1130,19 @@ def publish_source_artifacts( package_id: str | None = None, r2_bucket: str | None = None, r2_prefix: str | None = None, - wrangler_command: str = "npx wrangler", + wrangler_command: str = DEFAULT_WRANGLER_COMMAND, skip_hash_only: bool = False, + staging_dir: str | Path | None = None, ) -> RawArtifactPublishReport: """Upload manifest-declared raw source artifacts and record R2 locations. Only ``public`` artifacts are uploaded. A licensed or restricted entry is refused: no bytes are read or sent, and the entry carries a ``hash_only_access_refuses_bytes`` error unless ``skip_hash_only`` marks the - scan as deliberately mixed. + scan as deliberately mixed. A manifest that declares no kind (and is not + frozen kindless), or whose entries collide, is reported and skipped whole: + nothing under it is uploaded. A public microdata release's bytes are read + from the staging directory, never from beside the manifest. """ r2_bucket = r2_bucket or default_r2_raw_bucket() root_path = Path(root) @@ -1017,9 +1186,15 @@ def publish_source_artifacts( errors.append(f"Could not resolve R2 prefix for {manifest_path}: {exc}") continue - kind, kind_error = safe_manifest_kind(manifest) - if kind_error: - errors.append(f"{kind_error}: {manifest_path}") + kind, kind_error = safe_manifest_kind(manifest, manifest_path=manifest_path) + manifest_errors = [kind_error] if kind_error else [] + manifest_errors.extend(validate_manifest_files(manifest)) + if manifest_errors: + # Validate, then touch: a manifest Chronicle cannot classify or + # whose entries collide is reported and left alone; publishing any + # entry under it could ship bytes through the wrong record. + errors.extend(f"{code}: {manifest_path}" for code in manifest_errors) + continue updated = False for year, spec in files.items(): for file_spec in iter_file_specs(spec, kind=kind): @@ -1035,6 +1210,7 @@ def publish_source_artifacts( r2_prefix=resolved_r2_prefix, wrangler_command=wrangler_command, skip_hash_only=skip_hash_only, + staging_dir=staging_dir, ) entries.append(entry) if updated_spec is not None and isinstance(file_spec, dict): @@ -1100,8 +1276,15 @@ def inventory_source_artifacts( root: str | Path, *, manifest_filename: str = DEFAULT_MANIFEST_FILENAME, + staging_dir: str | Path | None = None, ) -> ArtifactInventoryReport: - """Inventory manifest-declared source artifacts under a root directory.""" + """Inventory manifest-declared source artifacts under a root directory. + + Reports every manifest-level defect (a missing or unknown kind, a vintage + under two key spellings, filename collisions) alongside the per-entry + codes, and treats a public microdata release's bytes as staged outside + the tree: a copy beside the manifest is an error, not an artifact. + """ root_path = Path(root) errors: list[str] = [] entries: list[ArtifactInventoryEntry] = [] @@ -1131,9 +1314,12 @@ def inventory_source_artifacts( if not isinstance(files, dict): errors.append(f"Manifest files must be a mapping: {manifest_path}") continue - kind, kind_error = safe_manifest_kind(manifest) + kind, kind_error = safe_manifest_kind(manifest, manifest_path=manifest_path) if kind_error: errors.append(f"{kind_error}: {manifest_path}") + errors.extend( + f"{code}: {manifest_path}" for code in validate_manifest_files(manifest) + ) for year, spec in files.items(): for file_spec in iter_file_specs(spec, kind=kind): entries.append( @@ -1143,6 +1329,7 @@ def inventory_source_artifacts( file_spec, manifest=manifest, kind=kind, + staging_dir=staging_dir, ) ) @@ -1449,13 +1636,346 @@ def _read_manifest(manifest_path: Path) -> dict[str, Any]: return payload -def _manifest_file_spec(payload: dict[str, Any], year: Any) -> dict[str, Any]: - """Return one manifest ``files`` entry, or an empty mapping.""" - files = payload.get("files") +def _infer_artifact_filename(source_url: str) -> str: + """Return the filename :func:`_read_artifact` would report, without I/O. + + The name is a pure function of the URL: the last path segment for http(s) + and ``file://`` URLs, the basename for a bare path. Resolving it before + the read lets every filename guard run before the publisher is touched. + """ + parsed = urlparse(source_url) + if parsed.scheme in ("http", "https"): + return _filename_from_url(source_url) + if parsed.scheme == "file": + return Path(unquote(parsed.path)).name + if not parsed.scheme: + return Path(source_url).name + raise ValueError(f"Unsupported source URL scheme: {parsed.scheme}") + + +@dataclass(frozen=True) +class ExpectedIdentity: + """The bytes a reviewed pin says the publisher serves, if any.""" + + sha256: str | None + size_bytes: int | None + + +def _expected_identity( + expected_sha256: str | None, + expected_size_bytes: int | None, +) -> ExpectedIdentity: + """Validate the expected-identity arguments before any I/O.""" + sha256 = expected_sha256.strip() if isinstance(expected_sha256, str) else None + if expected_sha256 is not None and ( + not sha256 or not _SHA256_KEY_SEGMENT.fullmatch(sha256) + ): + raise ExpectedArtifactIdentityError( + "--expected-sha256 must be a lowercase 64-character SHA-256 taken " + f"from a reviewed pin, not {expected_sha256!r}. Never invent a hash." + ) + if expected_size_bytes is not None and ( + isinstance(expected_size_bytes, bool) + or not isinstance(expected_size_bytes, int) + or expected_size_bytes <= 0 + ): + raise ExpectedArtifactIdentityError( + "--expected-size-bytes must be a positive integer, not " + f"{expected_size_bytes!r}." + ) + return ExpectedIdentity(sha256=sha256, size_bytes=expected_size_bytes) + + +def _assert_expected_identity( + expected: ExpectedIdentity, + *, + manifest_path: Path, + year: Any, + filename: str, + source_url: str, + sha256: str, + size_bytes: int, +) -> None: + """Refuse fetched bytes the reviewed pin does not cover. + + ``--record-revision`` does not override this: it governs what the manifest + records about bytes Chronicle chose to archive, and an expectation is the + statement that only reviewed bytes are archived at all. + """ + if expected.sha256 is None and expected.size_bytes is None: + return + if (expected.sha256 is None or expected.sha256 == sha256) and ( + expected.size_bytes is None or expected.size_bytes == size_bytes + ): + return + raise ExpectedArtifactIdentityError( + f"{manifest_path} entry {year!r} {filename}: the bytes served by " + f"{source_url} are not the bytes the reviewed pin covers. Expected " + f"sha256={expected.sha256 or 'unspecified'} " + f"size_bytes={expected.size_bytes or 'unspecified'}; fetched " + f"sha256={sha256} size_bytes={size_bytes}. The publisher is serving " + "bytes the pin does not describe. Chronicle will not archive an " + "unreviewed reissue; re-review the release and re-run with the pin you " + "reviewed." + ) + + +def _resolve_manifest_kind( + existing_manifest: dict[str, Any], + *, + manifest_path: Path, + requested_kind: str | None, +) -> str: + """Return the kind a fetch writes, refusing a conflict with the manifest. + + A manifest's kind is fixed once declared: an explicit ``--kind`` that + differs from it would reclassify every entry the manifest holds as a side + effect of one fetch. A kindless manifest with content is refused unless it + is frozen kindless byte for byte, and a stored kind Chronicle does not + recognise is never masked by the command line. + """ + requested = ( + normalize_manifest_kind({"kind": requested_kind}) + if requested_kind is not None + else None + ) + try: + stored = normalize_manifest_kind(existing_manifest, manifest_path=manifest_path) + except ManifestKindError as exc: + raise ManifestAccessError( + f"{exc} fetch-artifact will not add to a manifest whose kind it " + "cannot read; declare the kind by editing the manifest deliberately." + ) from exc + except ManifestAccessError as exc: + raise ManifestAccessError( + f"{manifest_path} declares an unknown manifest kind " + f"{existing_manifest.get('kind')!r}; expected one of " + f"{list(MANIFEST_KINDS)}. Fix the manifest before fetching into it." + ) from exc + if requested is None: + return stored + if existing_manifest and requested != stored: + raise ManifestAccessError( + f"{manifest_path} is a {stored} manifest; refusing to fetch into it " + f"as a {requested}. A manifest's kind is fixed once declared: " + f"register a {requested} in its own package directory, or migrate " + "this manifest deliberately." + ) + return requested + + +def _assert_manifest_valid_for_fetch( + manifest: dict[str, Any], + manifest_path: Path, + *, + kind: str, + package_dir: Path, +) -> None: + """Refuse to fetch into a manifest inventory would report as invalid. + + Uses the exact vocabulary ``inventory-artifacts`` and ``publish-raw`` + report, so a fetch never carries an invalid registration forward -- or + conceals one under a rewrite. + """ + codes: list[str] = list(validate_manifest_files(manifest)) + files = manifest.get("files") or {} + if isinstance(files, dict): + for key, spec in files.items(): + for file_spec in iter_file_specs(spec, kind=kind): + name = ( + file_spec.get("filename") if isinstance(file_spec, dict) else None + ) + exists = ( + bool(name) + and is_bare_filename(name) + and (package_dir / str(name)).exists() + ) + for code in validate_file_entry( + file_spec, + kind=kind, + manifest=manifest, + local_file_exists=exists, + ): + codes.append(f"{key!r}/{name}: {code}") + if codes: + raise ManifestAccessError( + f"{manifest_path} is not a valid {kind} manifest: " + f"{'; '.join(codes)}. fetch-artifact rewrites this manifest and " + "will not carry an invalid registration forward; " + "inventory-artifacts reports the same codes." + ) + + +def _release_fetch_evidence( + manifest_path: Path, + existing_manifest: dict[str, Any], + *, + filename: str, + package_dir: Path, + licence: str | None, + vintage: str | None, + publisher: str | None, + expected: ExpectedIdentity, + licence_evidence: Mapping[str, Any] | None, +) -> dict[str, str]: + """Return the ``licence_evidence`` block a public release fetch records. + + Bytes only with artifact-bound redistribution evidence: the licence must + be on the allowlist, the evidence must bind this artifact (by the reviewed + SHA-256) to that term, and the release must carry its publisher and + vintage. Public microdata is staged outside the package tree, so a file of + that name beside the manifest is refused as tracked microdata bytes. + """ + if not licence: + raise ManifestAccessError( + f"{manifest_path} registers a microdata release, so every entry " + "must record its publisher licence; pass --licence." + ) + if not (vintage and vintage.strip()): + raise ManifestAccessError( + f"{manifest_path} registers a microdata release, so every entry " + "must record its publisher vintage; pass --vintage." + ) + if not (publisher or existing_manifest.get("publisher")): + raise ManifestAccessError( + f"{manifest_path} registers a microdata release, so it must name " + "the publisher; pass --publisher." + ) + if expected.sha256 is None: + raise ManifestAccessError( + "A public microdata release is archived only against a reviewed " + "checksum that its licence evidence covers; pass --expected-sha256." + ) + if not is_redistributable_licence(licence): + raise ManifestAccessError( + f"licence {licence!r} is not on Chronicle's allowlist of " + f"redistributable terms {sorted(REDISTRIBUTABLE_LICENCES)}. A " + "public-download file without redistribution evidence is classed " + "licensed: register it hash-only with `chronicle register-artifact`." + ) + supplied = dict(licence_evidence or {}) + evidence = { + "issuer": str(supplied.get("issuer") or "").strip(), + "licence": licence, + "scope": str(supplied.get("scope") or "").strip(), + "url": str(supplied.get("url") or "").strip(), + "sha256": expected.sha256, + } + codes = licence_evidence_errors(evidence, licence=licence, sha256=expected.sha256) + if codes: + raise ManifestAccessError( + f"{manifest_path}: archiving {filename!r} needs licence evidence " + f"binding it to {licence!r}: {', '.join(codes)}. Pass " + "--licence-evidence-issuer, --licence-evidence-scope and a durable " + "--licence-evidence-url; the evidence covers --expected-sha256." + ) + if (package_dir / filename).exists(): + raise ManifestAccessError( + f"{package_dir / filename} exists beside the manifest. Public " + "microdata bytes are staged outside the package tree and uploaded " + "from there; a repository never holds them. Remove the file first." + ) + return evidence + + +def _select_vintage_entry( + payload: dict[str, Any], + *, + manifest_path: Path, + year: Any, + filename: str, + kind: str, +) -> tuple[Any, Any, dict[str, Any], int | None]: + """Locate the entry a fetch revises: ``(key, files[key], entry, index)``. + + The vintage key is whichever spelling the manifest already uses for + ``year`` (``2023`` and ``'2023'`` are one vintage). A publisher table's + vintage is one mapping, and that entry is its identity whatever filename + it records. A microdata release lists several files under one vintage, and + the entry is the one whose bare filename matches. ``index`` is the entry's + position in that list, or None. + """ + files = payload.get("files") if isinstance(payload, dict) else None + if files is None: + return year, None, {}, None if not isinstance(files, dict): - return {} - spec = files.get(year) - return spec if isinstance(spec, dict) else {} + raise MalformedManifestError( + f"{manifest_path} files must be a mapping; it is a " + f"{type(files).__name__}. Chronicle will not write into a manifest " + "it cannot read." + ) + try: + key = resolve_vintage_key(files, year) + except AmbiguousVintageKeyError as exc: + raise MalformedManifestError( + f"{manifest_path}: {exc} Chronicle will not choose which entry is " + "the record." + ) from exc + if key is None: + return year, None, {}, None + existing = files[key] + if isinstance(existing, dict): + return key, existing, existing, None + if isinstance(existing, list): + if kind != MICRODATA_RELEASE_KIND: + raise MalformedManifestError( + f"{manifest_path} entry {key!r} lists {len(existing)} files, but " + f"the manifest is a {kind} manifest. Only a kind: " + "microdata_release manifest may list several files under one " + "vintage (list_file_spec_requires_microdata_release_kind); " + "Chronicle will not add to or revise a vintage it cannot " + "validate." + ) + wanted = filename_key(filename) + matches = [ + (index, entry) + for index, entry in enumerate(existing) + if isinstance(entry, dict) and filename_key(entry.get("filename")) == wanted + ] + if len(matches) > 1: + raise MalformedManifestError( + f"{manifest_path} entry {key!r} lists {len(matches)} entries " + f"named {filename!r}; a fetch can revise exactly one, and " + "Chronicle will not guess which." + ) + if matches: + return key, existing, matches[0][1], matches[0][0] + return key, existing, {}, None + raise MalformedManifestError( + f"{manifest_path} entry {key!r} must be a mapping or a list of " + f"mappings; it is a {type(existing).__name__}." + ) + + +def _assert_table_vintage_is_revisable( + existing_value: Any, + recorded_identity: RecordedIdentity | None, + *, + manifest_path: Path, + year: Any, + filename: str, + release: bool, +) -> None: + """Refuse a second filename in a publisher-table vintage with no identity. + + A publisher table holds one file per vintage. When its entry identifies + bytes, a fetch under another name is judged by the recorded identity (and + ``--record-revision`` supersedes the entry). An entry that identifies + nothing -- no ``sha256``, no recorded object -- cannot be superseded, and + adding a second file would turn the mapping into a list every reader + refuses. + """ + if release or not isinstance(existing_value, dict) or recorded_identity: + return + recorded_name = existing_value.get("filename") + if recorded_name is None or filename_key(recorded_name) == filename_key(filename): + return + raise MalformedManifestError( + f"{manifest_path} entry {year!r} records {recorded_name!r} without a " + f"sha256; a publisher table holds one file per vintage, so {filename!r} " + "cannot be added and there is no recorded identity to supersede. " + "Register the vintage's bytes first or fix the entry by hand." + ) def _recorded_storage(spec: Any) -> dict[str, Any]: @@ -1771,14 +2291,37 @@ def _superseding_storage( return storage +#: Entry fields a fetch owns. Everything else an entry already records -- +#: notes, doi, study, access_route, a vintage the fetch did not restate -- is +#: carried forward when the fetch replaces that entry, so a re-fetch is never a +#: silent de-registration. +_FETCH_OWNED_FIELDS: frozenset[str] = frozenset( + { + "filename", + "source_url", + "access", + "licence", + "licence_evidence", + "sha256", + "size_bytes", + "fetched_at", + "verified_at", + "hash_source", + "attested_by", + "storage", + } +) + + def _upsert_manifest( manifest_path: Path, *, source_id: str, package_id: str, dataset: str, - source_page: str, - table: str, + source_page: str | None, + table: str | None, + publisher: str | None, year: int, filename: str, source_url: str, @@ -1788,43 +2331,58 @@ def _upsert_manifest( access: str, licence: str | None, kind: str, + vintage: str | None, + licence_evidence: Mapping[str, Any] | None, + expected: ExpectedIdentity, r2_location: ArtifactStorageLocation | None, record_revision: bool = False, ) -> None: + """Write one fetched entry into its manifest, in place. + + The guards fetch_source_artifact ran are repeated against the freshly + re-read manifest, so no caller can reach a false-provenance write by + another route. The entry the fetch revises is located by vintage key and + bare filename and replaced where it sits; a publisher-table vintage stays + one mapping and a release vintage stays a list, and every manifest this + command touches declares its kind. + """ payload = _read_manifest(manifest_path) + kind = _resolve_manifest_kind( + payload, manifest_path=manifest_path, requested_kind=kind + ) + _assert_no_hash_only_entry(payload, manifest_path, filename) payload.setdefault("source_id", source_id) payload.setdefault("package_id", package_id) + payload = _with_declared_kind(payload, kind) payload.setdefault("dataset", dataset) + if publisher: + payload.setdefault("publisher", publisher) payload.setdefault("source_page", source_page) payload.setdefault("table", table) if payload.get("files") is None: # setdefault keeps an explicit null (a bare ``files:`` line); the # entry below needs a mapping to record into. payload["files"] = {} - # Access is written explicitly on every entry this command touches, so a - # manifest never relies on the inferred ``public`` default once rewritten. - file_entry: dict[str, Any] = { - "filename": filename, - "source_url": source_url, - "access": access, - } - if licence: - file_entry["licence"] = licence - file_entry.update( - { - "sha256": sha256, - "size_bytes": size_bytes, - "fetched_at": fetched_at, - } - ) - if kind == MICRODATA_RELEASE_KIND: - payload["kind"] = kind - existing = ( - payload["files"].get(year) if isinstance(payload["files"], dict) else None + release = kind == MICRODATA_RELEASE_KIND + + key, existing_value, recorded_spec, index = _select_vintage_entry( + payload, + manifest_path=manifest_path, + year=year, + filename=filename, + kind=kind, ) - recorded_spec = _manifest_file_spec(payload, year) recorded_storage = _recorded_storage(recorded_spec) - identity = _recorded_identity(recorded_spec, manifest_path=manifest_path, year=year) + identity = _recorded_identity(recorded_spec, manifest_path=manifest_path, year=key) + _assert_expected_identity( + expected, + manifest_path=manifest_path, + year=key, + filename=filename, + source_url=source_url, + sha256=sha256, + size_bytes=size_bytes, + ) new_r2 = r2_location.to_dict() if r2_location is not None else None holds = identity is not None and identity.holds(sha256=sha256, filename=filename) if identity is not None and not holds and not record_revision: @@ -1834,7 +2392,7 @@ def _upsert_manifest( raise SourceArtifactRevisionError( _revision_error_message( manifest_path=manifest_path, - year=year, + year=key, filename=filename, identity=identity, sha256=sha256, @@ -1859,88 +2417,131 @@ def _upsert_manifest( storage = {**recorded_storage, "r2": new_r2} else: storage = dict(recorded_storage) + + # Access is written explicitly on every entry this command touches, so a + # manifest never relies on the inferred ``public`` default once rewritten. + file_entry: dict[str, Any] = {"filename": filename} + if release: + file_entry["access"] = access + file_entry["licence"] = licence + if licence_evidence: + file_entry["licence_evidence"] = dict(licence_evidence) + file_entry["vintage"] = vintage or recorded_spec.get("vintage") + file_entry["sha256"] = sha256 + file_entry["size_bytes"] = size_bytes + file_entry["source_url"] = source_url + file_entry["fetched_at"] = fetched_at + # Chronicle fetched and hashed these bytes itself. + file_entry["verified_at"] = fetched_at[:10] + file_entry["hash_source"] = HASH_SOURCE_CHRONICLE_FETCH + file_entry["attested_by"] = CHRONICLE_ATTESTER + else: + file_entry["source_url"] = source_url + file_entry["access"] = access + if licence: + file_entry["licence"] = licence + if vintage: + file_entry["vintage"] = vintage + file_entry["sha256"] = sha256 + file_entry["size_bytes"] = size_bytes + file_entry["fetched_at"] = fetched_at + for field, value in recorded_spec.items(): + if field not in _FETCH_OWNED_FIELDS and field not in file_entry: + file_entry[field] = value # An entry that has no storage to record carries no empty block: a # revision over a never-published entry supersedes nothing. if storage: file_entry["storage"] = storage - entries = list(existing) if isinstance(existing, list) else _as_entry_list(existing) - # A registration is a durable statement, so a fetch replaces only the entry - # for its own filename and never silently drops another one. Overwriting a - # hash-only registration is refused outright by _assert_no_hash_only_entry. - kept = [ - entry - for entry in entries - if not (isinstance(entry, dict) and entry.get("filename") == filename) - ] - if not kept and not isinstance(existing, list) and kind != MICRODATA_RELEASE_KIND: - # A single-file publisher table keeps the historical mapping shape. - payload["files"][year] = file_entry + + if existing_value is None: + payload["files"][key] = [file_entry] if release else file_entry + elif isinstance(existing_value, dict): + if not release: + # A publisher table holds one file per vintage; a rename under + # --record-revision supersedes that one entry, and the superseded + # key in storage.previous_r2 keeps the old name. + payload["files"][key] = file_entry + elif filename_key(existing_value.get("filename")) == filename_key(filename): + payload["files"][key] = [file_entry] + else: + payload["files"][key] = [existing_value, file_entry] else: - payload["files"][year] = [*kept, file_entry] + entries = list(existing_value) + if index is not None: + entries[index] = file_entry + else: + entries.append(file_entry) + payload["files"][key] = entries + # A release's bytes never enter the package directory, so the directory + # may not exist yet when its manifest is first written. + manifest_path.parent.mkdir(parents=True, exist_ok=True) manifest_path.write_text( yaml.safe_dump(payload, sort_keys=False, allow_unicode=True), encoding="utf-8", ) -def _as_entry_list(existing: Any) -> list[Any]: - """Return an existing ``files[year]`` value as a list of entries.""" - if existing is None: - return [] - return [existing] - - -def _load_manifest_payload(manifest_path: Path) -> dict[str, Any]: - """Load a manifest mapping, or an empty mapping when absent or unreadable.""" - if not manifest_path.exists(): - return {} - payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} - if not isinstance(payload, dict): - raise ValueError(f"Manifest must be a mapping: {manifest_path}") - return payload - - -def _iter_manifest_entries(manifest: dict[str, Any]) -> Iterator[tuple[Any, Any]]: - """Yield every ``(year, entry)`` a manifest declares, whatever its shape. - - Deliberately not gated on the manifest ``kind``: a guard must see the - entries a manifest actually holds, including a list under a manifest whose - kind is absent or misspelled. Reporting that shape as malformed is the - validator's job, not the guard's. - """ - files = manifest.get("files") - if not isinstance(files, dict): - return - for year, spec in files.items(): - for entry in spec if isinstance(spec, list) else (spec,): - yield year, entry +def _with_declared_kind(payload: dict[str, Any], kind: str) -> dict[str, Any]: + """Return ``payload`` declaring ``kind``, placed after its identity keys.""" + if "kind" in payload: + payload["kind"] = kind + return payload + ordered: dict[str, Any] = {} + inserted = False + for field, value in payload.items(): + ordered[field] = value + if field == "package_id" and not inserted: + ordered["kind"] = kind + inserted = True + if not inserted: + ordered["kind"] = kind + return ordered def _assert_no_hash_only_entry( manifest: dict[str, Any], manifest_path: Path, - year: Any, - filename: str | None, + filename: str, ) -> None: """Refuse to fetch bytes over an existing hash-only registration. The write target is a path in the package directory, so the search spans every vintage rather than the requested one: a licensed release registered - under one year must not be fetched into the tree under another. + under one year must not be fetched into the tree under another. Names are + compared as resolved, case-folded bare filenames, so no alias of a + registered name -- ``./adult.tab``, ``ADULT.TAB`` -- slips past. """ - if not filename: - return - for entry_year, spec in _iter_manifest_entries(manifest): - if not isinstance(spec, dict): + wanted = filename_key(filename) + for key, _index, spec in iter_manifest_entries(manifest): + if not isinstance(spec, dict) or spec.get("filename") is None: + continue + if filename_key(spec.get("filename")) != wanted: continue - if spec.get("filename") != filename: + declared = spec.get("access") + if declared is None: + # A release entry without an access class is reported by the + # strict manifest validation that follows; it is never read as + # public here. continue - access = safe_entry_access(spec) + try: + access = normalize_access(declared) + except ManifestAccessError: + raise ManifestAccessError( + f"{manifest_path} registers {spec.get('filename')!r} for " + f"{key!r} with access={declared!r}, which is not one of " + f"{list(ACCESS_CLASSES)}. Its bytes must not enter a Chronicle " + "store until the registration is fixed." + ) from None if is_hash_only(access): + requested = ( + f" (requested as {filename!r})" + if spec.get("filename") != filename + else "" + ) raise ManifestAccessError( - f"{manifest_path} registers {filename!r} for {entry_year} as " - f"access={access!r}. Its bytes must not enter a Chronicle " - "store; keep the hash-only registration." + f"{manifest_path} registers {spec.get('filename')!r}{requested} " + f"for {key!r} as access={access!r}. Its bytes must not enter a " + "Chronicle store; keep the hash-only registration." ) @@ -1979,6 +2580,7 @@ def _publish_raw_manifest_entry( r2_prefix: str, wrangler_command: str, skip_hash_only: bool = False, + staging_dir: str | Path | None = None, ) -> tuple[RawArtifactPublishEntry, dict[str, Any] | None]: errors: list[str] = [] if isinstance(spec, ListSpecRejected): @@ -2002,6 +2604,27 @@ def _publish_raw_manifest_entry( spec = {} errors.append("malformed_file_spec") filename = str(spec.get("filename") or "") + if filename and not is_bare_filename(filename): + # Refuse before resolving the path: a name that is not bare could + # address a file outside the package directory, or one another entry + # already governs. + return ( + RawArtifactPublishEntry( + manifest_path=str(manifest_path), + source_id=source_id, + package_id=package_id, + year=str(year), + filename=filename, + local_path=str(manifest_path.parent), + sha256=None, + size_bytes=None, + r2_location=None, + upload=None, + errors=(f"non_canonical_filename:{filename}",), + ), + None, + ) + kind = kind or safe_manifest_kind(manifest, manifest_path=manifest_path)[0] access = safe_entry_access(spec) if is_hash_only(access): # Refuse before touching bytes: no Chronicle store holds a licensed or @@ -2011,7 +2634,7 @@ def _publish_raw_manifest_entry( hash_only_errors = list( validate_file_entry( spec, - kind=kind or safe_manifest_kind(manifest)[0], + kind=kind, manifest=manifest, local_file_exists=(manifest_path.parent / filename).exists() if filename @@ -2040,21 +2663,35 @@ def _publish_raw_manifest_entry( errors.extend( validate_file_entry( spec, - kind=kind or safe_manifest_kind(manifest)[0], + kind=kind, manifest=manifest, local_file_exists=(manifest_path.parent / filename).exists() if filename else False, ) ) - artifact_path = manifest_path.parent / filename + release = kind == MICRODATA_RELEASE_KIND sha256_expected = spec.get("sha256") + if release and filename and sha256_expected: + # Public microdata is never read from beside its manifest: its bytes + # are staged outside the tree (validate_file_entry reports a copy in + # the tree as bytes_present_for_microdata_release_entry). + artifact_path = microdata_staging_path( + staging_dir=staging_dir, + source_id=source_id, + package_id=package_id, + year=year, + sha256=str(sha256_expected), + filename=filename, + ) + else: + artifact_path = manifest_path.parent / filename sha256_actual = None size_bytes = None if not filename: errors.append("missing_filename") elif not artifact_path.exists(): - errors.append("missing_file") + errors.append("staged_bytes_missing" if release else "missing_file") else: content = artifact_path.read_bytes() sha256_actual = hashlib.sha256(content).hexdigest() @@ -2209,6 +2846,7 @@ def _inventory_entry( *, manifest: dict[str, Any] | None = None, kind: str | None = None, + staging_dir: str | Path | None = None, ) -> ArtifactInventoryEntry: errors: list[str] = [] original_spec = spec @@ -2230,27 +2868,60 @@ def _inventory_entry( spec = {} errors.append("malformed_file_spec") filename = str(spec.get("filename") or "") - artifact_path = manifest_path.parent / filename - exists = bool(filename) and artifact_path.exists() + kind = kind or safe_manifest_kind(manifest, manifest_path=manifest_path)[0] + bare = bool(filename) and is_bare_filename(filename) + # A name that is not bare is reported by validate_file_entry and never + # resolved to a path, which could lie outside the package directory. + in_tree = bare and (manifest_path.parent / filename).exists() access = safe_entry_access(spec) hash_only = is_hash_only(access) + release = kind == MICRODATA_RELEASE_KIND errors.extend( validate_file_entry( original_spec, - kind=kind or safe_manifest_kind(manifest)[0], + kind=kind, manifest=manifest, - local_file_exists=exists, + local_file_exists=in_tree, ) ) sha256_expected = spec.get("sha256") + if release and not hash_only and bare and sha256_expected: + artifact_path = microdata_staging_path( + staging_dir=staging_dir, + source_id=str((manifest or {}).get("source_id") or ""), + package_id=str((manifest or {}).get("package_id") or ""), + year=year, + sha256=str(sha256_expected), + filename=filename, + ) + exists = artifact_path.exists() + else: + artifact_path = ( + manifest_path.parent / filename if bare else manifest_path.parent + ) + exists = in_tree sha256_actual = None - size_bytes = spec.get("size_bytes") if hash_only else None + size_bytes = spec.get("size_bytes") if hash_only or release else None if not filename: errors.append("missing_filename") + elif not bare: + pass elif hash_only: # A licensed or restricted registration is identity only: Chronicle # never holds the bytes, so a missing local file is the correct state. pass + elif release: + # A public release is archived, not committed: its registration is + # complete once the raw bucket records the object. Staged bytes are + # transient and checked when present. + if recorded_r2(spec) is None: + errors.append("r2_object_not_recorded") + if exists: + content = artifact_path.read_bytes() + sha256_actual = hashlib.sha256(content).hexdigest() + size_bytes = len(content) + if sha256_expected and sha256_actual != sha256_expected: + errors.append("checksum_mismatch") elif not exists: errors.append("missing_file") else: diff --git a/chronicle/harness.py b/chronicle/harness.py index 8ce68471..b450bfa0 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -53,8 +53,12 @@ ) from chronicle.registration import ( ACCESS_CLASSES, + HASH_ONLY_HASH_SOURCES, MANIFEST_KINDS, ArtifactRegistrationReport, + HashOnlyRegistrationError, + ManifestAccessError, + MicrodataReleaseNotParseableError, register_hash_only_artifact, ) from chronicle.sources.cells import ( @@ -350,6 +354,12 @@ def fetch_artifact_file( access: str = "public", licence: str | None = None, kind: str | None = None, + publisher: str | None = None, + vintage: str | None = None, + expected_sha256: str | None = None, + expected_size_bytes: int | None = None, + licence_evidence: dict[str, str] | None = None, + staging_dir: str | Path | None = None, upload_r2: bool = False, record_revision: bool = False, r2_bucket: str | None = None, @@ -377,6 +387,12 @@ def fetch_artifact_file( access=access, licence=licence, kind=kind, + publisher=publisher, + vintage=vintage, + expected_sha256=expected_sha256, + expected_size_bytes=expected_size_bytes, + licence_evidence=licence_evidence, + staging_dir=staging_dir, upload_r2=upload_r2, record_revision=record_revision, r2_bucket=r2_bucket, @@ -396,6 +412,11 @@ def register_artifact_file( licence: str, access: str, vintage: str, + hash_source: str, + attested_by: str, + attestation_evidence: str | None = None, + pinned_from: dict[str, str] | None = None, + verified_at: str | None = None, size_bytes: int | None = None, source_page: str | None = None, source_url: str | None = None, @@ -406,8 +427,6 @@ def register_artifact_file( table: str | None = None, publisher: str | None = None, fetched_at: str | None = None, - verified_at: str | None = None, - hash_source: str | None = None, notes: str | None = None, allow_reissue: bool = False, ) -> ArtifactRegistrationReport: @@ -422,6 +441,11 @@ def register_artifact_file( licence=licence, access=access, vintage=vintage, + hash_source=hash_source, + attested_by=attested_by, + attestation_evidence=attestation_evidence, + pinned_from=pinned_from, + verified_at=verified_at, size_bytes=size_bytes, source_page=source_page, source_url=source_url, @@ -432,8 +456,6 @@ def register_artifact_file( table=table, publisher=publisher, fetched_at=fetched_at, - verified_at=verified_at, - hash_source=hash_source, notes=notes, allow_reissue=allow_reissue, ) @@ -443,9 +465,14 @@ def inventory_artifact_files( root: str | Path, *, manifest_filename: str = "manifest.yaml", + staging_dir: str | Path | None = None, ) -> ArtifactInventoryReport: """Inventory local manifest-declared source artifacts.""" - return inventory_source_artifacts(root, manifest_filename=manifest_filename) + return inventory_source_artifacts( + root, + manifest_filename=manifest_filename, + staging_dir=staging_dir, + ) def publish_raw_artifact_files( @@ -458,6 +485,7 @@ def publish_raw_artifact_files( r2_prefix: str | None = None, wrangler_command: str = "npx wrangler", skip_hash_only: bool = False, + staging_dir: str | Path | None = None, ) -> RawArtifactPublishReport: """Publish manifest-declared raw source artifacts to R2.""" return publish_source_artifacts( @@ -469,6 +497,7 @@ def publish_raw_artifact_files( r2_prefix=r2_prefix, wrangler_command=wrangler_command, skip_hash_only=skip_hash_only, + staging_dir=staging_dir, ) @@ -572,6 +601,30 @@ def plan_pe_source_files( return report +def _licence_evidence_arguments(args: argparse.Namespace) -> dict[str, str] | None: + """Collect the licence-evidence flags, or None when none was passed.""" + evidence = { + "issuer": args.licence_evidence_issuer, + "scope": args.licence_evidence_scope, + "url": args.licence_evidence_url, + } + if all(value is None for value in evidence.values()): + return None + return {key: value for key, value in evidence.items() if value is not None} + + +def _pinned_from_arguments(args: argparse.Namespace) -> dict[str, str] | None: + """Collect the pinned-from flags, or None when none was passed.""" + pinned = { + "repository": args.pinned_from_repository, + "path": args.pinned_from_path, + "commit": args.pinned_from_commit, + } + if all(value is None for value in pinned.values()): + return None + return {key: value for key, value in pinned.items() if value is not None} + + def main(argv: list[str] | None = None) -> int: """Run the harness CLI.""" parser = argparse.ArgumentParser(description="Chronicle fact validation harness") @@ -977,7 +1030,10 @@ def main(argv: list[str] | None = None) -> int: ) artifact_parser.add_argument( "--filename", - help="Override artifact filename inferred from URL/path.", + help=( + "Override the artifact filename inferred from the URL. Must be a " + "bare filename; the artifact always lands under that name." + ), ) artifact_parser.add_argument( "--access", @@ -1002,8 +1058,64 @@ def main(argv: list[str] | None = None) -> int: help=( "Manifest kind to declare. Pass microdata_release to archive a " "public-use microdata release, which may hold several files under " - "one vintage and is never parsed by a source package. Defaults to " - "the existing manifest's kind, or publisher_table." + "one vintage and is never parsed by a source package. Must match " + "the existing manifest's kind; a conflicting kind is refused. " + "Omit to inherit it (publisher_table for a new manifest)." + ), + ) + artifact_parser.add_argument( + "--publisher", + help="Publishing body. Required for a microdata release.", + ) + artifact_parser.add_argument( + "--vintage", + help="Publisher vintage label. Required for a microdata release.", + ) + artifact_parser.add_argument( + "--expected-sha256", + help=( + "Lowercase 64-character SHA-256 the fetched bytes must have, from " + "a reviewed pin. The fetch refuses, before writing or uploading, " + "bytes that hash differently; --record-revision does not override " + "it. Required for a microdata release: the licence evidence covers " + "this checksum. Never invent one." + ), + ) + artifact_parser.add_argument( + "--expected-size-bytes", + type=int, + help="Size the fetched bytes must have, from the same reviewed pin.", + ) + artifact_parser.add_argument( + "--licence-evidence-issuer", + help=( + "Who issued the file under the allowlisted --licence. Required " + "for a microdata release." + ), + ) + artifact_parser.add_argument( + "--licence-evidence-scope", + help=( + "Statement of what the evidence covers, such as 'public-use file " + "of a federal agency'. Required for a microdata release." + ), + ) + artifact_parser.add_argument( + "--licence-evidence-url", + help=( + "Durable http(s) URL of the publisher's evidence that this file is " + "issued under --licence. Required for a microdata release." + ), + ) + artifact_parser.add_argument( + "--staging-dir", + type=Path, + default=None, + help=( + "Transient directory where a microdata release's bytes are staged " + "before upload, outside the repository. Defaults to " + "$CHRONICLE_MICRODATA_STAGING_DIR, else " + "~/.cache/policyengine-chronicle/microdata-staging." ), ) artifact_parser.add_argument( @@ -1144,13 +1256,46 @@ def main(argv: list[str] | None = None) -> int: "--fetched-at", help="When the authorized environment fetched the bytes, if known.", ) + registration_parser.add_argument( + "--hash-source", + required=True, + choices=list(HASH_ONLY_HASH_SOURCES), + help=( + "How the checksum is known: consumer_attested (the consumer " + "verified it against bytes it holds; pass --attested-by, " + "--attestation-evidence and --verified-at) or consumer_pin (it is " + "transcribed from a reviewed pin in the consumer's repository; pass " + "--attested-by and the three --pinned-from-* fields, and no " + "--verified-at)." + ), + ) + registration_parser.add_argument( + "--attested-by", + required=True, + help="The consumer that attests the checksum, such as PolicyEngine/microcosm.", + ) + registration_parser.add_argument( + "--attestation-evidence", + help="The consumer's evidence for a consumer_attested checksum.", + ) registration_parser.add_argument( "--verified-at", - help="When this checksum was verified against the reviewed pin.", + help=( + "When the consumer verified the checksum against the bytes " + "(consumer_attested only)." + ), ) registration_parser.add_argument( - "--hash-source", - help="Where the checksum came from, such as a consumer source manifest.", + "--pinned-from-repository", + help="Repository holding the consumer's pin, such as PolicyEngine/microcosm.", + ) + registration_parser.add_argument( + "--pinned-from-path", + help="Path of the pin inside that repository.", + ) + registration_parser.add_argument( + "--pinned-from-commit", + help="40-hex commit the pin was read from.", ) registration_parser.add_argument( "--notes", @@ -1180,6 +1325,16 @@ def main(argv: list[str] | None = None) -> int: default="manifest.yaml", help="Manifest filename to scan for.", ) + artifact_inventory_parser.add_argument( + "--staging-dir", + type=Path, + default=None, + help=( + "Where public microdata releases are staged. Defaults to " + "$CHRONICLE_MICRODATA_STAGING_DIR, else " + "~/.cache/policyengine-chronicle/microdata-staging." + ), + ) raw_publish_parser = subparsers.add_parser( "publish-raw", @@ -1204,6 +1359,17 @@ def main(argv: list[str] | None = None) -> int: "skipped rather than refused, so a mixed tree can be published." ), ) + raw_publish_parser.add_argument( + "--staging-dir", + type=Path, + default=None, + help=( + "Where public microdata releases are staged; their bytes are read " + "from there, never from beside the manifest. Defaults to " + "$CHRONICLE_MICRODATA_STAGING_DIR, else " + "~/.cache/policyengine-chronicle/microdata-staging." + ), + ) raw_publish_parser.add_argument( "--source-id", help="Override manifest source_id for scanned artifacts.", @@ -1535,15 +1701,21 @@ def main(argv: list[str] | None = None) -> int: return 0 if args.command == "build-suite": axiom_command = shlex.split(args.axiom_cli) if args.axiom_cli else None - report = build_source_suite_dir( - args.source, - args.out, - year=args.year, - axiom_command=axiom_command, - axiom_roots=args.axiom_root, - require_axiom_validation=args.require_axiom_validation, - replace=args.replace, - ) + try: + report = build_source_suite_dir( + args.source, + args.out, + year=args.year, + axiom_command=axiom_command, + axiom_roots=args.axiom_root, + require_axiom_validation=args.require_axiom_validation, + replace=args.replace, + ) + except (ManifestAccessError, MicrodataReleaseNotParseableError) as error: + # Refused before the output directory was touched: no source + # package parses a microdata release or a hash-only entry. + print(f"error: {error}", file=sys.stderr) + return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 if args.command == "build-bundle": @@ -1603,49 +1775,63 @@ def main(argv: list[str] | None = None) -> int: access=args.access, licence=args.licence, kind=args.kind, + publisher=args.publisher, + vintage=args.vintage, + expected_sha256=args.expected_sha256, + expected_size_bytes=args.expected_size_bytes, + licence_evidence=_licence_evidence_arguments(args), + staging_dir=args.staging_dir, upload_r2=args.upload_r2, record_revision=args.record_revision, r2_bucket=args.r2_bucket, r2_prefix=args.r2_prefix, wrangler_command=args.wrangler_command, ) - except SourceArtifactManifestError as error: + except (SourceArtifactManifestError, ManifestAccessError, ValueError) as error: print(f"error: {error}", file=sys.stderr) return 1 print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 if args.command == "register-artifact": - registration = register_artifact_file( - source_id=args.source_id, - package_id=args.package_id, - year=args.year, - output_dir=args.out_dir, - filename=args.filename, - sha256=args.sha256, - licence=args.licence, - access=args.access, - vintage=args.vintage, - size_bytes=args.size_bytes, - source_page=args.source_page, - source_url=args.source_url, - access_route=args.access_route, - doi=args.doi, - study=args.study, - dataset=args.dataset, - table=args.table, - publisher=args.publisher, - fetched_at=args.fetched_at, - verified_at=args.verified_at, - hash_source=args.hash_source, - notes=args.notes, - allow_reissue=args.allow_reissue, - ) + try: + registration = register_artifact_file( + source_id=args.source_id, + package_id=args.package_id, + year=args.year, + output_dir=args.out_dir, + filename=args.filename, + sha256=args.sha256, + licence=args.licence, + access=args.access, + vintage=args.vintage, + hash_source=args.hash_source, + attested_by=args.attested_by, + attestation_evidence=args.attestation_evidence, + pinned_from=_pinned_from_arguments(args), + verified_at=args.verified_at, + size_bytes=args.size_bytes, + source_page=args.source_page, + source_url=args.source_url, + access_route=args.access_route, + doi=args.doi, + study=args.study, + dataset=args.dataset, + table=args.table, + publisher=args.publisher, + fetched_at=args.fetched_at, + notes=args.notes, + allow_reissue=args.allow_reissue, + ) + except (HashOnlyRegistrationError, ManifestAccessError) as error: + print(f"error: {error}", file=sys.stderr) + return 1 print(json.dumps(registration.to_dict(), indent=2, sort_keys=True)) return 0 if registration.valid else 1 if args.command == "inventory-artifacts": report = inventory_artifact_files( args.root, manifest_filename=args.manifest, + staging_dir=args.staging_dir, ) print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 @@ -1659,6 +1845,7 @@ def main(argv: list[str] | None = None) -> int: r2_prefix=args.r2_prefix, wrangler_command=args.wrangler_command, skip_hash_only=args.skip_hash_only, + staging_dir=args.staging_dir, ) print(json.dumps(report.to_dict(), indent=2, sort_keys=True)) return 0 if report.valid else 1 From 060f2b6d31ae36c67aa30eeefcd3eb7b40e603af Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 01:59:13 -0400 Subject: [PATCH 36/42] Refuse hash-only entries at the byte reader, before any store is touched The source-package reader checked only the manifest kind, so a licensed or restricted entry in a publisher-table manifest could be read from the package tree, served from the content-addressed cache, or fetched and written into that cache, then parsed into rows, cells and facts. _read_source_artifact_content now refuses a hash-only entry first, SourceArtifactSpec.assert_parseable decides both carve-outs from the manifest alone, validate-package reports hash_only_artifact_not_parseable and manifest_kind_missing, build-suite validates before creating or replacing its output directory, and the reader refuses a vintage recorded under both key spellings instead of silently preferring the integer key. Co-Authored-By: Claude Fable 5.1 --- chronicle/source_package.py | 127 +++++++++++++++++++++++++++++++----- chronicle/suite.py | 4 ++ 2 files changed, 113 insertions(+), 18 deletions(-) diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 0fcee5e9..390fa9e7 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -34,8 +34,15 @@ from chronicle.env import env_flag, env_value from chronicle.epoch import SCHEMA_IDS, schema_id from chronicle.registration import ( + AmbiguousVintageKeyError, + ManifestAccessError, + ManifestKindError, MicrodataReleaseNotParseableError, + entry_access, + is_bare_filename, + is_hash_only, is_microdata_release, + resolve_vintage_key, ) from chronicle.sources.cells import ( SourceArtifactMetadata, @@ -853,13 +860,16 @@ def _source_artifact_metadata( raw_r2_uri=raw_r2.get("uri"), ) - def manifest_payload(self) -> dict[str, Any]: - """Load the artifact manifest this package spec points at.""" - manifest_path = files(self.resource_package).joinpath( + def manifest_resource(self) -> Any: + """Return the manifest file this package spec points at.""" + return files(self.resource_package).joinpath( self.resource_directory, self.manifest, ) - with manifest_path.open("r", encoding="utf-8") as file: + + def manifest_payload(self) -> dict[str, Any]: + """Load the artifact manifest this package spec points at.""" + with self.manifest_resource().open("r", encoding="utf-8") as file: return yaml.safe_load(file) or {} def assert_parseable_manifest(self) -> None: @@ -867,10 +877,12 @@ def assert_parseable_manifest(self) -> None: Microdata registration is manifest-level identity: no source package parses a release, and no microdata row, cell, or fact enters Chronicle - (``docs/adr-chronicle-raw-microdata-identity.md``). + (``docs/adr-chronicle-raw-microdata-identity.md``). A manifest that + declares no kind and is not frozen kindless is refused too: the reader + never assumes a publisher table. """ manifest = self.manifest_payload() - if is_microdata_release(manifest): + if is_microdata_release(manifest, manifest_path=self.manifest_resource()): raise MicrodataReleaseNotParseableError( f"{self.resource_directory}/{self.manifest} registers a " "microdata release. Registration is identity only: no source " @@ -878,14 +890,30 @@ def assert_parseable_manifest(self) -> None: "cells, or facts enter Chronicle." ) + def assert_parseable(self, year: int) -> dict[str, Any]: + """Return the entry a parse would read, refusing any it must not. + + Two carve-outs, decided from the manifest alone before any byte is + read: the manifest-level microdata-release kind, and the selected + entry's own access class -- a licensed or restricted entry is identity + only whatever manifest it sits in. + """ + self.assert_parseable_manifest() + manifest = self.manifest_payload() + spec = _year_mapping(manifest["files"], self.artifact_year or year) + _assert_entry_bytes_readable(spec) + return spec + def _artifact_content( self, year: int, ) -> tuple[bytes, str, str, dict[str, str]]: - manifest = self.manifest_payload() - if is_microdata_release(manifest): - self.assert_parseable_manifest() - spec = _year_mapping(manifest["files"], self.artifact_year or year) + spec = self.assert_parseable(year) + if not is_bare_filename(spec.get("filename")): + raise ValueError( + f"Source artifact filename must be a bare filename inside " + f"{self.resource_directory}, not {spec.get('filename')!r}." + ) artifact_path = files(self.resource_package).joinpath( self.resource_directory, spec["filename"], @@ -1239,10 +1267,14 @@ def validate_source_package( try: package.artifact.assert_parseable_manifest() - except MicrodataReleaseNotParseableError as exc: + except (MicrodataReleaseNotParseableError, ManifestKindError) as exc: errors.append( SourcePackageIssue( - code="microdata_release_not_parseable", + code=( + "microdata_release_not_parseable" + if isinstance(exc, MicrodataReleaseNotParseableError) + else "manifest_kind_missing" + ), message=str(exc), ) ) @@ -1261,6 +1293,29 @@ def validate_source_package( message=str(exc), ) ) + else: + try: + package.artifact.assert_parseable(year) + except ManifestAccessError as exc: + # Decided from the manifest alone: no package tree, cache, or + # publisher is consulted for an entry a parser must never read. + errors.append( + SourcePackageIssue( + code="hash_only_artifact_not_parseable", + message=str(exc), + ) + ) + return SourcePackageValidationReport( + package_id=package.package_id, + package_path=str(package.package_path), + year=year, + counts=counts, + errors=tuple(errors), + warnings=tuple(warnings), + ) + except (FileNotFoundError, KeyError, OSError, ValueError): + # Reported below by the artifact read itself. + pass try: package.artifact._artifact_content(year) @@ -2268,11 +2323,42 @@ def _required(payload: dict[str, Any], key: str, context: str) -> Any: def _year_mapping(files_by_year: dict[Any, Any], year: int) -> dict[str, str]: - if year in files_by_year: - return _single_year_spec(files_by_year[year], year) - if str(year) in files_by_year: - return _single_year_spec(files_by_year[str(year)], year) - raise ValueError(f"No source artifact for year {year}") + """Return the file spec for ``year``, whichever key spelling records it. + + ``2023`` and ``'2023'`` are one vintage; a manifest that records both is + refused rather than silently read through the integer key, which would + hide whichever entry -- often the one carrying the R2 history -- a writer + left under the other spelling. + """ + try: + key = resolve_vintage_key(files_by_year, year) + except AmbiguousVintageKeyError as exc: + raise ValueError(f"Source artifact for year {year}: {exc}") from exc + if key is None: + raise ValueError(f"No source artifact for year {year}") + return _single_year_spec(files_by_year[key], year) + + +def _assert_entry_bytes_readable(spec: Any) -> None: + """Refuse to read a hash-only entry's bytes from any store. + + This is the lowest byte-reader boundary: it runs before the package tree, + the content-addressed cache, or the publisher is consulted, so a licensed + or restricted entry is never read, cached, fetched, or parsed -- whatever + manifest kind it sits under. An access class Chronicle cannot parse is + refused too, never read as public. + """ + if not isinstance(spec, dict): + return + access = entry_access(spec) + if is_hash_only(access): + raise ManifestAccessError( + f"Source artifact {spec.get('filename')!r} is registered as " + f"access={access!r}. Its bytes must not enter a Chronicle store or " + "a parser: a licensed or restricted artifact is identity only, so " + "no source package reads, caches, fetches, or parses it " + "(docs/adr-chronicle-raw-microdata-identity.md)." + ) def _single_year_spec(spec: Any, year: int) -> dict[str, str]: @@ -2295,7 +2381,12 @@ def _read_source_artifact_content( artifact_path: Any, spec: dict[str, Any], ) -> bytes: - """Read a source artifact from package data, cache, or explicit fetch.""" + """Read a source artifact from package data, cache, or explicit fetch. + + Refuses a hash-only entry before touching any of the three: none of them + may hold its bytes, and the fetch branch would write them into the cache. + """ + _assert_entry_bytes_readable(spec) try: return artifact_path.read_bytes() except FileNotFoundError: diff --git a/chronicle/suite.py b/chronicle/suite.py index 0705c862..731b553d 100644 --- a/chronicle/suite.py +++ b/chronicle/suite.py @@ -289,6 +289,10 @@ def build_source_suite( source_package = try_load_source_package(source) source_id = source_package.package_id if source_package else source output_path = Path(output_dir) + if source_package is not None: + # Validate, then touch: a package whose artifact must not be parsed is + # refused before the output directory is created or replaced. + source_package.artifact.assert_parseable(year) _prepare_output_dir(output_path, replace=replace) reports_path = output_path / "reports" reports_path.mkdir(parents=True, exist_ok=True) From 29f8afc284a99d8b1c920a81e9e5536dcfa83416 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 01:59:13 -0400 Subject: [PATCH 37/42] Register consumer pins with their commit and plan the reviewed identity emit wrote a free-text hash_source and a Chronicle-side verified_at for checksums Chronicle never verified against bytes. Each registration is now a consumer_pin: attested_by PolicyEngine/microcosm, pinned_from the consumer manifest's repository, path and the last commit that changed it, no verified_at. The 15 committed FRS and SPI rows are re-emitted that way; their SHA-256 and size pins are byte-identical (verified against the microcosm blobs at commits 2fb2e2f8 and de7451bd). plan discarded publisher and vintage and printed the reviewed checksum as a comment. Every command now carries --publisher, --vintage, --filename when pinned, --expected-sha256 and --expected-size-bytes when Microcosm's pin is of the publisher bytes, the allowlisted licence US-Government-Work and its evidence issuer and scope. Anything unknown -- a URL, a publisher-bytes checksum, an evidence URL -- prints a TODO the CLI refuses, never a guess. Co-Authored-By: Claude Fable 5.1 --- db/data/dwp/frs_2023_24/manifest.yaml | 126 +++++--- .../spi_public_use_tape_2022_23/manifest.yaml | 9 +- .../be-silc-2023-registration-blocker.md | 6 +- scripts/register_microdata_releases.py | 286 +++++++++++++++--- 4 files changed, 338 insertions(+), 89 deletions(-) diff --git a/db/data/dwp/frs_2023_24/manifest.yaml b/db/data/dwp/frs_2023_24/manifest.yaml index fae85af5..604adda3 100644 --- a/db/data/dwp/frs_2023_24/manifest.yaml +++ b/db/data/dwp/frs_2023_24/manifest.yaml @@ -17,9 +17,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -37,9 +40,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -57,9 +63,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -77,9 +86,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -97,9 +109,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -117,9 +132,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -137,9 +155,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -157,9 +178,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -177,9 +201,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -197,9 +224,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -217,9 +247,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -237,9 +270,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -257,9 +293,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across @@ -277,9 +316,12 @@ files: credential grants access to them. doi: 10.5255/UKDA-SN-9367-2 study: UK Data Service SN 9367 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/source_stages.json - (frs_spine) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/source_stages.json + commit: 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 notes: Microcosm's frs_spine stage cites 'UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2'; its frs_employment, frs_council_tax, frs_education, and frs_legacy_proxies stages cite 'SN 9252' for the same 2023_24 tabs. The tabs are the same bytes across diff --git a/db/data/hmrc/spi_public_use_tape_2022_23/manifest.yaml b/db/data/hmrc/spi_public_use_tape_2022_23/manifest.yaml index 30eb6b24..3080319e 100644 --- a/db/data/hmrc/spi_public_use_tape_2022_23/manifest.yaml +++ b/db/data/hmrc/spi_public_use_tape_2022_23/manifest.yaml @@ -18,9 +18,12 @@ files: to them. doi: 10.5255/UKDA-SN-9422-1 study: UK Data Service SN 9422 - verified_at: '2026-09-02' - hash_source: PolicyEngine/microcosm packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json - (hmrc_spi_income) + hash_source: consumer_pin + attested_by: PolicyEngine/microcosm + pinned_from: + repository: PolicyEngine/microcosm + path: packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json + commit: de7451bd19ca46d2967e73cdf393908d29e72542 notes: 'Microcosm classes this artifact kind: private_microdata with access: private_local_input. Chronicle registers it restricted because the bytes are held only in a private mirror; if the UKDS terms for SN 9422 are confirmed as End User Licence it can diff --git a/docs/data-sources/be-silc-2023-registration-blocker.md b/docs/data-sources/be-silc-2023-registration-blocker.md index 0353713b..53ad7ae4 100644 --- a/docs/data-sources/be-silc-2023-registration-blocker.md +++ b/docs/data-sources/be-silc-2023-registration-blocker.md @@ -63,9 +63,13 @@ no `ledger-raw` key exists for them. --microcosm-root ~/PolicyEngine/microcosm \ --root db/data \ --release statbel-be-silc-2023 \ - emit --verified-at + emit ``` + The registration is a `consumer_pin`: it names PolicyEngine/microcosm as + the attester and records the consumer manifest's path and the commit the + pins were read from; it carries no `verified_at` of its own. + 3. Delete the `blocker` field from the `statbel-be-silc-2023` catalogue entry, and delete this document. diff --git a/scripts/register_microdata_releases.py b/scripts/register_microdata_releases.py index 6384878a..d4916b73 100644 --- a/scripts/register_microdata_releases.py +++ b/scripts/register_microdata_releases.py @@ -9,14 +9,22 @@ Write hash-only ``kind: microdata_release`` manifests for ``licensed`` and ``restricted`` releases. Every checksum, size, filename, and vintage is read verbatim from the Microcosm source-stages JSON; nothing is recomputed and - nothing is invented. A release Microcosm pins without a checksum is reported - as a blocker and never registered. + nothing is invented. Each registration is a ``consumer_pin``: it names the + consumer as the attester and records the repository, path, and commit the + pin was read from, and carries no verification date of its own. A release + Microcosm pins without a checksum is reported as a blocker and never + registered. ``plan`` Print the exact ``chronicle fetch-artifact ... --upload-r2`` commands to run from a networked machine for ``public`` releases, whose bytes Chronicle does - archive. Publisher URLs are copied verbatim from the Microcosm manifest; a - release whose manifest carries no URL prints a ``TODO`` instead of a guess. + archive. Every command carries the reviewed identity as arguments: the + publisher, the vintage, and -- when Microcosm's pin is of the publisher + bytes -- ``--expected-sha256`` and ``--expected-size-bytes``, so the fetch + refuses a reissue before archiving it. Publisher URLs are copied verbatim + from the Microcosm manifest; a release whose manifest carries no URL, no + publisher-bytes checksum, or no licence-evidence URL prints a ``TODO`` + instead of a guess, and that command cannot run until the TODO is filled. The catalogue below is the only authored content: it maps a Microcosm artifact onto Chronicle's ``{source_id, package_id, year, sha256, filename}`` identity @@ -27,7 +35,7 @@ python scripts/register_microdata_releases.py emit \\ --microcosm-root ~/PolicyEngine/microcosm \\ - --root db/data --verified-at 2026-09-02 + --root db/data python scripts/register_microdata_releases.py plan \\ --microcosm-root ~/PolicyEngine/microcosm --root db/data @@ -40,20 +48,40 @@ from dataclasses import dataclass, field import json from pathlib import Path +import re import shlex +import subprocess import sys from typing import Any # Allow `python scripts/register_microdata_releases.py` from a checkout. sys.path.insert(0, str(Path(__file__).resolve().parent.parent)) +from chronicle.artifacts import default_r2_raw_bucket # noqa: E402 from chronicle.registration import ( # noqa: E402 ACCESS_PUBLIC, + HASH_SOURCE_CONSUMER_PIN, MICRODATA_RELEASE_KIND, HashOnlyRegistrationError, register_hash_only_artifact, ) +#: The consumer whose reviewed pins every registration here transcribes. +CONSUMER_REPOSITORY = "PolicyEngine/microcosm" + +#: Placeholders a planned command prints where Microcosm pins nothing. Each is +#: refused by fetch-artifact as written, so a command carrying one cannot run +#: until a reviewer replaces it. +TODO_PUBLISHER_URL = "TODO_PUBLISHER_URL" +TODO_REVIEWED_SHA256 = "TODO_REVIEWED_SHA256" +TODO_EVIDENCE_URL = "TODO_EVIDENCE_URL" + +#: Allowlisted licence identifier for a public-use file of a U.S. federal +#: statistical agency (chronicle/licences.py). +US_GOVERNMENT_WORK = "US-Government-Work" + +_COMMIT_RE = re.compile(r"^[0-9a-f]{40}$") + #: Microcosm's per-artifact ``kind`` mapped onto Chronicle's access class. #: #: ``private_microdata`` maps to ``restricted`` rather than ``licensed``: the @@ -118,6 +146,12 @@ class Release: #: archive member, and such a hash must never be presented as the checksum #: a fetch should reproduce. pinned_sha_is_publisher_bytes: bool = True + #: Licence evidence for a public release: who issued the file under the + #: allowlisted term, the scope statement, and the durable evidence URL. + #: A missing URL prints a TODO in the plan; it is never guessed. + licence_evidence_issuer: str | None = None + licence_evidence_scope: str | None = None + licence_evidence_url: str | None = None #: Set when Microcosm pins the release without a checksum. blocker: str | None = None @@ -167,6 +201,16 @@ class Release: "a DOI. " + HASH_PROVENANCE ) +CENSUS_SCOPE = ( + "Public-use microdata file published by the U.S. Census Bureau, a federal " + "agency; a work of the United States Government under 17 U.S.C. §105." +) +FED_SCOPE = ( + "Public data set published by the Board of Governors of the Federal " + "Reserve System, a federal agency; a work of the United States Government " + "under 17 U.S.C. §105." +) + CATALOGUE: tuple[Release, ...] = ( *( Release( @@ -270,7 +314,9 @@ class Release: year=2023, table="CPS Annual Social and Economic Supplement 2023 public-use files", publisher="U.S. Census Bureau", - licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="U.S. Census Bureau", + licence_evidence_scope=CENSUS_SCOPE, access=ACCESS_PUBLIC, source_page="https://www.census.gov/programs-surveys/cps/data/datasets.html", ), @@ -287,7 +333,9 @@ class Release: year=2024, table="CPS basic monthly public-use files, January-December 2024", publisher="U.S. Census Bureau", - licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="U.S. Census Bureau", + licence_evidence_scope=CENSUS_SCOPE, access=ACCESS_PUBLIC, source_page="https://www2.census.gov/programs-surveys/cps/datasets/2024/basic/", notes=( @@ -311,7 +359,9 @@ class Release: year=2022, table="ACS 2022 1-Year PUMS household file", publisher="U.S. Census Bureau", - licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="U.S. Census Bureau", + licence_evidence_scope=CENSUS_SCOPE, access=ACCESS_PUBLIC, url_field="official_household_source", source_page="https://www.census.gov/programs-surveys/acs", @@ -335,7 +385,9 @@ class Release: year=2022, table="ACS 2022 1-Year PUMS person file", publisher="U.S. Census Bureau", - licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="U.S. Census Bureau", + licence_evidence_scope=CENSUS_SCOPE, access=ACCESS_PUBLIC, url_field="official_person_source", source_page="https://www.census.gov/programs-surveys/acs", @@ -356,9 +408,12 @@ class Release: year=2024, table="ACS 2024 1-Year PUMS household file", publisher="U.S. Census Bureau", - licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="U.S. Census Bureau", + licence_evidence_scope=CENSUS_SCOPE, access=ACCESS_PUBLIC, url_field="url", + vintage="2024", source_page="https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/", ), Release( @@ -371,9 +426,12 @@ class Release: year=2024, table="ACS 2024 1-Year PUMS person file", publisher="U.S. Census Bureau", - licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="U.S. Census Bureau", + licence_evidence_scope=CENSUS_SCOPE, access=ACCESS_PUBLIC, url_field="url", + vintage="2024", source_page="https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/", ), Release( @@ -390,7 +448,9 @@ class Release: year=2022, table="Survey of Consumer Finances 2022 summary extract", publisher="Board of Governors of the Federal Reserve System", - licence="Federal Reserve Board public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="Board of Governors of the Federal Reserve System", + licence_evidence_scope=FED_SCOPE, access=ACCESS_PUBLIC, source_page="https://www.federalreserve.gov/econres/scfindex.htm", ), @@ -408,7 +468,9 @@ class Release: year=2022, table="Survey of Consumer Finances 2022 full public data set", publisher="Board of Governors of the Federal Reserve System", - licence="Federal Reserve Board public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="Board of Governors of the Federal Reserve System", + licence_evidence_scope=FED_SCOPE, access=ACCESS_PUBLIC, source_page="https://www.federalreserve.gov/econres/scfindex.htm", notes=( @@ -431,7 +493,9 @@ class Release: year=2023, table="Survey of Income and Program Participation 2023 public-use file", publisher="U.S. Census Bureau", - licence="U.S. Census Bureau public-use file; U.S. Government work, no copyright", + licence=US_GOVERNMENT_WORK, + licence_evidence_issuer="U.S. Census Bureau", + licence_evidence_scope=CENSUS_SCOPE, access=ACCESS_PUBLIC, source_page="https://www.census.gov/programs-surveys/sipp.html", pinned_sha_is_publisher_bytes=False, @@ -586,22 +650,84 @@ def resolve( return resolved -def hash_source(release: Release) -> str: - """Return the provenance pointer recorded on a registration.""" - return f"PolicyEngine/microcosm {release.manifest} ({release.selector.stage})" +def pin_commit(microcosm_root: Path, relative: str) -> str: + """Return the commit the consumer's pin is read from, read-only. + + The pin is the manifest blob, so the commit recorded is the last one that + changed that file: it addresses exactly the bytes the registration + transcribes, and it is stable across later, unrelated commits so repeated + ``emit`` runs stay byte-identical. + """ + try: + completed = subprocess.run( + [ + "git", + "-C", + str(microcosm_root), + "log", + "-1", + "--format=%H", + "--", + relative, + ], + capture_output=True, + text=True, + check=True, + ) + except (OSError, subprocess.CalledProcessError) as exc: + raise CatalogueError( + f"Cannot read the commit of {relative} in {microcosm_root}: {exc}. " + "Pass --microcosm-commit with the reviewed commit." + ) from exc + commit = completed.stdout.strip() + if not _COMMIT_RE.match(commit): + raise CatalogueError( + f"{microcosm_root} records no commit for {relative}; pass " + "--microcosm-commit with the reviewed commit." + ) + return commit + + +def parse_pin_commits(values: Sequence[str]) -> dict[str, str]: + """Parse ``--microcosm-commit`` values into ``{manifest path or '*': commit}``.""" + commits: dict[str, str] = {} + for value in values: + path, separator, commit = value.rpartition("=") + key = path if separator else "*" + if not _COMMIT_RE.match(commit): + raise CatalogueError( + f"--microcosm-commit must name a 40-hex commit, not {value!r}." + ) + if key in commits and commits[key] != commit: + raise CatalogueError( + f"--microcosm-commit names two commits for {key!r}; pass one." + ) + commits[key] = commit + return commits + + +def pinned_from(release: Release, commit: str) -> dict[str, str]: + """Return the ``pinned_from`` block a consumer_pin registration records.""" + return { + "repository": CONSUMER_REPOSITORY, + "path": release.manifest, + "commit": commit, + } def emit( resolved: Sequence[ResolvedRelease], *, root: Path, - verified_at: str, + pin_commits: Mapping[str, str], allow_reissue: bool = False, ) -> tuple[list[dict[str, Any]], list[dict[str, str]]]: """Write hash-only manifests for every registrable non-public release. - Returns ``(registrations, blockers)``. A release Microcosm pins without a - checksum is a blocker, not a registration: no hash is ever invented. + ``pin_commits`` maps each consumer manifest path to the commit its pins + are read from. Returns ``(registrations, blockers)``. A release Microcosm + pins without a checksum is a blocker, not a registration: no hash is ever + invented. """ registrations: list[dict[str, Any]] = [] blockers: list[dict[str, str]] = [] @@ -636,6 +762,9 @@ def emit( licence=release.licence, access=release.access, vintage=item.vintage, + hash_source=HASH_SOURCE_CONSUMER_PIN, + attested_by=CONSUMER_REPOSITORY, + pinned_from=pinned_from(release, pin_commits[release.manifest]), size_bytes=item.size_bytes, source_page=release.source_page, access_route=release.access_route, @@ -643,8 +772,6 @@ def emit( study=release.study, table=release.table, publisher=release.publisher, - verified_at=verified_at, - hash_source=hash_source(release), notes=release.notes or HASH_PROVENANCE, allow_reissue=allow_reissue, ) @@ -668,7 +795,7 @@ def fetch_command( todos: list[str] = [] url = item.url if url is None: - url = "TODO_PUBLISHER_URL" + url = TODO_PUBLISHER_URL todos.append( f"{release.release_id}: Microcosm records no publisher URL " f"(field {release.url_field!r}); read it off {release.source_page}." @@ -689,11 +816,23 @@ def fetch_command( "--url", url, ] + if item.filename: + argv += ["--filename", item.filename] if release.source_page: argv += ["--source-page", release.source_page] + vintage = item.vintage + if not vintage: + todos.append( + f"{release.release_id}: Microcosm records no vintage; add " + "Release.vintage to the catalogue before running this command." + ) argv += [ "--table", release.table, + "--publisher", + release.publisher, + "--vintage", + vintage or "TODO_VINTAGE", "--access", ACCESS_PUBLIC, "--licence", @@ -702,22 +841,46 @@ def fetch_command( # several files share one vintage and no source package parses it. "--kind", MICRODATA_RELEASE_KIND, - "--upload-r2", - "--r2-bucket", - r2_bucket, ] + # The reviewed identity travels as arguments, never as a comment: the + # fetch refuses bytes that hash differently before archiving anything. if item.sha256 and release.pinned_sha_is_publisher_bytes: + argv += ["--expected-sha256", item.sha256] + if item.size_bytes: + argv += ["--expected-size-bytes", str(item.size_bytes)] + else: + argv += ["--expected-sha256", TODO_REVIEWED_SHA256] + if item.sha256: + todos.append( + f"{release.release_id}: Microcosm's pinned sha256 {item.sha256} " + "is NOT the publisher artifact's checksum. A public release is " + "archived only against a reviewed checksum its licence evidence " + "covers; review the publisher bytes and replace " + f"{TODO_REVIEWED_SHA256} before running this command." + ) + else: + todos.append( + f"{release.release_id}: Microcosm pins no checksum for this " + "release. A public release is archived only against a reviewed " + f"checksum; replace {TODO_REVIEWED_SHA256} before running this " + "command." + ) + argv += [ + "--licence-evidence-issuer", + release.licence_evidence_issuer or release.publisher, + "--licence-evidence-scope", + release.licence_evidence_scope or "TODO_EVIDENCE_SCOPE", + "--licence-evidence-url", + release.licence_evidence_url or TODO_EVIDENCE_URL, + ] + if not release.licence_evidence_url: todos.append( - f"{release.release_id}: expect sha256 {item.sha256}" - + (f" and size {item.size_bytes}" if item.size_bytes else "") - + " — fail the registration if the fetched bytes differ." - ) - elif item.sha256: - todos.append( - f"{release.release_id}: Microcosm's pinned sha256 {item.sha256} is " - "NOT the publisher artifact's checksum — do not use it to verify " - "this fetch. See the note below." + f"{release.release_id}: no durable licence-evidence URL is " + f"catalogued; replace {TODO_EVIDENCE_URL} with the publisher's " + "statement that this file is issued under " + f"{release.licence} before running this command." ) + argv += ["--upload-r2", "--r2-bucket", r2_bucket] if release.notes: todos.append(f"{release.release_id}: {release.notes}") return shlex.join(argv), todos @@ -777,11 +940,17 @@ def build_parser() -> argparse.ArgumentParser: help="Write hash-only manifests for licensed and restricted releases", ) emit_parser.add_argument( - "--verified-at", - required=True, + "--microcosm-commit", + action="append", + default=None, + metavar="[PATH=]COMMIT", help=( - "Date the pins were verified against Microcosm, as YYYY-MM-DD. " - "Required so repeated runs are byte-stable." + "Commit the consumer pins are read from, recorded as " + "pinned_from.commit on every registration. A bare COMMIT applies " + "to every consumer manifest; PATH=COMMIT (repeatable) names the " + "commit for one manifest path. Defaults to the last commit that " + "changed each consumer manifest, read from the checkout's git " + "history." ), ) emit_parser.add_argument( @@ -796,8 +965,11 @@ def build_parser() -> argparse.ArgumentParser: ) plan_parser.add_argument( "--r2-bucket", - default="ledger-raw", - help="Raw bucket the fetch should upload to.", + default=None, + help=( + "Raw bucket the fetch should upload to. Defaults to " + "$CHRONICLE_R2_RAW_BUCKET, else the ledger-era default." + ), ) return parser @@ -822,11 +994,35 @@ def main(argv: list[str] | None = None) -> int: return 1 if args.command == "emit": + try: + declared = parse_pin_commits(args.microcosm_commit or ()) + except CatalogueError as exc: + print(str(exc), file=sys.stderr) + return 2 + # Only a registrable release needs its pin's commit: a public release + # is fetched, not transcribed, and a blocked one is never registered. + registrable = sorted( + { + item.release.manifest + for item in resolved + if item.release.access != ACCESS_PUBLIC and not item.release.blocker + } + ) + try: + pin_commits = { + manifest: declared.get(manifest) + or declared.get("*") + or pin_commit(microcosm_root, manifest) + for manifest in registrable + } + except CatalogueError as exc: + print(str(exc), file=sys.stderr) + return 1 try: registrations, blockers = emit( resolved, root=args.root, - verified_at=args.verified_at, + pin_commits=pin_commits, allow_reissue=args.allow_reissue, ) except HashOnlyRegistrationError as exc: @@ -850,7 +1046,11 @@ def main(argv: list[str] | None = None) -> int: ) return 0 - commands, todos = plan(resolved, root=args.root, r2_bucket=args.r2_bucket) + commands, todos = plan( + resolved, + root=args.root, + r2_bucket=args.r2_bucket or default_r2_raw_bucket(), + ) if args.json: print(json.dumps({"commands": commands, "todos": todos}, indent=2)) return 0 From 98d376568cb56bbc36ddf0b173969577f087f866 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 01:59:13 -0400 Subject: [PATCH 38/42] Pin every refusal with hermetic tests and a synthetic consumer checkout The catalogue tests read a checkout under the user's home directory and skipped when it was absent, so CI never ran them, and the committed registrations were checked for shape rather than for the reviewed pins. tests/fixtures/microcosm now holds synthetic consumer manifests in Microcosm's shapes with the reviewed pins verbatim and decoy artifacts; emit reproduces both committed manifests byte for byte from it, plan matches a golden command list, every planned command parses against the real CLI with the reviewed identity as arguments, and a planned TODO is refused before any read. The committed pins are a golden mapping. Regression tests cover every gate finding: filename aliases, the byte reader, reclassification of an archived identity, kind conflicts and strict pre-fetch validation, inference before the read, in-place list updates under the recorded-identity guard, quoted year keys, the expected identity, and the explicit-kind rule with its frozen list and the no-tracked-microdata-bytes guard. Co-Authored-By: Claude Fable 5.1 --- tests/fixtures/microcosm/golden_plan.json | 36 + .../src/microcosm/build/be/source_stages.json | 27 + .../build/uk/hmrc_income_source_stages.json | 45 + .../src/microcosm/build/uk/source_stages.json | 212 ++ .../src/microcosm/build/us/source_stages.json | 112 + .../us_runtime/acs_2024_1yr_sources.json | 28 + tests/test_chronicle_manifest_kind.py | 314 ++ tests/test_chronicle_microdata_catalogue.py | 424 +++ .../test_chronicle_microdata_registration.py | 2547 ++++++++++++++--- 9 files changed, 3302 insertions(+), 443 deletions(-) create mode 100644 tests/fixtures/microcosm/golden_plan.json create mode 100644 tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/be/source_stages.json create mode 100644 tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json create mode 100644 tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/uk/source_stages.json create mode 100644 tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/us/source_stages.json create mode 100644 tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/us_runtime/acs_2024_1yr_sources.json create mode 100644 tests/test_chronicle_manifest_kind.py create mode 100644 tests/test_chronicle_microdata_catalogue.py diff --git a/tests/fixtures/microcosm/golden_plan.json b/tests/fixtures/microcosm/golden_plan.json new file mode 100644 index 00000000..9941d5fb --- /dev/null +++ b/tests/fixtures/microcosm/golden_plan.json @@ -0,0 +1,36 @@ +{ + "commands": [ + "uv run chronicle fetch-artifact --source-id census_cps --package-id census-cps-asec-2023 --year 2023 --out-dir db/data/census/cps_asec_2023 --url https://www2.census.gov/programs-surveys/cps/datasets/2023/march/asecpub23csv.zip --source-page https://www.census.gov/programs-surveys/cps/data/datasets.html --table 'CPS Annual Social and Economic Supplement 2023 public-use files' --publisher 'U.S. Census Bureau' --vintage '2023 ASEC / 2022 income reference year' --access public --licence US-Government-Work --kind microdata_release --expected-sha256 d2e000250782adfbdd7f29c82b66d866591a30f0d330496698ec19f9c784ce11 --expected-size-bytes 150165063 --licence-evidence-issuer 'U.S. Census Bureau' --licence-evidence-scope 'Public-use microdata file published by the U.S. Census Bureau, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw", + "uv run chronicle fetch-artifact --source-id census_cps --package-id census-cps-basic-monthly-2024 --year 2024 --out-dir db/data/census/cps_basic_monthly_2024 --url TODO_PUBLISHER_URL --source-page https://www2.census.gov/programs-surveys/cps/datasets/2024/basic/ --table 'CPS basic monthly public-use files, January-December 2024' --publisher 'U.S. Census Bureau' --vintage 2024 --access public --licence US-Government-Work --kind microdata_release --expected-sha256 TODO_REVIEWED_SHA256 --licence-evidence-issuer 'U.S. Census Bureau' --licence-evidence-scope 'Public-use microdata file published by the U.S. Census Bureau, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw", + "uv run chronicle fetch-artifact --source-id census_acs --package-id census-acs-pums-2022-1yr --year 2022 --out-dir db/data/census/acs_pums_2022_1yr --url https://www2.census.gov/programs-surveys/acs/data/pums/2022/1-Year/csv_hus.zip --source-page https://www.census.gov/programs-surveys/acs --table 'ACS 2022 1-Year PUMS household file' --publisher 'U.S. Census Bureau' --vintage 2022 --access public --licence US-Government-Work --kind microdata_release --expected-sha256 TODO_REVIEWED_SHA256 --licence-evidence-issuer 'U.S. Census Bureau' --licence-evidence-scope 'Public-use microdata file published by the U.S. Census Bureau, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw", + "uv run chronicle fetch-artifact --source-id census_acs --package-id census-acs-pums-2022-1yr --year 2022 --out-dir db/data/census/acs_pums_2022_1yr --url https://www2.census.gov/programs-surveys/acs/data/pums/2022/1-Year/csv_pus.zip --source-page https://www.census.gov/programs-surveys/acs --table 'ACS 2022 1-Year PUMS person file' --publisher 'U.S. Census Bureau' --vintage 2022 --access public --licence US-Government-Work --kind microdata_release --expected-sha256 TODO_REVIEWED_SHA256 --licence-evidence-issuer 'U.S. Census Bureau' --licence-evidence-scope 'Public-use microdata file published by the U.S. Census Bureau, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw", + "uv run chronicle fetch-artifact --source-id census_acs --package-id census-acs-pums-2024-1yr --year 2024 --out-dir db/data/census/acs_pums_2024_1yr --url https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/csv_hus.zip --filename csv_hus.zip --source-page https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/ --table 'ACS 2024 1-Year PUMS household file' --publisher 'U.S. Census Bureau' --vintage 2024 --access public --licence US-Government-Work --kind microdata_release --expected-sha256 8281008e53de98f0ef81e7a2ee5a8725991dda1ecfd2713ead73246425e515d0 --expected-size-bytes 251500587 --licence-evidence-issuer 'U.S. Census Bureau' --licence-evidence-scope 'Public-use microdata file published by the U.S. Census Bureau, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw", + "uv run chronicle fetch-artifact --source-id census_acs --package-id census-acs-pums-2024-1yr --year 2024 --out-dir db/data/census/acs_pums_2024_1yr --url https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/csv_pus.zip --filename csv_pus.zip --source-page https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/ --table 'ACS 2024 1-Year PUMS person file' --publisher 'U.S. Census Bureau' --vintage 2024 --access public --licence US-Government-Work --kind microdata_release --expected-sha256 afdc6d90c6e2f0bab365ed32d95ba4c4d8ac651162f46ac7861295b2dc469894 --expected-size-bytes 602847146 --licence-evidence-issuer 'U.S. Census Bureau' --licence-evidence-scope 'Public-use microdata file published by the U.S. Census Bureau, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw", + "uv run chronicle fetch-artifact --source-id federal_reserve --package-id federal-reserve-scf-2022 --year 2022 --out-dir db/data/federal_reserve/scf_2022 --url https://www.federalreserve.gov/econres/files/scfp2022s.zip --source-page https://www.federalreserve.gov/econres/scfindex.htm --table 'Survey of Consumer Finances 2022 summary extract' --publisher 'Board of Governors of the Federal Reserve System' --vintage 2022 --access public --licence US-Government-Work --kind microdata_release --expected-sha256 3bb4d890ae2463ff6039ec7692e375f544dd98a55a37ca2cb2340354b9cc9d80 --licence-evidence-issuer 'Board of Governors of the Federal Reserve System' --licence-evidence-scope 'Public data set published by the Board of Governors of the Federal Reserve System, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw", + "uv run chronicle fetch-artifact --source-id federal_reserve --package-id federal-reserve-scf-2022 --year 2022 --out-dir db/data/federal_reserve/scf_2022 --url https://www.federalreserve.gov/econres/files/scf2022s.zip --source-page https://www.federalreserve.gov/econres/scfindex.htm --table 'Survey of Consumer Finances 2022 full public data set' --publisher 'Board of Governors of the Federal Reserve System' --vintage 2022 --access public --licence US-Government-Work --kind microdata_release --expected-sha256 TODO_REVIEWED_SHA256 --licence-evidence-issuer 'Board of Governors of the Federal Reserve System' --licence-evidence-scope 'Public data set published by the Board of Governors of the Federal Reserve System, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw", + "uv run chronicle fetch-artifact --source-id census_sipp --package-id census-sipp-2023 --year 2023 --out-dir db/data/census/sipp_2023 --url TODO_PUBLISHER_URL --source-page https://www.census.gov/programs-surveys/sipp.html --table 'Survey of Income and Program Participation 2023 public-use file' --publisher 'U.S. Census Bureau' --vintage 2023 --access public --licence US-Government-Work --kind microdata_release --expected-sha256 TODO_REVIEWED_SHA256 --licence-evidence-issuer 'U.S. Census Bureau' --licence-evidence-scope 'Public-use microdata file published by the U.S. Census Bureau, a federal agency; a work of the United States Government under 17 U.S.C. \u00a7105.' --licence-evidence-url TODO_EVIDENCE_URL --upload-r2 --r2-bucket ledger-raw" + ], + "todos": [ + "census-cps-asec-2023: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "census-cps-basic-monthly-2024: Microcosm records no publisher URL (field 'locator'); read it off https://www2.census.gov/programs-surveys/cps/datasets/2024/basic/.", + "census-cps-basic-monthly-2024: Microcosm pins no checksum for this release. A public release is archived only against a reviewed checksum; replace TODO_REVIEWED_SHA256 before running this command.", + "census-cps-basic-monthly-2024: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "census-cps-basic-monthly-2024: Microcosm pins the twelve monthly files as the locator string 'jan24pub through dec24pub' with no per-file URL, filename extension, or checksum. The publisher directory is the stage source; the twelve filenames must be read off that directory before the fetch commands can be completed.", + "census-acs-pums-2022-household: Microcosm's pinned sha256 0b319b496f19a6913066f9c5ea572edfda3d78a187be6f375846617d0b441bd4 is NOT the publisher artifact's checksum. A public release is archived only against a reviewed checksum its licence evidence covers; review the publisher bytes and replace TODO_REVIEWED_SHA256 before running this command.", + "census-acs-pums-2022-household: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "census-acs-pums-2022-household: The Microcosm artifact's own sha256 belongs to the derived acs_2022.h5, not to the publisher zip; only the publisher URL is reused here. The fetch computes the release checksum.", + "census-acs-pums-2022-person: Microcosm's pinned sha256 0b319b496f19a6913066f9c5ea572edfda3d78a187be6f375846617d0b441bd4 is NOT the publisher artifact's checksum. A public release is archived only against a reviewed checksum its licence evidence covers; review the publisher bytes and replace TODO_REVIEWED_SHA256 before running this command.", + "census-acs-pums-2022-person: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "census-acs-pums-2022-person: The Microcosm artifact's own sha256 belongs to the derived acs_2022.h5, not to the publisher zip; only the publisher URL is reused here. The fetch computes the release checksum.", + "census-acs-pums-2024-household: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "census-acs-pums-2024-person: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "federal-reserve-scf-2022-summary: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "federal-reserve-scf-2022-full: Microcosm pins no checksum for this release. A public release is archived only against a reviewed checksum; replace TODO_REVIEWED_SHA256 before running this command.", + "federal-reserve-scf-2022-full: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "federal-reserve-scf-2022-full: Microcosm records no checksum for this zip: 'Full-file SHA-256 pending one network-enabled provisioning fetch'. The fetch below computes and registers it.", + "census-sipp-2023: Microcosm records no publisher URL (field 'locator'); read it off https://www.census.gov/programs-surveys/sipp.html.", + "census-sipp-2023: Microcosm's pinned sha256 5c30439e365fc26483318ef61d1d8f4bb2f0e9d6bb47c22c06756a7698733ee2 is NOT the publisher artifact's checksum. A public release is archived only against a reviewed checksum its licence evidence covers; review the publisher bytes and replace TODO_REVIEWED_SHA256 before running this command.", + "census-sipp-2023: no durable licence-evidence URL is catalogued; replace TODO_EVIDENCE_URL with the publisher's statement that this file is issued under US-Government-Work before running this command.", + "census-sipp-2023: Microcosm reaches this file through an immutable Hugging Face mirror (revision 21280dca5995e978d706740a8a4b9b7860cfd7b6) and records no Census URL, so the publisher URL must be read off the SIPP dataset page before the fetch. Microcosm's pinned sha256 5c30439e365fc26483318ef61d1d8f4bb2f0e9d6bb47c22c06756a7698733ee2 and size 3726010471 are for the mirrored pu2023.csv member, not for whatever archive the Census page serves, so they are not the checksum this fetch should be expected to reproduce." + ] +} diff --git a/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/be/source_stages.json b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/be/source_stages.json new file mode 100644 index 00000000..9ea696a9 --- /dev/null +++ b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/be/source_stages.json @@ -0,0 +1,27 @@ +{ + "snapshot_of": { + "repository": "PolicyEngine/microcosm", + "commit": "fe2f92f8c79d36b897e6ae4d7e4b3c33fc7ebd28", + "note": "Synthetic consumer manifests in Microcosm's shapes, carrying the reviewed pins verbatim; decoy artifacts exercise selection." + }, + "version": "fixture", + "country": "be", + "policy": "synthetic Microcosm BE source-stages manifest", + "stages": [ + { + "stage": "silc_load", + "survey": "BE-SILC (Statbel national SILC)", + "source": "https://statbel.fgov.be/en/themes/households/poverty-and-living-conditions", + "grain": "person", + "artifacts": [ + { + "kind": "restricted_microdata", + "format": "csv_or_spss", + "vintage": "2023", + "locator": "Statbel BE-SILC scientific-use files: D (household register), R (personal register), H (household data), P (personal data)", + "licence": "Statbel/Eurostat scientific-use; restricted" + } + ] + } + ] +} diff --git a/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json new file mode 100644 index 00000000..85411e3f --- /dev/null +++ b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json @@ -0,0 +1,45 @@ +{ + "snapshot_of": { + "repository": "PolicyEngine/microcosm", + "commit": "fe2f92f8c79d36b897e6ae4d7e4b3c33fc7ebd28", + "note": "Synthetic consumer manifests in Microcosm's shapes, carrying the reviewed pins verbatim; decoy artifacts exercise selection." + }, + "version": "fixture", + "country": "uk", + "policy": "synthetic Microcosm HMRC income source-stages manifest", + "stages": [ + { + "stage": "hmrc_spi_income", + "survey": "Survey of Personal Incomes Public Use Tape 2022-23", + "source": "https://example.invalid/collated_tables.ods", + "grain": "person", + "artifacts": [ + { + "role": "qrf_donor", + "kind": "private_microdata", + "format": "tab_delimited", + "survey": "Survey of Personal Incomes Public Use Tape 2022-23", + "vintage": "2022-23", + "tax_year_start": 2022, + "ukds_study_number": "SN 9422", + "doi": "10.5255/UKDA-SN-9422-1", + "filename": "put2223uk.tab", + "sha256": "5ef829461060c91a2a47be59ad541d9b519fc3976d66ca80d4920f711bb96f66", + "size_bytes": 141323762, + "access": "private_local_input", + "locator": "caller-supplied local input", + "runtime_sha256_required": true + }, + { + "role": "published_fact_surface", + "kind": "administrative_table", + "format": "ods", + "vintage": "2023-24", + "locator": "https://example.invalid/collated_tables.ods", + "sha256": "1111111111111111111111111111111111111111111111111111111111111111", + "size_bytes": 2 + } + ] + } + ] +} diff --git a/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/uk/source_stages.json b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/uk/source_stages.json new file mode 100644 index 00000000..22d59100 --- /dev/null +++ b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/uk/source_stages.json @@ -0,0 +1,212 @@ +{ + "snapshot_of": { + "repository": "PolicyEngine/microcosm", + "commit": "fe2f92f8c79d36b897e6ae4d7e4b3c33fc7ebd28", + "note": "Synthetic consumer manifests in Microcosm's shapes, carrying the reviewed pins verbatim; decoy artifacts exercise selection." + }, + "version": "fixture", + "country": "uk", + "policy": "synthetic Microcosm source-stages manifest for Chronicle tests", + "stages": [ + { + "stage": "frs_spine", + "survey": "Family Resources Survey 2023-24", + "source": "Department for Work and Pensions Family Resources Survey 2023-24, UK Data Service SN 9367, DOI 10.5255/UKDA-SN-9367-2", + "grain": "household", + "artifacts": [ + { + "role": "frs_table", + "table": "accounts", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "accounts.tab", + "sha256": "c5e31932bfd06087f835d2c83c0984c85a93409bf5ef85b699cb0958abcba1ea", + "size_bytes": 1807921, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "adult", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "adult.tab", + "sha256": "e09f9647d03585c81a528636028b2ed495f8f1fbcf64c5e7b4fe521b67367e06", + "size_bytes": 35323384, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "benefits", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "benefits.tab", + "sha256": "ff30d054cc659bcf23b44c492d98cfd701c0bfdb63e8e9aa9769b490ba9d636b", + "size_bytes": 4460292, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "benunit", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "benunit.tab", + "sha256": "88946815eace8561516d5cbb442c27e319c1e90abc381fb2338f0126e3b9e05b", + "size_bytes": 21213867, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "child", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "child.tab", + "sha256": "b5dc84fe8b002ee925e61fae23fed27b11537af9fb174f1d07d9cc1748b9702e", + "size_bytes": 2913156, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "chldcare", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "chldcare.tab", + "sha256": "566e0ebca1d5e2f3e424e556c91f4cb583d17dadfdfa59feb3841eda7e5976a3", + "size_bytes": 273837, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "extchild", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "extchild.tab", + "sha256": "8d358d7ee66ee4a7ceab87b4f24fbbf21ac86dc038dc7831e51fb271f96a57ec", + "size_bytes": 18677, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "househol", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "househol.tab", + "sha256": "5fd26b8b675f33b3b30c9ac789a18da17de734790f77e00ded287d1c3a187b30", + "size_bytes": 12387117, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "job", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "job.tab", + "sha256": "88b77ffe06865f029f713bb1d55ff12bdea8a1234de5bc293e72458fe64f3a74", + "size_bytes": 10934873, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "maint", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "maint.tab", + "sha256": "f2dc924eb5a51b0c357791693d15b431327dc39c6421011efb313d88bf839695", + "size_bytes": 15440, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "mortgage", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "mortgage.tab", + "sha256": "ce36b477d67837c469608a0d68f7ef269ac04758974235f1157d2f6b92cdbfdc", + "size_bytes": 631783, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "oddjob", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "oddjob.tab", + "sha256": "b4ba3dd3151f73a01422983c60514a3e38458ddfa4fb33ae4ed0326873406305", + "size_bytes": 5165, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "penprov", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "penprov.tab", + "sha256": "ee001461c40306ec24b38b2881e1774121114266a2ee449d606cd0a811c37731", + "size_bytes": 522313, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "pension", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "pension.tab", + "sha256": "150d6fad1fce81254fb7aea1526fbb00b63d4027d6e2ac4c26bb90aea3127eb7", + "size_bytes": 1225838, + "runtime_sha256_required": true + }, + { + "kind": "administrative_table", + "format": "ods", + "vintage": "2023-24", + "locator": "https://example.invalid/decoy_table.ods", + "sha256": "0000000000000000000000000000000000000000000000000000000000000000", + "size_bytes": 1 + } + ] + }, + { + "stage": "frs_employment", + "survey": "Family Resources Survey 2023-24", + "source": "UK Data Service SN 9252; local licensed 2023_24 tabs.", + "grain": "person", + "artifacts": [ + { + "role": "frs_table", + "table": "adult", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "adult.tab", + "sha256": "e09f9647d03585c81a528636028b2ed495f8f1fbcf64c5e7b4fe521b67367e06", + "size_bytes": 35323384, + "runtime_sha256_required": true + }, + { + "role": "frs_table", + "table": "job", + "kind": "licensed_microdata", + "format": "tab", + "vintage": "2023_24", + "locator": "job.tab", + "sha256": "88b77ffe06865f029f713bb1d55ff12bdea8a1234de5bc293e72458fe64f3a74", + "size_bytes": 10934873, + "runtime_sha256_required": true + } + ] + } + ] +} diff --git a/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/us/source_stages.json b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/us/source_stages.json new file mode 100644 index 00000000..1523b8b1 --- /dev/null +++ b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/us/source_stages.json @@ -0,0 +1,112 @@ +{ + "snapshot_of": { + "repository": "PolicyEngine/microcosm", + "commit": "fe2f92f8c79d36b897e6ae4d7e4b3c33fc7ebd28", + "note": "Synthetic consumer manifests in Microcosm's shapes, carrying the reviewed pins verbatim; decoy artifacts exercise selection." + }, + "version": "fixture", + "country": "us", + "policy": "synthetic Microcosm US source-stages manifest", + "stages": [ + { + "stage": "weeks_unemployed_input", + "survey": "Census CPS ASEC", + "source": "https://www.census.gov/programs-surveys/cps.html", + "grain": "person", + "artifacts": [ + { + "kind": "public_microdata", + "format": "zip_csv", + "vintage": "2023 ASEC / 2022 income reference year", + "locator": "https://www2.census.gov/programs-surveys/cps/datasets/2023/march/asecpub23csv.zip", + "sha256": "d2e000250782adfbdd7f29c82b66d866591a30f0d330496698ec19f9c784ce11", + "size_bytes": 150165063, + "member": "pppub23.csv" + }, + { + "kind": "official_data_dictionary", + "format": "pdf", + "vintage": "2023", + "locator": "https://example.invalid/asec2023_ddl_pub_full.pdf" + } + ] + }, + { + "stage": "org_wages", + "survey": "CPS ORG", + "source": "https://www2.census.gov/programs-surveys/cps/datasets/2024/basic/", + "grain": "person", + "artifacts": [ + { + "kind": "public_microdata", + "format": "cps_basic_monthly_csv_or_zip", + "vintage": "2024", + "locator": "jan24pub through dec24pub; HRMIS 4 and 8" + }, + { + "kind": "administrative_table", + "format": "published_table", + "vintage": "2024", + "locator": "https://example.invalid/union2.t01.1" + } + ] + }, + { + "stage": "acs_rent", + "survey": "Census ACS 2022", + "source": "https://www.census.gov/programs-surveys/acs", + "grain": "household", + "artifacts": [ + { + "kind": "public_microdata", + "format": "census_asec_hdf5", + "vintage": "2022-2024 pooled", + "locator": "SHA-locked Census CPS ASEC person and household tables" + }, + { + "kind": "versioned_derived_microdata", + "format": "hdf5_arrays", + "vintage": "2022", + "locator": "Census ACS 2022 PUMS processed person/household arrays (acs_2022.h5)", + "sha256": "0b319b496f19a6913066f9c5ea572edfda3d78a187be6f375846617d0b441bd4", + "official_person_source": "https://www2.census.gov/programs-surveys/acs/data/pums/2022/1-Year/csv_pus.zip", + "official_household_source": "https://www2.census.gov/programs-surveys/acs/data/pums/2022/1-Year/csv_hus.zip" + } + ] + }, + { + "stage": "scf_wealth", + "survey": "Fed SCF 2022 + Census SIPP 2023", + "source": "https://www.federalreserve.gov/econres/scfindex.htm", + "grain": "household", + "artifacts": [ + { + "kind": "public_microdata", + "format": "stata_in_zip", + "vintage": "2022", + "locator": "https://www.federalreserve.gov/econres/files/scfp2022s.zip", + "member": "rscfp2022.dta", + "sha256": "3bb4d890ae2463ff6039ec7692e375f544dd98a55a37ca2cb2340354b9cc9d80" + }, + { + "kind": "public_microdata", + "format": "stata_in_zip", + "vintage": "2022", + "locator": "https://www.federalreserve.gov/econres/files/scf2022s.zip", + "member": "p22i6.dta", + "expected_rows": 22975 + }, + { + "kind": "public_microdata", + "format": "pipe_delimited_csv", + "vintage": "2023", + "locator": "Census SIPP 2023 public-use file; immutable Hugging Face mirror revision 21280dca5995e978d706740a8a4b9b7860cfd7b6", + "member": "pu2023.csv", + "revision": "21280dca5995e978d706740a8a4b9b7860cfd7b6", + "sha256": "5c30439e365fc26483318ef61d1d8f4bb2f0e9d6bb47c22c06756a7698733ee2", + "size_bytes": 3726010471 + } + ] + } + ] +} diff --git a/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/us_runtime/acs_2024_1yr_sources.json b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/us_runtime/acs_2024_1yr_sources.json new file mode 100644 index 00000000..389950b1 --- /dev/null +++ b/tests/fixtures/microcosm/packages/microcosm-build/src/microcosm/build/us_runtime/acs_2024_1yr_sources.json @@ -0,0 +1,28 @@ +{ + "snapshot_of": { + "repository": "PolicyEngine/microcosm", + "commit": "fe2f92f8c79d36b897e6ae4d7e4b3c33fc7ebd28", + "note": "Synthetic consumer manifests in Microcosm's shapes, carrying the reviewed pins verbatim; decoy artifacts exercise selection." + }, + "version": "fixture", + "spine": "acs_2024_1yr", + "vintage": "2024", + "verified_on": "2026-08-01", + "source_directory": "https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/", + "artifacts": [ + { + "role": "household", + "filename": "csv_hus.zip", + "url": "https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/csv_hus.zip", + "sha256": "8281008e53de98f0ef81e7a2ee5a8725991dda1ecfd2713ead73246425e515d0", + "size_bytes": 251500587 + }, + { + "role": "person", + "filename": "csv_pus.zip", + "url": "https://www2.census.gov/programs-surveys/acs/data/pums/2024/1-Year/csv_pus.zip", + "sha256": "afdc6d90c6e2f0bab365ed32d95ba4c4d8ac651162f46ac7861295b2dc469894", + "size_bytes": 602847146 + } + ] +} diff --git a/tests/test_chronicle_manifest_kind.py b/tests/test_chronicle_manifest_kind.py new file mode 100644 index 00000000..e5409b0b --- /dev/null +++ b/tests/test_chronicle_manifest_kind.py @@ -0,0 +1,314 @@ +"""The explicit-kind rule and the repository guard for microdata bytes. + +Every manifest created or modified after the microdata-identity ADR declares +``kind``. Manifests that predate the rule are frozen, byte for byte, in +``chronicle/grandfathered_manifests.py``; any other kindless manifest is an +error at every entry point -- ``fetch-artifact``, ``publish-raw``, +``inventory-artifacts``, ``validate-package`` and the source-package byte +reader -- and never a publisher table by default. A ``kind: microdata_release`` +package directory holds only manifests: public release bytes are archived in +R2 from a staging directory outside the tree, never committed. +""" + +from __future__ import annotations + +import subprocess +import sys +import uuid +from pathlib import Path + +import pytest +import yaml + +from chronicle.artifacts import ( + fetch_source_artifact, + inventory_source_artifacts, + publish_source_artifacts, +) +from chronicle.grandfathered_manifests import ( + GRANDFATHERED_KINDLESS_MANIFESTS, + grandfathered_manifest_key, + is_grandfathered_manifest, + manifest_digest, +) +from chronicle.registration import ( + ManifestAccessError, + ManifestKindError, + manifest_kind, + safe_manifest_kind, +) +from chronicle.source_package import SourceArtifactSpec, validate_source_package + + +REPO_ROOT = Path(__file__).resolve().parents[1] +FREEZE_SIZE = 161 + + +def _tracked_manifests() -> list[Path]: + listed = subprocess.run( + ["git", "-C", str(REPO_ROOT), "ls-files", "db/data"], + capture_output=True, + text=True, + check=True, + ).stdout.split() + return [ + REPO_ROOT / path + for path in listed + if Path(path).name.startswith("manifest") and path.endswith((".yaml", ".yml")) + ] + + +def _kindless(path: Path) -> bool: + payload = yaml.safe_load(path.read_text()) or {} + return isinstance(payload, dict) and "kind" not in payload + + +def test_every_kindless_manifest_in_the_tree_is_frozen_unmodified(): + tracked = _tracked_manifests() + assert tracked + kindless = { + path.relative_to(REPO_ROOT).as_posix(): path + for path in tracked + if _kindless(path) + } + + missing = sorted(set(kindless) - set(GRANDFATHERED_KINDLESS_MANIFESTS)) + assert missing == [], ( + "kindless manifests outside the frozen list; declare `kind:` in them: " + f"{missing}" + ) + modified = sorted( + key + for key, path in kindless.items() + if manifest_digest(path) != GRANDFATHERED_KINDLESS_MANIFESTS[key] + ) + assert modified == [], ( + "frozen manifests modified without declaring `kind:`; declare it and " + f"drop them from the frozen list: {modified}" + ) + + +def test_the_frozen_list_only_shrinks(): + # Entries leave the list once a manifest declares its kind; none is added. + assert len(GRANDFATHERED_KINDLESS_MANIFESTS) <= FREEZE_SIZE + stale = sorted( + key + for key in GRANDFATHERED_KINDLESS_MANIFESTS + if not (REPO_ROOT / key).exists() or not _kindless(REPO_ROOT / key) + ) + assert stale == [], f"drop from the frozen list, the kind is declared: {stale}" + + +def _frozen_copy(tmp_path: Path) -> tuple[Path, str]: + key = "db/data/irs_soi/table_1_2/manifest.yaml" + assert key in GRANDFATHERED_KINDLESS_MANIFESTS + copy = tmp_path / key + copy.parent.mkdir(parents=True) + copy.write_bytes((REPO_ROOT / key).read_bytes()) + return copy, key + + +def test_a_frozen_manifest_reads_as_a_publisher_table_until_it_is_modified(tmp_path): + copy, key = _frozen_copy(tmp_path) + payload = yaml.safe_load(copy.read_text()) + + assert grandfathered_manifest_key(copy) == key + assert is_grandfathered_manifest(copy) + assert manifest_kind(payload, manifest_path=copy) == "publisher_table" + assert safe_manifest_kind(payload, manifest_path=copy) == ("publisher_table", None) + + copy.write_bytes(copy.read_bytes() + b"# touched\n") + assert not is_grandfathered_manifest(copy) + with pytest.raises(ManifestKindError, match="declares no kind"): + manifest_kind(payload, manifest_path=copy) + assert safe_manifest_kind(payload, manifest_path=copy) == ( + "publisher_table", + "manifest_kind_missing", + ) + + +def test_a_fetch_into_a_frozen_manifest_declares_its_kind(tmp_path): + copy, _key = _frozen_copy(tmp_path) + source = tmp_path / "24in12ms.xls" + source.write_bytes(b"the next vintage") + + report = fetch_source_artifact( + str(source), + source_id="irs_soi", + package_id="soi-table-1-2", + year=2099, + output_dir=copy.parent, + ) + payload = yaml.safe_load(copy.read_text()) + + assert report.valid + assert payload["kind"] == "publisher_table" + assert list(payload)[:3] == ["source_id", "package_id", "kind"] + # The manifest left the freeze by declaring itself. + assert not is_grandfathered_manifest(copy) + assert manifest_kind(payload, manifest_path=copy) == "publisher_table" + + +def _kindless_package(tmp_path: Path) -> Path: + package = tmp_path / "db" / "data" / "dwp" / "new_tables" + package.mkdir(parents=True) + (package / "table.ods").write_bytes(b"a table") + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-new-tables", + "files": { + 2023: { + "filename": "table.ods", + "source_url": "https://publisher.example/table.ods", + } + }, + }, + sort_keys=False, + ) + ) + return package + + +def test_a_new_kindless_manifest_is_refused_by_fetch_before_reading( + tmp_path, monkeypatch +): + package = _kindless_package(tmp_path) + original = (package / "manifest.yaml").read_bytes() + reads: list[str] = [] + monkeypatch.setattr( + "chronicle.artifacts._read_artifact", + lambda url: reads.append(url) or (b"x", "other.ods"), + ) + + for kind in (None, "publisher_table", "microdata_release"): + with pytest.raises(ManifestAccessError, match="declares no kind"): + fetch_source_artifact( + "https://publisher.example/other.ods", + source_id="dwp", + package_id="dwp-new-tables", + year=2024, + output_dir=package, + kind=kind, + licence="OGL-UK-3.0", + ) + + assert reads == [] + assert (package / "manifest.yaml").read_bytes() == original + + +def test_a_new_kindless_manifest_is_reported_and_skipped_by_the_sweeps( + tmp_path, monkeypatch +): + package = _kindless_package(tmp_path) + original = (package / "manifest.yaml").read_bytes() + monkeypatch.setattr( + "chronicle.artifacts._upload_r2_object", + lambda *args, **kwargs: (_ for _ in ()).throw(AssertionError("no upload")), + ) + + inventory = inventory_source_artifacts(tmp_path / "db" / "data") + published = publish_source_artifacts(tmp_path / "db" / "data") + + assert not inventory.valid + assert any(error.startswith("manifest_kind_missing") for error in inventory.errors) + assert not published.valid + assert published.entries == () + assert any(error.startswith("manifest_kind_missing") for error in published.errors) + assert (package / "manifest.yaml").read_bytes() == original + + +def test_a_new_kindless_manifest_is_refused_by_the_byte_reader(tmp_path, monkeypatch): + package_name = f"chronicle_kind_{uuid.uuid4().hex}" + resource_dir = tmp_path / "pkgroot" / package_name / "data" / "dwp" / "tables" + resource_dir.mkdir(parents=True) + (resource_dir / "table.csv").write_bytes(b"a,b\n1,2\n") + (resource_dir / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-tables", + "files": {2023: {"filename": "table.csv", "source_url": "x"}}, + } + ) + ) + monkeypatch.syspath_prepend(str(tmp_path / "pkgroot")) + monkeypatch.delitem(sys.modules, package_name, raising=False) + spec = SourceArtifactSpec( + source_name="dwp", + source_table="Tables", + resource_package=package_name, + resource_directory="data/dwp/tables", + manifest="manifest.yaml", + vintage="2023", + extracted_at="2026-09-02", + extraction_method="none", + parser="delimited_text_full_rows", + artifact_year=2023, + ) + + with pytest.raises(ManifestKindError): + spec.assert_parseable_manifest() + with pytest.raises(ManifestKindError): + spec.build_source_rows(2023) + + package_dir = tmp_path / "package" + package_dir.mkdir() + (package_dir / "source_package.yaml").write_text( + yaml.safe_dump( + { + "schema_version": "ledger.source_package.v1", + "package_id": "dwp-tables-parse-attempt", + "label": "A kindless manifest", + "artifact": { + "source_name": "dwp", + "source_table": "Tables", + "resource_package": package_name, + "resource_directory": "data/dwp/tables", + "manifest": "manifest.yaml", + "vintage": "2023", + "extracted_at": "2026-09-02", + "extraction_method": "none", + "parser": "delimited_text_full_rows", + "artifact_year": 2023, + }, + "record_sets": [], + }, + sort_keys=False, + ) + ) + report = validate_source_package(package_dir, year=2023) + assert not report.valid + assert [issue.code for issue in report.errors] == ["manifest_kind_missing"] + + +def test_no_tracked_microdata_bytes(): + """The repository guard: a release package holds manifests and nothing else.""" + release_dirs = { + path.parent + for path in _tracked_manifests() + if (yaml.safe_load(path.read_text()) or {}).get("kind") == "microdata_release" + } + assert release_dirs + + tracked = subprocess.run( + ["git", "-C", str(REPO_ROOT), "ls-files", "db/data"], + capture_output=True, + text=True, + check=True, + ).stdout.split() + for directory in release_dirs: + relative = directory.relative_to(REPO_ROOT).as_posix() + files = sorted( + Path(path).name + for path in tracked + if Path(path).parent.as_posix() == relative + ) + assert files == ["manifest.yaml"], f"{relative} tracks microdata bytes: {files}" + assert sorted(item.name for item in directory.iterdir()) == ["manifest.yaml"] + + +def test_the_committed_tree_passes_the_kind_rule(): + report = inventory_source_artifacts(REPO_ROOT / "db" / "data") + assert not [error for error in report.errors if "manifest_kind" in error] diff --git a/tests/test_chronicle_microdata_catalogue.py b/tests/test_chronicle_microdata_catalogue.py new file mode 100644 index 00000000..9f9d8213 --- /dev/null +++ b/tests/test_chronicle_microdata_catalogue.py @@ -0,0 +1,424 @@ +"""Hermetic tests for ``scripts/register_microdata_releases.py``. + +The catalogue is resolved against ``tests/fixtures/microcosm``: synthetic +consumer manifests in Microcosm's two shapes (staged ``source_stages.json`` +files and the flat ACS runtime manifest) that carry the reviewed pins +verbatim beside decoy artifacts the selectors must ignore. Nothing here reads +a checkout outside the repository, and nothing skips: a resolution failure is +a failure. + +``emit`` must reproduce the two committed hash-only manifests byte for byte +from the fixture, and ``plan`` must print exactly the golden commands in +``tests/fixtures/microcosm/golden_plan.json`` -- every reviewed identity as an +argument, every unknown as a ``TODO`` that the real CLI refuses to run. +""" + +from __future__ import annotations + +import argparse +import contextlib +import json +import shlex +import subprocess +import sys +from pathlib import Path + +import pytest +import yaml + +from chronicle.harness import main as harness_main + + +REPO_ROOT = Path(__file__).resolve().parents[1] +FIXTURE_ROOT = REPO_ROOT / "tests" / "fixtures" / "microcosm" +GOLDEN_PLAN = FIXTURE_ROOT / "golden_plan.json" +FRS_MANIFEST = REPO_ROOT / "db" / "data" / "dwp" / "frs_2023_24" / "manifest.yaml" +SPI_MANIFEST = ( + REPO_ROOT / "db" / "data" / "hmrc" / "spi_public_use_tape_2022_23" / "manifest.yaml" +) +UK_STAGES = "packages/microcosm-build/src/microcosm/build/uk/source_stages.json" +UK_HMRC_STAGES = ( + "packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json" +) +#: The commits the committed registrations were transcribed from: the last +#: commit that changed each consumer manifest. +PIN_COMMITS = { + UK_STAGES: "2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77", + UK_HMRC_STAGES: "de7451bd19ca46d2967e73cdf393908d29e72542", +} +PIN_COMMIT_ARGS = [ + arg + for path, commit in PIN_COMMITS.items() + for arg in ("--microcosm-commit", f"{path}={commit}") +] + +sys.path.insert(0, str(REPO_ROOT / "scripts")) +import register_microdata_releases as script # noqa: E402 + + +def _run(argv: list[str], capsys) -> tuple[int, str, str]: + exit_code = script.main(argv) + captured = capsys.readouterr() + return exit_code, captured.out, captured.err + + +def _harness_parser(monkeypatch) -> argparse.ArgumentParser: + """Capture the real ``chronicle`` harness parser.""" + captured: dict[str, argparse.ArgumentParser] = {} + + def capture(self, args=None, namespace=None): + captured["parser"] = self + raise SystemExit(0) + + monkeypatch.setattr(argparse.ArgumentParser, "parse_args", capture) + with contextlib.suppress(SystemExit): + harness_main(["--help"]) + monkeypatch.undo() + return captured["parser"] + + +# -------------------------------------------------------------------------- +# The fixture checkout +# -------------------------------------------------------------------------- + + +def test_fixture_checkout_holds_every_catalogued_manifest(): + catalogued = sorted({release.manifest for release in script.CATALOGUE}) + assert catalogued + for relative in catalogued: + path = FIXTURE_ROOT / relative + assert path.exists(), relative + payload = json.loads(path.read_text()) + assert payload["snapshot_of"]["repository"] == "PolicyEngine/microcosm" + + +def test_selection_ignores_decoys_and_resolves_every_catalogue_entry(): + resolved = script.resolve(FIXTURE_ROOT, script.CATALOGUE) + by_id = {item.release.release_id: item for item in resolved} + frs = yaml.safe_load(FRS_MANIFEST.read_text())["files"][2023] + committed = {entry["filename"]: entry for entry in frs} + + assert len(by_id) == len(script.CATALOGUE) + for tab in script.FRS_TABS: + item = by_id[f"dwp-frs-2023-24:{tab}"] + expected = committed[f"{tab}.tab"] + assert item.stage["stage"] == "frs_spine" + assert item.artifact["kind"] == "licensed_microdata" + assert (item.filename, item.sha256, item.size_bytes, item.vintage) == ( + f"{tab}.tab", + expected["sha256"], + expected["size_bytes"], + expected["vintage"], + ) + spi = by_id["hmrc-spi-public-use-tape-2022-23:put2223uk"] + spi_committed = yaml.safe_load(SPI_MANIFEST.read_text())["files"][2022][0] + assert (spi.filename, spi.sha256, spi.size_bytes, spi.vintage) == ( + "put2223uk.tab", + spi_committed["sha256"], + spi_committed["size_bytes"], + spi_committed["vintage"], + ) + assert spi.url is None + silc = by_id["statbel-be-silc-2023"] + assert (silc.sha256, silc.size_bytes, silc.filename) == (None, None, None) + assert by_id["census-cps-basic-monthly-2024"].url is None + assert by_id["census-sipp-2023"].url is None + assert by_id["federal-reserve-scf-2022-full"].sha256 is None + assert by_id["census-acs-pums-2024-household"].filename == "csv_hus.zip" + + +def test_selection_accepts_agreeing_duplicates_and_refuses_conflicts(): + payload = json.loads((FIXTURE_ROOT / UK_STAGES).read_text()) + selector = script.ArtifactSelector( + kind="licensed_microdata", match={"table": "adult"} + ) + stage, artifact = script.select_artifact(payload, selector, release_id="x") + # adult.tab is listed by two stages with identical bytes; the first wins. + assert stage["stage"] == "frs_spine" + + conflicting = json.loads(json.dumps(payload)) + conflicting["stages"][1]["artifacts"][0]["sha256"] = "f" * 64 + with pytest.raises(script.CatalogueError, match="conflicting values"): + script.select_artifact(conflicting, selector, release_id="x") + + with pytest.raises(script.CatalogueError, match="no Microcosm artifact matches"): + script.select_artifact( + payload, + script.ArtifactSelector(kind="licensed_microdata", match={"table": "nope"}), + release_id="x", + ) + + +def test_resolve_refuses_a_missing_consumer_manifest(tmp_path): + with pytest.raises(script.CatalogueError, match="manifest not found"): + script.resolve(tmp_path / "no-such-checkout", script.CATALOGUE[:1]) + + +# -------------------------------------------------------------------------- +# emit +# -------------------------------------------------------------------------- + + +def test_emit_from_the_fixture_reproduces_the_committed_manifests_byte_for_byte( + tmp_path, capsys +): + root = tmp_path / "data" + + exit_code, out, err = _run( + [ + "--microcosm-root", + str(FIXTURE_ROOT), + "--root", + str(root), + "--json", + "emit", + *PIN_COMMIT_ARGS, + ], + capsys, + ) + payload = json.loads(out) + + assert exit_code == 0, err + assert len(payload["registrations"]) == 15 + assert [blocker["release"] for blocker in payload["blockers"]] == [ + "statbel-be-silc-2023" + ] + assert "No hash is invented" in payload["blockers"][0]["reason"] + assert all(r["hash_source"] == "consumer_pin" for r in payload["registrations"]) + assert all(r["r2_location"] is None for r in payload["registrations"]) + written = sorted( + path.relative_to(root).as_posix() for path in root.rglob("*") if path.is_file() + ) + assert written == [ + "dwp/frs_2023_24/manifest.yaml", + "hmrc/spi_public_use_tape_2022_23/manifest.yaml", + ] + assert (root / "dwp/frs_2023_24/manifest.yaml").read_bytes() == ( + FRS_MANIFEST.read_bytes() + ) + assert (root / "hmrc/spi_public_use_tape_2022_23/manifest.yaml").read_bytes() == ( + SPI_MANIFEST.read_bytes() + ) + + +def test_emit_is_idempotent_over_the_committed_manifests(tmp_path, capsys): + root = tmp_path / "data" + for manifest in (FRS_MANIFEST, SPI_MANIFEST): + target = root / manifest.relative_to(REPO_ROOT / "db" / "data") + target.parent.mkdir(parents=True) + target.write_bytes(manifest.read_bytes()) + + exit_code, out, _err = _run( + [ + "--microcosm-root", + str(FIXTURE_ROOT), + "--root", + str(root), + "--json", + "emit", + *PIN_COMMIT_ARGS, + ], + capsys, + ) + + assert exit_code == 0 + assert all(r["replaced"] for r in json.loads(out)["registrations"]) + assert ( + root / "dwp/frs_2023_24/manifest.yaml" + ).read_bytes() == FRS_MANIFEST.read_bytes() + + +def test_emit_refuses_a_pin_that_drifted_from_the_committed_one(tmp_path, capsys): + root = tmp_path / "data" + target = root / "dwp" / "frs_2023_24" / "manifest.yaml" + target.parent.mkdir(parents=True) + target.write_bytes(FRS_MANIFEST.read_bytes()) + checkout = tmp_path / "consumer" + for path in FIXTURE_ROOT.rglob("*.json"): + copy = checkout / path.relative_to(FIXTURE_ROOT) + copy.parent.mkdir(parents=True, exist_ok=True) + copy.write_bytes(path.read_bytes()) + stages = json.loads((checkout / UK_STAGES).read_text()) + for stage in stages["stages"]: + for artifact in stage["artifacts"]: + if artifact.get("table") == "adult": + artifact["sha256"] = "f" * 64 + (checkout / UK_STAGES).write_text(json.dumps(stages)) + + exit_code, _out, err = _run( + [ + "--microcosm-root", + str(checkout), + "--root", + str(root), + "emit", + *PIN_COMMIT_ARGS, + ], + capsys, + ) + + assert exit_code == 1 + assert "pass --allow-reissue" in err + assert target.read_bytes() == FRS_MANIFEST.read_bytes() + + +def test_emit_needs_a_commit_it_can_read_or_be_told(tmp_path, capsys): + exit_code, _out, err = _run( + [ + "--microcosm-root", + str(FIXTURE_ROOT), + "--root", + str(tmp_path / "data"), + "emit", + ], + capsys, + ) + assert exit_code == 1 + assert "pass --microcosm-commit" in err + assert not (tmp_path / "data").exists() + + exit_code, _out, err = _run( + [ + "--microcosm-root", + str(FIXTURE_ROOT), + "--root", + str(tmp_path / "data"), + "emit", + "--microcosm-commit", + "not-a-commit", + ], + capsys, + ) + assert exit_code == 2 + assert "40-hex commit" in err + + +def test_pin_commit_reads_the_last_commit_that_changed_the_file(tmp_path): + repo = tmp_path / "consumer" + (repo / "build").mkdir(parents=True) + git = ["git", "-C", str(repo)] + subprocess.run([*git, "init", "-q"], check=True) + subprocess.run([*git, "config", "user.email", "t@example.com"], check=True) + subprocess.run([*git, "config", "user.name", "t"], check=True) + (repo / "build" / "stages.json").write_text("{}") + subprocess.run([*git, "add", "."], check=True) + subprocess.run([*git, "commit", "-q", "-m", "pin"], check=True) + pinned = subprocess.run( + [*git, "rev-parse", "HEAD"], capture_output=True, text=True, check=True + ).stdout.strip() + (repo / "other.txt").write_text("unrelated") + subprocess.run([*git, "add", "."], check=True) + subprocess.run([*git, "commit", "-q", "-m", "unrelated"], check=True) + + assert script.pin_commit(repo, "build/stages.json") == pinned + with pytest.raises(script.CatalogueError, match="records no commit"): + script.pin_commit(repo, "build/missing.json") + + +# -------------------------------------------------------------------------- +# plan +# -------------------------------------------------------------------------- + + +def test_plan_matches_the_golden_commands_exactly(capsys): + exit_code, out, _err = _run( + ["--microcosm-root", str(FIXTURE_ROOT), "--root", "db/data", "--json", "plan"], + capsys, + ) + golden = json.loads(GOLDEN_PLAN.read_text()) + + assert exit_code == 0 + assert json.loads(out) == golden + assert len(golden["commands"]) == 9 + + +def test_every_planned_command_parses_with_the_reviewed_identity_as_arguments( + monkeypatch, +): + parser = _harness_parser(monkeypatch) + golden = json.loads(GOLDEN_PLAN.read_text()) + + for command in golden["commands"]: + argv = shlex.split(command) + assert argv[:3] == ["uv", "run", "chronicle"] + namespace = parser.parse_args(argv[3:]) + assert namespace.command == "fetch-artifact" + assert namespace.access == "public" + assert namespace.kind == "microdata_release" + assert namespace.licence == "US-Government-Work" + assert namespace.publisher + assert namespace.vintage + assert namespace.expected_sha256 + assert namespace.licence_evidence_issuer + assert namespace.licence_evidence_scope + assert namespace.licence_evidence_url + assert namespace.upload_r2 is True + + +def test_plan_prints_a_todo_and_never_a_guess(capsys): + golden = json.loads(GOLDEN_PLAN.read_text()) + by_release = { + shlex.split(command)[shlex.split(command).index("--package-id") + 1]: command + for command in golden["commands"] + } + + # Every reviewed publisher checksum travels as an argument. + assert ( + "--expected-sha256 d2e000250782adfbdd7f29c82b66d866591a30f0d330496698ec19f9c784ce11" + in by_release["census-cps-asec-2023"] + ) + assert "--expected-size-bytes 150165063" in by_release["census-cps-asec-2023"] + # Unknowns are TODOs the CLI refuses: no URL, no publisher-bytes checksum, + # no catalogued evidence URL. + assert "--url TODO_PUBLISHER_URL" in by_release["census-cps-basic-monthly-2024"] + assert "--url TODO_PUBLISHER_URL" in by_release["census-sipp-2023"] + for package_id in ( + "census-cps-basic-monthly-2024", + "census-acs-pums-2022-1yr", + "federal-reserve-scf-2022", + "census-sipp-2023", + ): + assert any( + "--expected-sha256 TODO_REVIEWED_SHA256" in command + for package, command in by_release.items() + if package == package_id + ) + assert all("TODO_EVIDENCE_URL" in command for command in golden["commands"]) + assert not any("TODO_VINTAGE" in command for command in golden["commands"]) + # The derived-h5 hash Microcosm pins for ACS 2022 never masquerades as + # the publisher checksum. + assert ( + "0b319b496f19a6913066f9c5ea572edfda3d78a187be6f375846617d0b441bd4" + not in "\n".join(golden["commands"]) + ) + + +def test_a_planned_command_with_a_todo_is_refused_before_anything_is_read( + tmp_path, monkeypatch, capsys +): + golden = json.loads(GOLDEN_PLAN.read_text()) + reads: list[str] = [] + + def unexpected_read(source_url): + reads.append(source_url) + raise AssertionError("a planned command with a TODO must not read") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + monkeypatch.setattr( + "chronicle.artifacts._upload_r2_object", + lambda *args, **kwargs: (_ for _ in ()).throw(AssertionError("no upload")), + ) + + for command in golden["commands"]: + argv = shlex.split(command)[3:] + out_index = argv.index("--out-dir") + 1 + argv[out_index] = str(tmp_path / argv[out_index]) + exit_code = harness_main([*argv, "--staging-dir", str(tmp_path / "staging")]) + captured = capsys.readouterr() + assert exit_code == 1 + assert captured.out == "" + assert "error:" in captured.err + + assert reads == [] + assert not (tmp_path / "db").exists() + assert not (tmp_path / "staging").exists() diff --git a/tests/test_chronicle_microdata_registration.py b/tests/test_chronicle_microdata_registration.py index f608b1bc..53797e35 100644 --- a/tests/test_chronicle_microdata_registration.py +++ b/tests/test_chronicle_microdata_registration.py @@ -4,32 +4,50 @@ stores the bytes of only the ones a publisher permits it to redistribute (``docs/adr-chronicle-raw-microdata-identity.md``). These tests pin the whole refusal surface: which access classes may carry bytes, which commands refuse -them, and that a hash-only registration is a complete, valid artifact record -with no local file and no R2 key. +them, that every refusal happens before any filesystem or network side effect, +and that a hash-only registration is a complete, valid artifact record with no +local file and no R2 key. + +The catalogue and the checked-in synthetic consumer manifests are covered in +``tests/test_chronicle_microdata_catalogue.py``; the explicit-kind rule and the +repository guard in ``tests/test_chronicle_manifest_kind.py``. """ from __future__ import annotations import hashlib import json +import sys +import uuid from pathlib import Path import pytest import yaml from chronicle.artifacts import ( + ArtifactCommandResult, + ExpectedArtifactIdentityError, + MalformedManifestError, + SourceArtifactRevisionError, fetch_source_artifact, inventory_source_artifacts, + microdata_staging_path, publish_source_artifacts, ) from chronicle.harness import main as harness_main from chronicle.registration import ( ACCESS_CLASSES, + HASH_SOURCES, + AmbiguousVintageKeyError, + ArtifactFilenameError, HashOnlyRegistrationError, ListSpecRejected, ManifestAccessError, MicrodataReleaseNotParseableError, + bare_filename, entry_access, + filename_key, + is_bare_filename, is_hash_only, is_microdata_release, iter_file_specs, @@ -37,21 +55,63 @@ normalize_access, register_hash_only_artifact, registration_id, + resolve_vintage_key, safe_entry_access, stores_bytes, validate_file_entry, + validate_manifest_files, + vintage_key_forms, +) +from chronicle.source_package import ( + SOURCE_ARTIFACT_CACHE_ENV, + SOURCE_ARTIFACT_FETCH_ENV, + SourceArtifactSpec, + _read_source_artifact_content, + validate_source_package, ) -from chronicle.source_package import SourceArtifactSpec, validate_source_package REPO_ROOT = Path(__file__).resolve().parents[1] FRS_PACKAGE = REPO_ROOT / "db" / "data" / "dwp" / "frs_2023_24" SPI_PACKAGE = REPO_ROOT / "db" / "data" / "hmrc" / "spi_public_use_tape_2022_23" -# Read-only consumer checkout the emit/plan script resolves pins against. -MICROCOSM_ROOT = Path.home() / "PolicyEngine" / "populace" # A syntactically valid checksum that identifies no real publisher bytes. FIXTURE_SHA = "a" * 64 +OTHER_SHA = "b" * 64 +FIXTURE_COMMIT = "2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77" + +ATTESTED = { + "hash_source": "consumer_attested", + "attested_by": "PolicyEngine/microcosm", + "attestation_evidence": ( + "microcosm uk/source_stages.json frs_spine pin, verified against the " + "licensed copy" + ), + "verified_at": "2026-09-02", +} +PINNED = { + "hash_source": "consumer_pin", + "attested_by": "PolicyEngine/microcosm", + "pinned_from": { + "repository": "PolicyEngine/microcosm", + "path": "packages/microcosm-build/src/microcosm/build/uk/source_stages.json", + "commit": FIXTURE_COMMIT, + }, +} + +# Evidence binding a public release to an allowlisted term. The URL is a +# test value; the check is that it is a durable http(s) location. +EVIDENCE = { + "issuer": "U.S. Census Bureau", + "scope": ( + "Public-use microdata file published by the U.S. Census Bureau, a " + "federal agency; a work of the United States Government under 17 " + "U.S.C. §105." + ), + "url": "https://evidence.example/census/public-use-files", +} +PUBLIC_BYTES = b"public household pums" +PUBLIC_SHA = hashlib.sha256(PUBLIC_BYTES).hexdigest() def _register(output_dir: Path, **overrides: object) -> object: @@ -68,14 +128,106 @@ def _register(output_dir: Path, **overrides: object) -> object: "vintage": "2023_24", "size_bytes": 35323384, "doi": "10.5255/UKDA-SN-9367-2", - "verified_at": "2026-09-02", + **ATTESTED, } kwargs.update(overrides) return register_hash_only_artifact(**kwargs) # type: ignore[arg-type] +def _manifest(output_dir: Path) -> dict: + return yaml.safe_load((output_dir / "manifest.yaml").read_text()) + + +def _refuse_read(monkeypatch, message: str = "the publisher was read") -> list: + """Make any publisher read an ordering violation; return the read log.""" + reads: list[str] = [] + + def unexpected_read(source_url): + reads.append(source_url) + raise AssertionError(f"ORDERING VIOLATION: {message} before the refusal") + + monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + return reads + + +def _serve(monkeypatch, content: bytes, filename: str | None = None) -> list: + """Serve ``content`` for any URL, recording each read.""" + reads: list[str] = [] + + def fake_read(source_url): + reads.append(source_url) + return content, filename or Path(source_url).name + + monkeypatch.setattr("chronicle.artifacts._read_artifact", fake_read) + return reads + + +def _record_uploads(monkeypatch) -> list[tuple[str, str]]: + uploads: list[tuple[str, str]] = [] + + def fake_upload(location, local_path, *, wrangler_command): + uploads.append((location.uri, str(local_path))) + return ArtifactCommandResult( + command=("stub",), returncode=0, stdout="ok", stderr="" + ) + + monkeypatch.setattr("chronicle.artifacts._upload_r2_object", fake_upload) + return uploads + + +def _forbid_uploads(monkeypatch) -> None: + monkeypatch.setattr( + "chronicle.artifacts._upload_r2_object", + lambda *args, **kwargs: (_ for _ in ()).throw( + AssertionError("nothing may be uploaded on the way to a refusal") + ), + ) + + +def _fetch_table(source: Path, output_dir: Path, **overrides: object): + kwargs: dict[str, object] = { + "source_id": "irs_soi", + "package_id": "soi-table-1-2", + "year": 2023, + "output_dir": output_dir, + } + kwargs.update(overrides) + return fetch_source_artifact(str(source), **kwargs) # type: ignore[arg-type] + + +def _fetch_release( + output_dir: Path, + *, + staging_dir: Path, + filename: str = "csv_hus.zip", + content: bytes = PUBLIC_BYTES, + **overrides: object, +): + """Fetch a public microdata release with complete evidence.""" + kwargs: dict[str, object] = { + "source_id": "census_acs", + "package_id": "census-acs-pums-2022-1yr", + "year": 2022, + "output_dir": output_dir, + "filename": filename, + "licence": "US-Government-Work", + "access": "public", + "kind": "microdata_release", + "publisher": "U.S. Census Bureau", + "vintage": "2022", + "expected_sha256": hashlib.sha256(content).hexdigest(), + "licence_evidence": EVIDENCE, + "staging_dir": staging_dir, + } + kwargs.update(overrides) + return fetch_source_artifact( + f"https://publisher.example/pums/{filename}", + **kwargs, # type: ignore[arg-type] + ) + + # -------------------------------------------------------------------------- -# Access and kind vocabularies +# Access, kind, filename and vintage vocabularies # -------------------------------------------------------------------------- @@ -89,7 +241,7 @@ def test_only_public_access_may_carry_bytes(access): assert is_hash_only(access) is (access != "public") -def test_absent_access_is_inferred_public(): +def test_absent_access_is_inferred_public_for_a_table_entry(): assert normalize_access(None) == "public" assert entry_access({"filename": "table.xlsx"}) == "public" @@ -104,12 +256,19 @@ def test_unknown_access_class_falls_back_to_restricted_not_public(): assert safe_entry_access({"access": "internal"}) == "restricted" -def test_manifest_kind_defaults_to_publisher_table_and_rejects_unknown(): +def test_manifest_kind_is_explicit_except_for_an_absent_manifest(): assert manifest_kind(None) == "publisher_table" assert manifest_kind({}) == "publisher_table" assert is_microdata_release({"kind": "microdata_release"}) is True with pytest.raises(ManifestAccessError, match="Unknown manifest kind"): manifest_kind({"kind": "microdata_rows"}) + # A manifest with content and no kind is an error, never a table. + with pytest.raises(ManifestAccessError, match="declares no kind"): + manifest_kind({"files": {2023: {"filename": "table.xlsx"}}}) + + +def test_hash_sources_are_the_closed_contract_set(): + assert HASH_SOURCES == ("chronicle_fetch", "consumer_attested", "consumer_pin") def test_registration_identity_is_the_contract_tuple(): @@ -125,12 +284,50 @@ def test_registration_identity_is_the_contract_tuple(): ) +@pytest.mark.parametrize( + "alias", + [ + "./adult.tab", + "sub/../adult.tab", + "adult.tab/", + "../pkg/adult.tab", + "/x/adult.tab", + ".", + "..", + " adult.tab", + "", + ], +) +def test_bare_filename_refuses_every_alias(alias): + assert not is_bare_filename(alias) + with pytest.raises(ArtifactFilenameError, match="bare filename"): + bare_filename(alias) + + +def test_bare_filename_accepts_a_plain_name_and_keys_case_folded(): + assert bare_filename("adult.tab") == "adult.tab" + assert filename_key("ADULT.TAB") == filename_key("adult.tab") + assert filename_key("./Adult.tab") == "adult.tab" + + +def test_vintage_key_forms_pair_the_two_spellings_of_a_year_only(): + assert vintage_key_forms(2023) == (2023, "2023") + assert vintage_key_forms("2023") == ("2023", 2023) + assert vintage_key_forms("A_1") == ("A_1",) + assert vintage_key_forms("0123") == ("0123",) + assert resolve_vintage_key({"2023": {}}, 2023) == "2023" + assert resolve_vintage_key({2023: {}}, "2023") == 2023 + assert resolve_vintage_key({}, 2023) is None + with pytest.raises(AmbiguousVintageKeyError, match="both keys"): + resolve_vintage_key({2023: {}, "2023": {}}, 2023) + + # -------------------------------------------------------------------------- -# Manifest entry validation +# Entry and manifest validation vocabulary # -------------------------------------------------------------------------- -def test_microdata_release_entry_requires_access_and_licence(): +def test_microdata_release_entry_requires_access_licence_and_attestation(): errors = validate_file_entry( {"filename": "adult.tab", "sha256": FIXTURE_SHA}, kind="microdata_release", @@ -140,6 +337,21 @@ def test_microdata_release_entry_requires_access_and_licence(): assert "missing_access" in errors assert "missing_licence" in errors + assert "missing_hash_source" in errors + + +def _attested_entry(**mutation: object) -> dict: + entry = { + "filename": "adult.tab", + "access": "licensed", + "licence": "UK Data Service End User Licence", + "vintage": "2023_24", + "sha256": FIXTURE_SHA, + "doi": "10.5255/UKDA-SN-9367-2", + **ATTESTED, + } + entry.update(mutation) + return {key: value for key, value in entry.items() if value is not None} @pytest.mark.parametrize( @@ -151,24 +363,17 @@ def test_microdata_release_entry_requires_access_and_licence(): ({"sha256": FIXTURE_SHA.upper()}, "malformed_sha256"), ({"vintage": None}, "missing_vintage"), ({"doi": None}, "missing_access_route"), - ({"verified_at": None}, "missing_verification_timestamp"), + ({"hash_source": None}, "missing_hash_source"), + ({"hash_source": "transcribed"}, "unknown_hash_source:transcribed"), + ({"attested_by": None}, "missing_attested_by"), + ({"attestation_evidence": None}, "missing_attestation_evidence"), + ({"verified_at": None}, "missing_verified_at"), + ({"filename": "./adult.tab"}, "non_canonical_filename:./adult.tab"), ], ) def test_hash_only_entry_reports_each_missing_field(mutation, expected_code): - entry = { - "filename": "adult.tab", - "access": "licensed", - "licence": "UK Data Service End User Licence", - "vintage": "2023_24", - "sha256": FIXTURE_SHA, - "doi": "10.5255/UKDA-SN-9367-2", - "verified_at": "2026-09-02", - } - entry.update(mutation) - entry = {key: value for key, value in entry.items() if value is not None} - errors = validate_file_entry( - entry, + _attested_entry(**mutation), kind="microdata_release", manifest={}, local_file_exists=False, @@ -177,17 +382,53 @@ def test_hash_only_entry_reports_each_missing_field(mutation, expected_code): assert expected_code in errors +@pytest.mark.parametrize( + ("mutation", "expected_code"), + [ + ({"pinned_from": None}, "missing_pinned_from"), + ({"pinned_from": "microcosm@abc"}, "malformed_pinned_from"), + ( + {"pinned_from": {"repository": "PolicyEngine/microcosm"}}, + "pinned_from_missing_field:commit", + ), + ( + {"pinned_from": {**PINNED["pinned_from"], "commit": "abc123"}}, + "malformed_pinned_from_commit", + ), + ({"verified_at": "2026-09-02"}, "verified_at_forbidden_for_consumer_pin"), + ], +) +def test_consumer_pin_entry_reports_its_own_fields(mutation, expected_code): + entry = _attested_entry(attestation_evidence=None, verified_at=None, **PINNED) + entry.update(mutation) + entry = {key: value for key, value in entry.items() if value is not None} + + errors = validate_file_entry( + entry, kind="microdata_release", manifest={}, local_file_exists=False + ) + + assert expected_code in errors + + +def test_chronicle_fetch_entry_is_attested_by_chronicle_with_a_date(): + entry = _attested_entry( + hash_source="chronicle_fetch", attested_by="microcosm", verified_at=None + ) + errors = validate_file_entry( + entry, kind="microdata_release", manifest={}, local_file_exists=False + ) + assert "attested_by_not_chronicle" in errors + assert "missing_verified_at" in errors + + def test_hash_only_entry_flags_bytes_and_r2_locations(): - entry = { - "filename": "adult.tab", - "access": "restricted", - "licence": "Statbel/Eurostat scientific-use", - "vintage": "2023", - "sha256": FIXTURE_SHA, - "doi": "10.5255/UKDA-SN-9422-1", - "verified_at": "2026-09-02", - "storage": {"r2": {"bucket": "ledger-raw", "key": "raw/x/y/z"}}, - } + entry = _attested_entry( + access="restricted", + storage={ + "r2": {"bucket": "ledger-raw", "key": "raw/x/y/z"}, + "previous_r2": [{"uri": "r2://ledger-raw/raw/x/y/old"}], + }, + ) errors = validate_file_entry( entry, @@ -198,9 +439,10 @@ def test_hash_only_entry_flags_bytes_and_r2_locations(): assert "bytes_present_for_hash_only_entry" in errors assert "r2_location_for_hash_only_entry" in errors + assert "r2_history_for_hash_only_entry" in errors -def test_public_entry_needs_no_licence_or_access_route(): +def test_public_table_entry_needs_no_licence_access_route_or_attestation(): assert ( validate_file_entry( {"filename": "table.xlsx", "sha256": FIXTURE_SHA}, @@ -212,6 +454,114 @@ def test_public_entry_needs_no_licence_or_access_route(): ) +def _public_release_entry(**mutation: object) -> dict: + entry = { + "filename": "csv_hus.zip", + "access": "public", + "licence": "US-Government-Work", + "licence_evidence": { + **EVIDENCE, + "licence": "US-Government-Work", + "sha256": PUBLIC_SHA, + }, + "vintage": "2022", + "sha256": PUBLIC_SHA, + "hash_source": "chronicle_fetch", + "attested_by": "chronicle", + "verified_at": "2026-09-03", + } + entry.update(mutation) + return {key: value for key, value in entry.items() if value is not None} + + +def test_public_release_entry_is_complete_with_allowlisted_evidence(): + assert ( + validate_file_entry( + _public_release_entry(), + kind="microdata_release", + manifest={}, + local_file_exists=False, + ) + == () + ) + + +@pytest.mark.parametrize( + ("mutation", "expected_code"), + [ + ( + {"licence": "U.S. Census Bureau public-use file"}, + "licence_not_redistributable:U.S. Census Bureau public-use file", + ), + ({"licence_evidence": None}, "missing_licence_evidence"), + ({"licence_evidence": "see website"}, "malformed_licence_evidence"), + ( + { + "licence_evidence": { + **EVIDENCE, + "licence": "US-Government-Work", + "sha256": PUBLIC_SHA, + "url": "", + } + }, + "licence_evidence_missing_field:url", + ), + ( + { + "licence_evidence": { + **EVIDENCE, + "licence": "CC0-1.0", + "sha256": PUBLIC_SHA, + } + }, + "licence_evidence_licence_mismatch", + ), + ( + { + "licence_evidence": { + **EVIDENCE, + "licence": "US-Government-Work", + "sha256": OTHER_SHA, + } + }, + "licence_evidence_sha256_mismatch", + ), + ( + { + "licence_evidence": { + **EVIDENCE, + "licence": "US-Government-Work", + "sha256": PUBLIC_SHA, + "url": "see the website", + } + }, + "licence_evidence_url_not_durable", + ), + ({"vintage": None}, "missing_vintage"), + ({"sha256": None}, "missing_sha256"), + ({"hash_source": None}, "missing_hash_source"), + ], +) +def test_public_release_entry_reports_missing_evidence(mutation, expected_code): + errors = validate_file_entry( + _public_release_entry(**mutation), + kind="microdata_release", + manifest={}, + local_file_exists=False, + ) + assert expected_code in errors + + +def test_public_release_bytes_beside_the_manifest_are_a_repository_violation(): + errors = validate_file_entry( + _public_release_entry(), + kind="microdata_release", + manifest={}, + local_file_exists=True, + ) + assert "bytes_present_for_microdata_release_entry" in errors + + def test_list_file_spec_expands_only_for_a_microdata_release(): specs = [{"filename": "adult.tab"}, {"filename": "child.tab"}] @@ -228,32 +578,95 @@ def test_list_file_spec_expands_only_for_a_microdata_release(): ) == ("list_file_spec_requires_microdata_release_kind",) +def test_manifest_level_validation_reports_keys_names_and_collisions(): + manifest = { + "files": { + 2023: [ + {"filename": "adult.tab", "access": "licensed", "sha256": FIXTURE_SHA}, + {"filename": "./adult.tab", "access": "public", "sha256": OTHER_SHA}, + {"filename": "child.tab", "sha256": FIXTURE_SHA}, + {"filename": "child.tab", "sha256": FIXTURE_SHA}, + ], + "2023": {"filename": "other.csv"}, + 2022: {"filename": "table.xlsx", "sha256": FIXTURE_SHA}, + 2021: {"filename": "TABLE.xlsx", "sha256": OTHER_SHA}, + } + } + + errors = validate_manifest_files(manifest) + + assert "duplicate_vintage_key:2023" in errors + assert "non_canonical_filename:./adult.tab" in errors + assert "filename_collision:adult.tab" in errors + assert "duplicate_filename_in_vintage:child.tab" in errors + # Two public entries, one path, different bytes: the tree can hold one. + assert "filename_collision:table.xlsx" in errors + + +def test_manifest_level_validation_accepts_the_same_file_under_two_keys(): + # The SSA manifests register one file under a year and a label key. + manifest = { + "files": { + 2024: {"filename": "ssa.csv", "sha256": FIXTURE_SHA}, + "extracted_targets": {"filename": "ssa.csv", "sha256": FIXTURE_SHA}, + } + } + assert validate_manifest_files(manifest) == () + + +def test_manifest_level_validation_accepts_a_hash_only_reissue(): + manifest = { + "files": { + 2023: [ + {"filename": "adult.tab", "access": "licensed", "sha256": FIXTURE_SHA}, + {"filename": "adult.tab", "access": "licensed", "sha256": OTHER_SHA}, + ] + } + } + assert validate_manifest_files(manifest) == () + assert validate_manifest_files({"files": ["not", "a", "mapping"]}) == ( + "files_not_a_mapping", + ) + + # -------------------------------------------------------------------------- # register-artifact # -------------------------------------------------------------------------- def test_register_writes_identity_without_bytes_or_an_r2_key(tmp_path): - output_dir = tmp_path / "dwp" / "frs_2023_24" + output_dir = tmp_path / "pkg" report = _register(output_dir) - - manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) + manifest = _manifest(output_dir) entry = manifest["files"][2023][0] assert report.valid - assert report.registration == (f"dwp/dwp-frs-2023-24/2023/{FIXTURE_SHA}/adult.tab") - assert report.to_dict()["r2_location"] is None + assert report.replaced is False + assert report.hash_source == "consumer_attested" + assert report.attested_by == "PolicyEngine/microcosm" + assert report.registration == f"dwp/dwp-frs-2023-24/2023/{FIXTURE_SHA}/adult.tab" assert manifest["kind"] == "microdata_release" assert entry["access"] == "licensed" - assert entry["licence"] == "UK Data Service End User Licence" assert entry["sha256"] == FIXTURE_SHA - assert entry["size_bytes"] == 35323384 - assert entry["vintage"] == "2023_24" + assert entry["hash_source"] == "consumer_attested" + assert entry["attested_by"] == "PolicyEngine/microcosm" + assert entry["attestation_evidence"] == ATTESTED["attestation_evidence"] assert entry["verified_at"] == "2026-09-02" assert "storage" not in entry - # The only file the registration creates is the manifest itself. - assert sorted(p.name for p in output_dir.iterdir()) == ["manifest.yaml"] + assert sorted(path.name for path in output_dir.iterdir()) == ["manifest.yaml"] + + +def test_register_a_consumer_pin_records_where_it_was_read(tmp_path): + output_dir = tmp_path / "pkg" + + _register(output_dir, attestation_evidence=None, verified_at=None, **PINNED) + entry = _manifest(output_dir)["files"][2023][0] + + assert entry["hash_source"] == "consumer_pin" + assert entry["pinned_from"] == PINNED["pinned_from"] + assert "verified_at" not in entry + assert list(entry).index("hash_source") < list(entry).index("pinned_from") def test_register_refuses_public_access(tmp_path): @@ -273,13 +686,35 @@ def test_register_never_invents_a_hash(tmp_path, sha256): [ ({"licence": ""}, "must record the publisher licence"), ({"vintage": ""}, "must record the artifact vintage"), - ({"verified_at": None}, "must record when the checksum was"), ({"filename": "../adult.tab"}, "must be a bare filename"), + ({"filename": "./adult.tab"}, "must be a bare filename"), + ({"hash_source": "chronicle_fetch"}, "does not apply"), + ({"hash_source": "guessed"}, "Unknown hash_source"), + ({"attested_by": ""}, "pass --attested-by"), + ({"attestation_evidence": None}, "pass --attestation-evidence"), + ({"verified_at": None}, "pass --verified-at"), ], ) def test_register_refuses_an_incomplete_registration(tmp_path, overrides, expected): with pytest.raises(HashOnlyRegistrationError, match=expected): _register(tmp_path / "pkg", **overrides) + assert not (tmp_path / "pkg").exists() + + +@pytest.mark.parametrize( + ("overrides", "expected"), + [ + ({"pinned_from": None}, "pass --pinned-from-repository"), + ({"pinned_from": {"repository": "PolicyEngine/microcosm"}}, "40-hex commit"), + ({"verified_at": "2026-09-02"}, "carries no verified_at"), + ], +) +def test_register_refuses_an_incomplete_consumer_pin(tmp_path, overrides, expected): + kwargs = {"hash_source": None, "attestation_evidence": None, "verified_at": None} + kwargs.update(PINNED) + kwargs.update(overrides) + with pytest.raises(HashOnlyRegistrationError, match=expected): + _register(tmp_path / "pkg", **kwargs) def test_register_refuses_without_an_access_route(tmp_path): @@ -297,6 +732,24 @@ def test_register_refuses_while_the_bytes_are_present(tmp_path): def test_register_refuses_a_publisher_table_manifest(tmp_path): + output_dir = tmp_path / "pkg" + output_dir.mkdir() + (output_dir / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "publisher_table", + "files": {2023: {"filename": "table.ods", "sha256": FIXTURE_SHA}}, + } + ) + ) + + with pytest.raises(HashOnlyRegistrationError, match="publisher_table manifest"): + _register(output_dir) + + +def test_register_refuses_a_kindless_manifest_with_content(tmp_path): output_dir = tmp_path / "pkg" output_dir.mkdir() (output_dir / "manifest.yaml").write_text( @@ -308,24 +761,27 @@ def test_register_refuses_a_publisher_table_manifest(tmp_path): } ) ) + original = (output_dir / "manifest.yaml").read_bytes() - with pytest.raises(HashOnlyRegistrationError, match="is a publisher_table"): + with pytest.raises(HashOnlyRegistrationError, match="declares no kind"): _register(output_dir) + assert (output_dir / "manifest.yaml").read_bytes() == original + def test_register_refuses_a_manifest_for_a_different_source(tmp_path): output_dir = tmp_path / "pkg" _register(output_dir) with pytest.raises(HashOnlyRegistrationError, match="declares source_id='dwp'"): - _register(output_dir, source_id="hmrc") + _register(output_dir, source_id="ons", filename="other.tab") def test_register_is_idempotent_and_byte_stable(tmp_path): output_dir = tmp_path / "pkg" - _register(output_dir) first = (output_dir / "manifest.yaml").read_bytes() + report = _register(output_dir) assert report.replaced is True @@ -335,128 +791,1047 @@ def test_register_is_idempotent_and_byte_stable(tmp_path): def test_register_refuses_a_reissue_unless_it_is_asked_for(tmp_path): output_dir = tmp_path / "pkg" _register(output_dir) + original = (output_dir / "manifest.yaml").read_bytes() - with pytest.raises(HashOnlyRegistrationError, match="already registers"): - _register(output_dir, sha256="b" * 64) + with pytest.raises(HashOnlyRegistrationError, match="pass --allow-reissue"): + _register(output_dir, sha256=OTHER_SHA) + assert (output_dir / "manifest.yaml").read_bytes() == original - _register(output_dir, sha256="b" * 64, allow_reissue=True) - manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) - entries = manifest["files"][2023] + report = _register(output_dir, sha256=OTHER_SHA, allow_reissue=True) + entries = _manifest(output_dir)["files"][2023] - # A reissue is a new publisher release, so both registrations survive. - assert [entry["sha256"] for entry in entries] == [FIXTURE_SHA, "b" * 64] + assert report.replaced is False + assert [entry["sha256"] for entry in entries] == [FIXTURE_SHA, OTHER_SHA] def test_register_keeps_distinct_files_under_one_vintage(tmp_path): output_dir = tmp_path / "pkg" + _register(output_dir, filename="adult.tab") + _register(output_dir, filename="child.tab", sha256=OTHER_SHA) - _register(output_dir, filename="adult.tab", sha256=FIXTURE_SHA) - _register(output_dir, filename="child.tab", sha256="c" * 64) + entries = _manifest(output_dir)["files"][2023] - manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) + assert [entry["filename"] for entry in entries] == ["adult.tab", "child.tab"] - assert [entry["filename"] for entry in manifest["files"][2023]] == [ - "adult.tab", - "child.tab", - ] +def test_reregistering_the_current_pin_stays_idempotent_after_a_reissue(tmp_path): + output_dir = tmp_path / "pkg" + _register(output_dir, sha256=FIXTURE_SHA) + _register(output_dir, sha256=OTHER_SHA, allow_reissue=True) -# -------------------------------------------------------------------------- -# fetch-artifact -# -------------------------------------------------------------------------- + report = _register(output_dir, sha256=OTHER_SHA) + assert report.replaced is True + entries = _manifest(output_dir)["files"][2023] + assert [entry["sha256"] for entry in entries] == [FIXTURE_SHA, OTHER_SHA] -def test_fetch_refuses_a_hash_only_access_class(tmp_path, monkeypatch): - source = tmp_path / "adult.tab" - source.write_bytes(b"licensed microdata") - def unexpected_read(_source_url): - raise AssertionError("A refused access class must not read the artifact") +# Finding 7: a quoted string year key is the same vintage as the integer. - monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) - with pytest.raises(ManifestAccessError, match="fetch-artifact stores bytes"): - fetch_source_artifact( - str(source), - source_id="dwp", - package_id="dwp-frs-2023-24", - year=2023, - output_dir=tmp_path / "pkg", - access="licensed", +def _quoted_year_release(tmp_path: Path, **entry_overrides: object) -> Path: + output_dir = tmp_path / "pkg" + output_dir.mkdir() + entry = _attested_entry(**entry_overrides) + (output_dir / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "microdata_release", + "files": {"2023": [entry]}, + }, + sort_keys=False, ) + ) + return output_dir -def test_fetch_refuses_to_pull_bytes_over_a_hash_only_registration( - tmp_path, monkeypatch -): - output_dir = tmp_path / "pkg" - _register(output_dir) - original_manifest = (output_dir / "manifest.yaml").read_bytes() - source = tmp_path / "adult.tab" - source.write_bytes(b"licensed microdata") +def test_register_sees_a_registration_under_a_quoted_year_key(tmp_path): + output_dir = _quoted_year_release(tmp_path) + original = (output_dir / "manifest.yaml").read_bytes() - def unexpected_read(_source_url): - raise AssertionError("A refused fetch must not read the artifact") + with pytest.raises(HashOnlyRegistrationError, match="already registers"): + _register(output_dir, sha256=OTHER_SHA) + assert (output_dir / "manifest.yaml").read_bytes() == original - monkeypatch.setattr("chronicle.artifacts._read_artifact", unexpected_read) + report = _register(output_dir) + manifest = _manifest(output_dir) - with pytest.raises(ManifestAccessError, match="Its bytes must not enter"): - fetch_source_artifact( - str(source), - source_id="dwp", - package_id="dwp-frs-2023-24", - year=2023, - output_dir=output_dir, - filename="adult.tab", - access="public", - ) + assert report.replaced is True + # The manifest's own key spelling is retained; no parallel key appears. + assert list(manifest["files"]) == ["2023"] - assert not (output_dir / "adult.tab").exists() - assert (output_dir / "manifest.yaml").read_bytes() == original_manifest +def test_register_refuses_a_vintage_recorded_under_both_key_spellings(tmp_path): + output_dir = _quoted_year_release(tmp_path) + manifest = _manifest(output_dir) + manifest["files"][2023] = [_attested_entry(filename="child.tab")] + (output_dir / "manifest.yaml").write_text(yaml.safe_dump(manifest, sort_keys=False)) + original = (output_dir / "manifest.yaml").read_bytes() + + with pytest.raises(HashOnlyRegistrationError, match="duplicate_vintage_key"): + _register(output_dir, filename="job.tab", sha256=OTHER_SHA) + + assert (output_dir / "manifest.yaml").read_bytes() == original -def test_fetch_writes_the_access_class_explicitly(tmp_path): - source = tmp_path / "table.xlsx" - source.write_bytes(b"publisher table") - output_dir = tmp_path / "pkg" - fetch_source_artifact( - str(source), - source_id="irs_soi", - package_id="soi-table-1-2", +# Finding 3: a public identity with an object in R2 is not reclassified. + + +def _archived_release(tmp_path: Path, monkeypatch, *, filename="asecpub23csv.zip"): + """A public release whose bytes were archived, then cleaned up locally.""" + output_dir = tmp_path / "pkg" + staging = tmp_path / "staging" + uploads = _record_uploads(monkeypatch) + _serve(monkeypatch, PUBLIC_BYTES) + _fetch_release( + output_dir, + staging_dir=staging, + filename=filename, + source_id="census_cps", + package_id="census-cps-asec-2023", year=2023, - output_dir=output_dir, + upload_r2=True, ) + assert uploads + staged = microdata_staging_path( + staging_dir=staging, + source_id="census_cps", + package_id="census-cps-asec-2023", + year=2023, + sha256=PUBLIC_SHA, + filename=filename, + ) + staged.unlink() + return output_dir, uploads - manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) - assert manifest["files"][2023]["access"] == "public" +@pytest.mark.parametrize("access", ["licensed", "restricted"]) +def test_register_refuses_to_reclassify_an_identity_r2_still_holds( + tmp_path, monkeypatch, access +): + output_dir, uploads = _archived_release(tmp_path, monkeypatch) + original = (output_dir / "manifest.yaml").read_bytes() + + with pytest.raises(HashOnlyRegistrationError, match="records the R2 object"): + _register( + output_dir, + source_id="census_cps", + package_id="census-cps-asec-2023", + filename="asecpub23csv.zip", + sha256=PUBLIC_SHA, + access=access, + licence="Some licence", + vintage="2023", + ) + assert (output_dir / "manifest.yaml").read_bytes() == original + entry = _manifest(output_dir)["files"][2023][0] + assert entry["storage"]["r2"]["uri"] == uploads[0][0] + inventory = inventory_source_artifacts(output_dir, staging_dir=tmp_path / "staging") + assert inventory.counts["r2_link_count"] == 1 -def test_fetch_into_a_microdata_release_manifest_requires_a_licence( + +def test_register_refuses_to_reclassify_an_unarchived_public_entry(tmp_path): + output_dir = tmp_path / "pkg" + output_dir.mkdir() + (output_dir / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "census_cps", + "package_id": "census-cps-asec-2023", + "kind": "microdata_release", + "files": {2023: [_public_release_entry(filename="asecpub23csv.zip")]}, + }, + sort_keys=False, + ) + ) + original = (output_dir / "manifest.yaml").read_bytes() + + with pytest.raises(HashOnlyRegistrationError, match="explicit decision"): + _register( + output_dir, + source_id="census_cps", + package_id="census-cps-asec-2023", + filename="ASECPUB23CSV.ZIP", + sha256=PUBLIC_SHA, + licence="Some licence", + vintage="2023", + ) + + assert (output_dir / "manifest.yaml").read_bytes() == original + + +def test_register_refuses_a_collision_with_a_public_alias(tmp_path): + # Finding 1 from the registration side: a manifest never holds one path + # under two access classes, whatever spelling the other entry uses. + output_dir = tmp_path / "pkg" + output_dir.mkdir() + (output_dir / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "microdata_release", + "files": {2022: [_public_release_entry(filename="Adult.tab")]}, + }, + sort_keys=False, + ) + ) + + with pytest.raises(HashOnlyRegistrationError, match="explicit decision"): + _register(output_dir) + + +# -------------------------------------------------------------------------- +# fetch-artifact +# -------------------------------------------------------------------------- + + +def test_fetch_refuses_a_hash_only_access_class(tmp_path, monkeypatch): + source = tmp_path / "adult.tab" + source.write_bytes(b"licensed microdata") + _refuse_read(monkeypatch) + + with pytest.raises(ManifestAccessError, match="fetch-artifact stores bytes"): + fetch_source_artifact( + str(source), + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=tmp_path / "pkg", + access="licensed", + ) + + +def test_fetch_refuses_to_pull_bytes_over_a_hash_only_registration( tmp_path, monkeypatch ): output_dir = tmp_path / "pkg" _register(output_dir) - source = tmp_path / "codebook.pdf" - source.write_bytes(b"public codebook") + original_manifest = (output_dir / "manifest.yaml").read_bytes() + source = tmp_path / "adult.tab" + source.write_bytes(b"licensed microdata") + _refuse_read(monkeypatch) - monkeypatch.setattr( - "chronicle.artifacts._read_artifact", - lambda _url: (_ for _ in ()).throw( - AssertionError("A refused fetch must not read the artifact") + with pytest.raises(ManifestAccessError, match="Its bytes must not enter"): + fetch_source_artifact( + str(source), + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=output_dir, + filename="adult.tab", + access="public", + ) + + assert not (output_dir / "adult.tab").exists() + assert (output_dir / "manifest.yaml").read_bytes() == original_manifest + + +# Finding 5: the inferred filename is refused before the read, not after. + + +@pytest.mark.parametrize("year", [2023, 2022], ids=["same-vintage", "other-vintage"]) +def test_fetch_without_filename_refuses_a_hash_only_registration_before_reading( + tmp_path, monkeypatch, year +): + output_dir = tmp_path / "pkg" + _register(output_dir) + original_manifest = (output_dir / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch, "the artifact was downloaded") + + with pytest.raises(ManifestAccessError, match="Its bytes must not enter"): + fetch_source_artifact( + "https://beta.ukdataservice.ac.uk/Umbraco/Surface/Download/9367/adult.tab", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=year, + output_dir=output_dir, + licence="UK Data Service End User Licence", + access="public", + ) + + assert reads == [] + assert (output_dir / "manifest.yaml").read_bytes() == original_manifest + assert sorted(path.name for path in output_dir.iterdir()) == ["manifest.yaml"] + + +def test_fetch_refuses_an_uninferrable_or_unsupported_source_before_reading( + tmp_path, monkeypatch +): + reads = _refuse_read(monkeypatch) + common = { + "source_id": "irs_soi", + "package_id": "soi-table-1-2", + "year": 2023, + "output_dir": tmp_path / "pkg", + } + with pytest.raises(ArtifactFilenameError, match="inferred from the URL"): + fetch_source_artifact("https://publisher.example/", **common) + with pytest.raises(ValueError, match="Unsupported source URL scheme"): + fetch_source_artifact("ftp://publisher.example/table.xlsx", **common) + assert reads == [] + assert not (tmp_path / "pkg").exists() + + +# Finding 1: no alias of a registered name slips past the byte boundary. + +ALIASES = { + "dot-slash": lambda out: "./adult.tab", + "dot-dot-segment": lambda out: "sub/../adult.tab", + "trailing-slash": lambda out: "adult.tab/", + "parent-then-back": lambda out: f"../{out.name}/adult.tab", + "absolute": lambda out: str(out / "adult.tab"), + "case": lambda out: "ADULT.TAB", +} + + +@pytest.mark.parametrize("alias", ALIASES.values(), ids=ALIASES.keys()) +def test_fetch_refuses_every_alias_of_a_hash_only_registration_before_reading( + tmp_path, monkeypatch, alias +): + output_dir = tmp_path / "pkg" + _register(output_dir) + original_manifest = (output_dir / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + _forbid_uploads(monkeypatch) + + with pytest.raises(ManifestAccessError): + fetch_source_artifact( + "https://publisher.example/frs/adult.tab", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=output_dir, + filename=alias(output_dir), + licence="Open Government Licence v3", + access="public", + upload_r2=True, + ) + + assert reads == [] + assert sorted(path.name for path in output_dir.iterdir()) == ["manifest.yaml"] + assert (output_dir / "manifest.yaml").read_bytes() == original_manifest + + +def test_fetch_refuses_a_path_that_escapes_the_package_directory(tmp_path, monkeypatch): + reads = _refuse_read(monkeypatch) + with pytest.raises(ArtifactFilenameError, match="bare filename"): + _fetch_table( + tmp_path / "table.xlsx", + tmp_path / "pkg", + filename="../../escaped.xlsx", + ) + assert reads == [] + assert not (tmp_path / "escaped.xlsx").exists() + + +def test_fetch_writes_the_access_class_and_kind_explicitly(tmp_path): + source = tmp_path / "table.xlsx" + source.write_bytes(b"publisher table") + output_dir = tmp_path / "pkg" + + _fetch_table(source, output_dir) + manifest = _manifest(output_dir) + + assert list(manifest)[:3] == ["source_id", "package_id", "kind"] + assert manifest["kind"] == "publisher_table" + assert manifest["files"][2023]["access"] == "public" + + +def test_fetch_into_a_microdata_release_manifest_requires_the_evidence( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + _register(output_dir) + original = (output_dir / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + base = { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "year": 2023, + "output_dir": output_dir, + "filename": "codebook.pdf", + "access": "public", + "licence": "OGL-UK-3.0", + "publisher": "Department for Work and Pensions", + "vintage": "2023_24", + "expected_sha256": FIXTURE_SHA, + "licence_evidence": {**EVIDENCE, "issuer": "DWP"}, + "staging_dir": tmp_path / "staging", + } + cases = [ + ({"licence": None}, "pass --licence"), + ({"vintage": None}, "pass --vintage"), + ({"publisher": None}, "pass --publisher"), + ({"expected_sha256": None}, "pass --expected-sha256"), + ( + {"licence": "UK Data Service End User Licence"}, + "not on Chronicle's allowlist", + ), + ({"licence_evidence": None}, "licence_evidence_missing_field:issuer"), + ( + {"licence_evidence": {**EVIDENCE, "url": "ask the archive"}}, + "url_not_durable", ), + ] + for overrides, expected in cases: + kwargs = {**base, **overrides} + with pytest.raises(ManifestAccessError, match=expected): + fetch_source_artifact("https://publisher.example/codebook.pdf", **kwargs) + + assert reads == [] + assert (output_dir / "manifest.yaml").read_bytes() == original + + +def test_fetch_refuses_a_registration_recorded_under_another_vintage( + tmp_path, monkeypatch +): + # The write target is a path in the package directory, not a year, so a + # registration under 2022 must still block a fetch requested for 2023. + package = tmp_path / "pkg" + _register(package, year=2022, vintage="2022_23") + reads = _refuse_read(monkeypatch) + + with pytest.raises(ManifestAccessError, match="for 2022 as access='licensed'"): + fetch_source_artifact( + "https://publisher.example/adult.tab", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=package, + filename="adult.tab", + licence="OGL-UK-3.0", + access="public", + ) + + assert reads == [] + assert not (package / "adult.tab").exists() + + +def test_fetch_refuses_a_list_entry_in_a_manifest_without_a_kind(tmp_path, monkeypatch): + # A missing kind must not make the guard blind to the entries the + # manifest actually holds: the byte boundary is refused first. + package = tmp_path / "pkg" + package.mkdir() + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "files": {2023: [_attested_entry()]}, + } + ) ) + reads = _refuse_read(monkeypatch) - with pytest.raises(ManifestAccessError, match="must record its publisher licence"): + with pytest.raises(ManifestAccessError, match="Its bytes must not enter"): fetch_source_artifact( - str(source), + "https://publisher.example/adult.tab", source_id="dwp", package_id="dwp-frs-2023-24", year=2023, - output_dir=output_dir, - filename="codebook.pdf", + output_dir=package, + filename="adult.tab", + access="public", + ) + + assert reads == [] + assert not (package / "adult.tab").exists() + + +# Finding 4: the existing manifest is validated strictly before any I/O. + + +def test_fetch_refuses_a_kind_that_conflicts_with_the_manifest(tmp_path, monkeypatch): + release = tmp_path / "release" + _register(release) + table = tmp_path / "table" + (tmp_path / "t.xlsx").write_bytes(b"table") + _fetch_table(tmp_path / "t.xlsx", table) + reads = _refuse_read(monkeypatch) + _forbid_uploads(monkeypatch) + before = {path: (path / "manifest.yaml").read_bytes() for path in (release, table)} + + with pytest.raises(ManifestAccessError, match="is a microdata_release manifest"): + fetch_source_artifact( + "https://publisher.example/codebook.pdf", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=release, + kind="publisher_table", + upload_r2=True, + ) + with pytest.raises(ManifestAccessError, match="is a publisher_table manifest"): + fetch_source_artifact( + "https://publisher.example/22in05ira.xlsx", + source_id="irs_soi", + package_id="soi-table-1-2", + year=2022, + output_dir=table, + kind="microdata_release", + licence="US-Government-Work", + upload_r2=True, + ) + + assert reads == [] + for path, original in before.items(): + assert (path / "manifest.yaml").read_bytes() == original + assert not (path / "codebook.pdf").exists() + + +def test_fetch_refuses_a_stored_unknown_kind_even_with_an_explicit_kind( + tmp_path, monkeypatch +): + package = tmp_path / "pkg" + package.mkdir() + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "microdata_rows", + "files": {2023: {"filename": "table.ods", "sha256": FIXTURE_SHA}}, + } + ) + ) + original = (package / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + + for kind in (None, "publisher_table"): + with pytest.raises(ManifestAccessError, match="unknown manifest kind"): + fetch_source_artifact( + "https://publisher.example/codebook.pdf", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=package, + kind=kind, + ) + + assert reads == [] + assert (package / "manifest.yaml").read_bytes() == original + + +def test_fetch_refuses_a_release_entry_without_an_access_class(tmp_path, monkeypatch): + package = tmp_path / "pkg" + package.mkdir() + entry = _attested_entry() + del entry["access"] + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "microdata_release", + "publisher": "DWP", + "files": {2023: [entry]}, + } + ) + ) + original = (package / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + _forbid_uploads(monkeypatch) + + with pytest.raises(ManifestAccessError, match="missing_access"): + fetch_source_artifact( + "https://publisher.example/adult.tab", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=package, + filename="adult.tab", + licence="OGL-UK-3.0", + vintage="2023_24", + expected_sha256=FIXTURE_SHA, + licence_evidence={**EVIDENCE, "issuer": "DWP"}, + upload_r2=True, + ) + + assert reads == [] + assert (package / "manifest.yaml").read_bytes() == original + assert not (package / "adult.tab").exists() + + +def test_fetch_refuses_an_invalid_manifest_before_reading(tmp_path, monkeypatch): + package = tmp_path / "pkg" + package.mkdir() + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "microdata_release", + "files": { + 2023: [ + _attested_entry(filename="adult.tab"), + _public_release_entry(filename="./adult.tab"), + ] + }, + } + ) + ) + original = (package / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + + with pytest.raises(ManifestAccessError, match="not a valid microdata_release"): + _fetch_release( + package, + staging_dir=tmp_path / "staging", + filename="job.tab", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + ) + + assert reads == [] + assert (package / "manifest.yaml").read_bytes() == original + + +def test_fetch_never_drops_an_existing_registration(tmp_path, monkeypatch): + output_dir = tmp_path / "pkg" + _register(output_dir) + _serve(monkeypatch, b"a public codebook in the same package") + + _fetch_release( + output_dir, + staging_dir=tmp_path / "staging", + filename="codebook.pdf", + content=b"a public codebook in the same package", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + licence="OGL-UK-3.0", + publisher="Department for Work and Pensions", + vintage="2023_24", + licence_evidence={**EVIDENCE, "issuer": "DWP"}, + ) + + entries = _manifest(output_dir)["files"][2023] + + assert [entry["filename"] for entry in entries] == ["adult.tab", "codebook.pdf"] + assert entries[0]["access"] == "licensed" + assert entries[0]["sha256"] == FIXTURE_SHA + assert entries[0]["attestation_evidence"] == ATTESTED["attestation_evidence"] + + +# -------------------------------------------------------------------------- +# Public microdata releases: bytes only with evidence, staged outside the tree +# -------------------------------------------------------------------------- + + +def test_fetch_archives_a_public_release_from_a_staging_directory( + tmp_path, monkeypatch +): + output_dir = tmp_path / "db" / "data" / "census" / "acs_pums_2022_1yr" + staging = tmp_path / "staging" + uploads = _record_uploads(monkeypatch) + _serve(monkeypatch, PUBLIC_BYTES) + + report = _fetch_release(output_dir, staging_dir=staging, upload_r2=True) + person = b"public person pums" + _serve(monkeypatch, person) + _fetch_release( + output_dir, + staging_dir=staging, + filename="csv_pus.zip", + content=person, + upload_r2=True, + ) + manifest = _manifest(output_dir) + entries = manifest["files"][2022] + + staged = microdata_staging_path( + staging_dir=staging, + source_id="census_acs", + package_id="census-acs-pums-2022-1yr", + year=2022, + sha256=PUBLIC_SHA, + filename="csv_hus.zip", + ) + assert report.valid + assert Path(report.local_path) == staged + assert staged.read_bytes() == PUBLIC_BYTES + # No release bytes ever land beside the manifest. + assert sorted(path.name for path in output_dir.iterdir()) == ["manifest.yaml"] + assert uploads[0][1] == str(staged) + assert uploads[0][0].endswith(f"/2022/{PUBLIC_SHA}/csv_hus.zip") + assert manifest["kind"] == "microdata_release" + assert manifest["publisher"] == "U.S. Census Bureau" + assert [entry["filename"] for entry in entries] == ["csv_hus.zip", "csv_pus.zip"] + first = entries[0] + assert first["access"] == "public" + assert first["licence"] == "US-Government-Work" + assert first["licence_evidence"] == { + **EVIDENCE, + "licence": "US-Government-Work", + "sha256": PUBLIC_SHA, + } + assert first["vintage"] == "2022" + assert first["hash_source"] == "chronicle_fetch" + assert first["attested_by"] == "chronicle" + assert first["verified_at"] == first["fetched_at"][:10] + assert first["storage"]["r2"]["uri"] == uploads[0][0] + + inventory = inventory_source_artifacts(output_dir, staging_dir=staging) + assert inventory.valid + assert inventory.counts["hash_only_count"] == 0 + assert inventory.counts["r2_link_count"] == 2 + assert all(entry.exists for entry in inventory.entries) + + +def test_a_public_release_without_a_recorded_object_is_incomplete( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + staging = tmp_path / "staging" + _serve(monkeypatch, PUBLIC_BYTES) + _fetch_release(output_dir, staging_dir=staging) + + inventory = inventory_source_artifacts(output_dir, staging_dir=staging) + + assert not inventory.valid + assert inventory.entries[0].errors == ("r2_object_not_recorded",) + + +# Finding 8: the fetch refuses bytes the reviewed pin does not cover. + + +def test_fetch_refuses_a_reissue_before_any_side_effect(tmp_path, monkeypatch): + output_dir = tmp_path / "pkg" + staging = tmp_path / "staging" + _serve(monkeypatch, b"silently re-published bytes") + _forbid_uploads(monkeypatch) + + with pytest.raises(ExpectedArtifactIdentityError) as raised: + _fetch_release( + output_dir, + staging_dir=staging, + expected_sha256=PUBLIC_SHA, + upload_r2=True, + ) + + message = str(raised.value) + assert PUBLIC_SHA in message + assert hashlib.sha256(b"silently re-published bytes").hexdigest() in message + assert "unreviewed reissue" in message + assert not output_dir.exists() + assert not staging.exists() + + +def test_fetch_refuses_a_size_that_disagrees_with_the_pin(tmp_path, monkeypatch): + _serve(monkeypatch, PUBLIC_BYTES) + with pytest.raises(ExpectedArtifactIdentityError, match="size_bytes=1"): + _fetch_release( + tmp_path / "pkg", + staging_dir=tmp_path / "staging", + expected_size_bytes=1, + ) + assert not (tmp_path / "pkg").exists() + + +def test_record_revision_does_not_override_the_expected_identity(tmp_path, monkeypatch): + output_dir = tmp_path / "pkg" + staging = tmp_path / "staging" + _serve(monkeypatch, PUBLIC_BYTES) + _fetch_release(output_dir, staging_dir=staging) + original = (output_dir / "manifest.yaml").read_bytes() + _serve(monkeypatch, b"revised bytes") + + with pytest.raises(ExpectedArtifactIdentityError): + _fetch_release( + output_dir, + staging_dir=staging, + expected_sha256=PUBLIC_SHA, + record_revision=True, + ) + + assert (output_dir / "manifest.yaml").read_bytes() == original + + +def test_an_expectation_that_contradicts_the_record_is_refused_before_reading( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + staging = tmp_path / "staging" + _serve(monkeypatch, PUBLIC_BYTES) + _fetch_release(output_dir, staging_dir=staging) + reads = _refuse_read(monkeypatch) + + with pytest.raises(ExpectedArtifactIdentityError, match="already records"): + _fetch_release( + output_dir, + staging_dir=staging, + expected_sha256=OTHER_SHA, + licence_evidence=EVIDENCE, + ) + + assert reads == [] + + +@pytest.mark.parametrize("bad", ["", "abc", FIXTURE_SHA.upper()]) +def test_fetch_never_accepts_an_invented_expected_hash(tmp_path, monkeypatch, bad): + reads = _refuse_read(monkeypatch) + with pytest.raises(ExpectedArtifactIdentityError, match="Never invent a hash"): + _fetch_table(tmp_path / "t.xlsx", tmp_path / "pkg", expected_sha256=bad) + assert reads == [] + + +def test_a_table_fetch_honours_an_expected_hash_before_writing(tmp_path, monkeypatch): + source = tmp_path / "table.xlsx" + source.write_bytes(b"publisher table") + output_dir = tmp_path / "pkg" + + with pytest.raises(ExpectedArtifactIdentityError): + _fetch_table(source, output_dir, expected_sha256=OTHER_SHA) + assert not output_dir.exists() + + report = _fetch_table( + source, + output_dir, + expected_sha256=hashlib.sha256(b"publisher table").hexdigest(), + expected_size_bytes=len(b"publisher table"), + ) + assert report.valid + + +def test_fetch_refuses_release_bytes_already_tracked_beside_the_manifest( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + output_dir.mkdir() + (output_dir / "csv_hus.zip").write_bytes(PUBLIC_BYTES) + reads = _refuse_read(monkeypatch) + + with pytest.raises(ManifestAccessError, match="staged outside the package tree"): + _fetch_release(output_dir, staging_dir=tmp_path / "staging") + + assert reads == [] + + +# Finding 6: a release vintage is a list, and every entry keeps its identity. + + +def test_refetching_a_release_file_with_different_bytes_is_refused( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + staging = tmp_path / "staging" + uploads = _record_uploads(monkeypatch) + _serve(monkeypatch, PUBLIC_BYTES) + _fetch_release(output_dir, staging_dir=staging, upload_r2=True) + person = b"public person pums" + _serve(monkeypatch, person) + _fetch_release( + output_dir, + staging_dir=staging, + filename="csv_pus.zip", + content=person, + upload_r2=True, + ) + original = (output_dir / "manifest.yaml").read_bytes() + revised = b"household pums, silently re-published" + reads = _refuse_read(monkeypatch) + uploads_before = list(uploads) + + # The list entry's recorded identity is live: a reviewed pin for other + # bytes contradicts it, and the contradiction is refused before the read. + with pytest.raises( + ExpectedArtifactIdentityError, match="already records" + ) as raised: + _fetch_release(output_dir, staging_dir=staging, content=revised, upload_r2=True) + + assert PUBLIC_SHA in str(raised.value) + assert "csv_hus.zip" in str(raised.value) + assert reads == [] + assert (output_dir / "manifest.yaml").read_bytes() == original + assert uploads == uploads_before + + # Served bytes that differ from the pin the record and the evidence agree + # on are refused after the read, before any write or upload. + _serve(monkeypatch, revised) + with pytest.raises(ExpectedArtifactIdentityError, match="unreviewed reissue"): + _fetch_release( + output_dir, + staging_dir=staging, + content=revised, + upload_r2=True, + expected_sha256=PUBLIC_SHA, + ) + assert (output_dir / "manifest.yaml").read_bytes() == original + assert uploads == uploads_before + assert not microdata_staging_path( + staging_dir=staging, + source_id="census_acs", + package_id="census-acs-pums-2022-1yr", + year=2022, + sha256=hashlib.sha256(revised).hexdigest(), + filename="csv_hus.zip", + ).exists() + + +def test_refetching_identical_release_bytes_preserves_the_entry_in_place( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + staging = tmp_path / "staging" + _record_uploads(monkeypatch) + _serve(monkeypatch, PUBLIC_BYTES) + _fetch_release(output_dir, staging_dir=staging, upload_r2=True) + person = b"public person pums" + _serve(monkeypatch, person) + _fetch_release( + output_dir, + staging_dir=staging, + filename="csv_pus.zip", + content=person, + upload_r2=True, + ) + before = _manifest(output_dir)["files"][2022] + + monkeypatch.setenv("CHRONICLE_R2_RAW_BUCKET", "chronicle-raw") + _serve(monkeypatch, PUBLIC_BYTES) + report = _fetch_release(output_dir, staging_dir=staging, upload_r2=True) + after = _manifest(output_dir)["files"][2022] + + assert report.valid + assert [entry["filename"] for entry in after] == ["csv_hus.zip", "csv_pus.zip"] + # The recorded object is history: the backfill copy does not restate it. + assert after[0]["storage"] == before[0]["storage"] + assert after[0]["storage"]["r2"]["bucket"] == "ledger-raw" + assert after[1] == before[1] + + +def test_record_revision_of_a_release_file_supersedes_only_that_entry( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + staging = tmp_path / "staging" + _record_uploads(monkeypatch) + _serve(monkeypatch, PUBLIC_BYTES) + _fetch_release(output_dir, staging_dir=staging, upload_r2=True) + person = b"public person pums" + _serve(monkeypatch, person) + _fetch_release( + output_dir, + staging_dir=staging, + filename="csv_pus.zip", + content=person, + upload_r2=True, + ) + before = _manifest(output_dir)["files"][2022] + revised = b"household pums, revised" + _serve(monkeypatch, revised) + + _fetch_release( + output_dir, + staging_dir=staging, + content=revised, + upload_r2=True, + record_revision=True, + ) + after = _manifest(output_dir)["files"][2022] + + assert [entry["filename"] for entry in after] == ["csv_hus.zip", "csv_pus.zip"] + assert after[0]["sha256"] == hashlib.sha256(revised).hexdigest() + assert after[0]["licence_evidence"]["sha256"] == after[0]["sha256"] + assert [entry["uri"] for entry in after[0]["storage"]["previous_r2"]] == [ + before[0]["storage"]["r2"]["uri"] + ] + assert after[1] == before[1] + + +def test_a_second_file_never_turns_a_publisher_table_vintage_into_a_list( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + (tmp_path / "22in05ira.xlsx").write_bytes(b"IRA table 5") + _fetch_table(tmp_path / "22in05ira.xlsx", output_dir, year=2022) + original = (output_dir / "manifest.yaml").read_bytes() + (tmp_path / "22in05ira_rev.xlsx").write_bytes(b"IRA table 5, renamed") + + with pytest.raises(SourceArtifactRevisionError): + _fetch_table(tmp_path / "22in05ira_rev.xlsx", output_dir, year=2022) + assert (output_dir / "manifest.yaml").read_bytes() == original + + _fetch_table( + tmp_path / "22in05ira_rev.xlsx", output_dir, year=2022, record_revision=True + ) + revised = _manifest(output_dir)["files"][2022] + + assert isinstance(revised, dict) + assert revised["filename"] == "22in05ira_rev.xlsx" + + +def test_a_second_file_over_an_unidentified_table_entry_is_refused( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + output_dir.mkdir() + (output_dir / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "irs_soi", + "package_id": "soi-table-1-2", + "kind": "publisher_table", + "files": {2022: {"filename": "22in05ira.xlsx", "source_url": "x"}}, + } ) + ) + original = (output_dir / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + + with pytest.raises(MalformedManifestError, match="one file per vintage"): + _fetch_table(tmp_path / "other.xlsx", output_dir, year=2022) + + assert reads == [] + assert (output_dir / "manifest.yaml").read_bytes() == original + + +# Finding 7 on the fetch side. + + +def test_fetch_refuses_a_revision_recorded_under_a_quoted_year_key( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + (tmp_path / "t.xlsx").write_bytes(b"first") + _fetch_table(tmp_path / "t.xlsx", output_dir, year=2022) + manifest = _manifest(output_dir) + manifest["files"] = {"2022": manifest["files"][2022]} + (output_dir / "manifest.yaml").write_text(yaml.safe_dump(manifest, sort_keys=False)) + original = (output_dir / "manifest.yaml").read_bytes() + (tmp_path / "t.xlsx").write_bytes(b"second") + + with pytest.raises(SourceArtifactRevisionError, match="entry '2022'"): + _fetch_table(tmp_path / "t.xlsx", output_dir, year=2022) + assert (output_dir / "manifest.yaml").read_bytes() == original + + (tmp_path / "t.xlsx").write_bytes(b"first") + _fetch_table(tmp_path / "t.xlsx", output_dir, year=2022) + assert list(_manifest(output_dir)["files"]) == ["2022"] + + +def test_fetch_refuses_a_vintage_recorded_under_both_key_spellings( + tmp_path, monkeypatch +): + output_dir = tmp_path / "pkg" + (tmp_path / "t.xlsx").write_bytes(b"first") + _fetch_table(tmp_path / "t.xlsx", output_dir, year=2022) + manifest = _manifest(output_dir) + manifest["files"]["2022"] = dict(manifest["files"][2022]) + (output_dir / "manifest.yaml").write_text(yaml.safe_dump(manifest, sort_keys=False)) + original = (output_dir / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + + with pytest.raises(ManifestAccessError, match="duplicate_vintage_key"): + _fetch_table(tmp_path / "t.xlsx", output_dir, year=2022) + + assert reads == [] + assert (output_dir / "manifest.yaml").read_bytes() == original # -------------------------------------------------------------------------- @@ -475,11 +1850,7 @@ def test_publish_raw_refuses_hash_only_entries_without_reading_bytes( tmp_path, monkeypatch ): root = _hash_only_tree(tmp_path) - - def unexpected_upload(*args, **kwargs): - raise AssertionError("A hash-only registration must never be uploaded") - - monkeypatch.setattr("chronicle.artifacts._upload_r2_object", unexpected_upload) + _forbid_uploads(monkeypatch) report = publish_source_artifacts(root) @@ -497,12 +1868,7 @@ def test_publish_raw_skip_hash_only_reports_the_skip_without_failing( tmp_path, monkeypatch ): root = _hash_only_tree(tmp_path) - monkeypatch.setattr( - "chronicle.artifacts._upload_r2_object", - lambda *args, **kwargs: (_ for _ in ()).throw( - AssertionError("A hash-only registration must never be uploaded") - ), - ) + _forbid_uploads(monkeypatch) report = publish_source_artifacts(root, skip_hash_only=True) @@ -517,16 +1883,77 @@ def test_publish_raw_leaves_the_manifest_untouched(tmp_path, monkeypatch): root = _hash_only_tree(tmp_path) manifest_path = root / "dwp" / "frs_2023_24" / "manifest.yaml" original = manifest_path.read_bytes() - monkeypatch.setattr( - "chronicle.artifacts._upload_r2_object", - lambda *args, **kwargs: (_ for _ in ()).throw(AssertionError("no upload")), - ) + _forbid_uploads(monkeypatch) publish_source_artifacts(root, skip_hash_only=True) assert manifest_path.read_bytes() == original +def test_publish_raw_reports_a_violation_even_when_skipping(tmp_path, monkeypatch): + root = _hash_only_tree(tmp_path) + (root / "dwp" / "frs_2023_24" / "adult.tab").write_bytes(b"leaked bytes") + _forbid_uploads(monkeypatch) + + report = publish_source_artifacts(root, skip_hash_only=True) + + # --skip-hash-only turns off the refusal, not the contract check. + assert not report.valid + assert "bytes_present_for_hash_only_entry" in report.entries[0].errors + + +@pytest.mark.parametrize("skip_hash_only", [False, True]) +def test_publish_raw_never_uploads_a_hash_only_file_through_a_public_alias( + tmp_path, monkeypatch, skip_hash_only +): + # Finding 1: a hand-edited public alias of the licensed entry must not + # carry its bytes to the bucket; the manifest is refused whole. + root = _hash_only_tree(tmp_path) + package = root / "dwp" / "frs_2023_24" + (package / "adult.tab").write_bytes(b"leaked licensed bytes") + manifest = _manifest(package) + manifest["files"][2023].append( + { + **_public_release_entry(filename="./adult.tab"), + "sha256": hashlib.sha256(b"leaked licensed bytes").hexdigest(), + } + ) + (package / "manifest.yaml").write_text(yaml.safe_dump(manifest, sort_keys=False)) + original = (package / "manifest.yaml").read_bytes() + _forbid_uploads(monkeypatch) + + report = publish_source_artifacts(root, skip_hash_only=skip_hash_only) + + assert not report.valid + assert report.entries == () + assert any("filename_collision:adult.tab" in error for error in report.errors) + assert any("non_canonical_filename:./adult.tab" in error for error in report.errors) + assert (package / "manifest.yaml").read_bytes() == original + + +def test_publish_raw_uploads_a_staged_release_and_never_the_tree(tmp_path, monkeypatch): + output_dir = tmp_path / "data" / "census" / "acs_pums_2022_1yr" + staging = tmp_path / "staging" + _serve(monkeypatch, PUBLIC_BYTES) + _fetch_release(output_dir, staging_dir=staging) + uploads = _record_uploads(monkeypatch) + + report = publish_source_artifacts(tmp_path / "data", staging_dir=staging) + + assert report.valid + assert report.counts["uploaded_count"] == 1 + assert uploads[0][1].startswith(str(staging)) + assert ( + _manifest(output_dir)["files"][2022][0]["storage"]["r2"]["uri"] + == (uploads[0][0]) + ) + + (output_dir / "csv_hus.zip").write_bytes(PUBLIC_BYTES) + report = publish_source_artifacts(tmp_path / "data", staging_dir=staging) + assert not report.valid + assert "bytes_present_for_microdata_release_entry" in report.entries[0].errors + + # -------------------------------------------------------------------------- # inventory-artifacts # -------------------------------------------------------------------------- @@ -570,7 +1997,33 @@ def test_inventory_rejects_a_list_entry_outside_a_microdata_release(tmp_path): { "source_id": "dwp", "package_id": "dwp-tables", - "files": {2023: [{"filename": "a.ods", "sha256": FIXTURE_SHA}]}, + "kind": "publisher_table", + "files": {2023: [{"filename": "a.ods", "sha256": FIXTURE_SHA}]}, + } + ) + ) + + report = inventory_source_artifacts(tmp_path / "data") + + assert not report.valid + assert report.entries[0].errors == ( + "list_file_spec_requires_microdata_release_kind", + ) + + +def test_inventory_reports_manifest_level_defects(tmp_path): + package = tmp_path / "data" / "dwp" / "tables" + package.mkdir(parents=True) + (package / "manifest.yaml").write_text( + yaml.safe_dump( + { + "source_id": "dwp", + "package_id": "dwp-tables", + "kind": "publisher_table", + "files": { + 2023: {"filename": "../a.ods", "sha256": FIXTURE_SHA}, + "2023": {"filename": "b.ods", "sha256": FIXTURE_SHA}, + }, } ) ) @@ -578,13 +2031,20 @@ def test_inventory_rejects_a_list_entry_outside_a_microdata_release(tmp_path): report = inventory_source_artifacts(tmp_path / "data") assert not report.valid - assert report.entries[0].errors == ( - "list_file_spec_requires_microdata_release_kind", + assert any( + error.startswith("duplicate_vintage_key:2023") for error in report.errors ) + assert any( + error.startswith("non_canonical_filename:../a.ods") for error in report.errors + ) + # A non-bare name is never resolved to a path outside the package. + first = next(entry for entry in report.entries if entry.filename == "../a.ods") + assert first.local_path == str(package) + assert "missing_file" not in first.errors # -------------------------------------------------------------------------- -# Source packages never parse a microdata release +# Source packages never parse a microdata release or a hash-only entry # -------------------------------------------------------------------------- @@ -608,13 +2068,16 @@ def test_source_artifact_spec_refuses_to_parse_a_registered_release(): spec._artifact_content(2023) -def test_year_mapping_refuses_a_multi_file_vintage(): +def test_year_mapping_refuses_a_multi_file_vintage_and_both_key_spellings(): from chronicle.source_package import _year_mapping with pytest.raises(ValueError, match="list of 2 entries"): _year_mapping( {2023: [{"filename": "adult.tab"}, {"filename": "child.tab"}]}, 2023 ) + with pytest.raises(ValueError, match="both keys"): + _year_mapping({2023: {"filename": "a"}, "2023": {"filename": "b"}}, 2023) + assert _year_mapping({"2023": {"filename": "a"}}, 2023) == {"filename": "a"} def test_validate_package_reports_a_microdata_release_carve_out(tmp_path): @@ -650,6 +2113,200 @@ def test_validate_package_reports_a_microdata_release_carve_out(tmp_path): assert "microdata_release_not_parseable" in {issue.code for issue in report.errors} +# Finding 2: the byte reader refuses a hash-only entry under any manifest kind. + +LICENSED_BYTES = b"sernum\tage\n1\t45\n2\t31\n" + + +class _UnreadableArtifactPath: + """Bytes sit in the package tree; a refusal must not read them.""" + + def read_bytes(self) -> bytes: + raise AssertionError("a hash-only entry's bytes must not be read") + + +def _isolated_reader(tmp_path: Path, monkeypatch) -> Path: + cache_root = tmp_path / "cache" + monkeypatch.setenv(SOURCE_ARTIFACT_CACHE_ENV, str(cache_root)) + monkeypatch.setattr( + "chronicle.source_package._fetch_source_artifact_content", + lambda _url: (_ for _ in ()).throw( + AssertionError("a hash-only entry must never be fetched") + ), + ) + return cache_root + + +def _licensed_table_spec( + tmp_path: Path, monkeypatch, *, kind: str | None, access: str +) -> tuple[SourceArtifactSpec, Path]: + """A publisher-table package whose single entry is hash-only. + + The resource package is a namespace package on ``sys.path`` with a unique + name, so ``importlib.resources.files`` resolves it exactly as it resolves + the repository's ``db`` package. + """ + package_name = f"chronicle_test_{uuid.uuid4().hex}" + resource_dir = tmp_path / "pkgroot" / package_name / "data" / "dwp" / "frs" + resource_dir.mkdir(parents=True) + manifest: dict = {"source_id": "dwp", "package_id": "dwp-frs"} + if kind is not None: + manifest["kind"] = kind + entry = _attested_entry( + access=access, + sha256=hashlib.sha256(LICENSED_BYTES).hexdigest(), + source_url="https://ukdataservice.example/adult.tab", + ) + manifest["files"] = {2023: entry} + (resource_dir / "manifest.yaml").write_text( + yaml.safe_dump(manifest, sort_keys=False) + ) + (resource_dir / "adult.tab").write_bytes(LICENSED_BYTES) + monkeypatch.syspath_prepend(str(tmp_path / "pkgroot")) + monkeypatch.delitem(sys.modules, package_name, raising=False) + spec = SourceArtifactSpec( + source_name="dwp", + source_table="Family Resources Survey", + resource_package=package_name, + resource_directory="data/dwp/frs", + manifest="manifest.yaml", + vintage="2023_24", + extracted_at="2026-09-02", + extraction_method="none", + parser="delimited_text_full_rows", + delimiter="\t", + artifact_year=2023, + ) + return spec, resource_dir + + +@pytest.mark.parametrize("access", ["licensed", "restricted"]) +def test_byte_reader_refuses_a_hash_only_entry_before_any_store( + tmp_path, monkeypatch, access +): + cache_root = _isolated_reader(tmp_path, monkeypatch) + entry = _attested_entry(access=access, source_url="https://x.example/adult.tab") + cached = cache_root / FIXTURE_SHA / "adult.tab" + cached.parent.mkdir(parents=True) + cached.write_bytes(LICENSED_BYTES) + monkeypatch.setenv(SOURCE_ARTIFACT_FETCH_ENV, "1") + + with pytest.raises(ManifestAccessError, match=f"access={access!r}"): + _read_source_artifact_content(_UnreadableArtifactPath(), entry) + + +def test_byte_reader_treats_an_unknown_access_class_as_unreadable( + tmp_path, monkeypatch +): + _isolated_reader(tmp_path, monkeypatch) + entry = _attested_entry(access="internal") + with pytest.raises(ManifestAccessError, match="Unknown access class 'internal'"): + _read_source_artifact_content(_UnreadableArtifactPath(), entry) + + +@pytest.mark.parametrize("kind", ["publisher_table"]) +@pytest.mark.parametrize("access", ["licensed", "restricted"]) +def test_artifact_content_refuses_a_hash_only_mapping_entry( + tmp_path, monkeypatch, kind, access +): + _isolated_reader(tmp_path, monkeypatch) + spec, _resource_dir = _licensed_table_spec( + tmp_path, monkeypatch, kind=kind, access=access + ) + + with pytest.raises(ManifestAccessError, match="identity only"): + spec.assert_parseable(2023) + with pytest.raises(ManifestAccessError): + spec._artifact_content(2023) + with pytest.raises(ManifestAccessError): + spec.build_source_rows(2023) + + +def test_validate_package_reports_a_hash_only_entry_without_reading( + tmp_path, monkeypatch +): + _isolated_reader(tmp_path, monkeypatch) + spec, _resource_dir = _licensed_table_spec( + tmp_path, monkeypatch, kind="publisher_table", access="licensed" + ) + package_dir = tmp_path / "package" + package_dir.mkdir() + (package_dir / "source_package.yaml").write_text( + yaml.safe_dump( + { + "schema_version": "ledger.source_package.v1", + "package_id": "dwp-frs-parse-attempt", + "label": "Attempt to parse a hash-only table entry", + "artifact": { + "source_name": spec.source_name, + "source_table": spec.source_table, + "resource_package": spec.resource_package, + "resource_directory": spec.resource_directory, + "manifest": "manifest.yaml", + "vintage": "2023_24", + "extracted_at": "2026-09-02", + "extraction_method": "none", + "parser": "delimited_text_full_rows", + "delimiter": "\t", + "artifact_year": 2023, + }, + "record_sets": [], + }, + sort_keys=False, + ) + ) + + report = validate_source_package(package_dir, year=2023) + + assert not report.valid + assert [issue.code for issue in report.errors] == [ + "hash_only_artifact_not_parseable" + ] + + +def test_build_suite_refuses_a_hash_only_entry_before_touching_the_output( + tmp_path, monkeypatch +): + from chronicle.suite import build_source_suite + + _isolated_reader(tmp_path, monkeypatch) + spec, _resource_dir = _licensed_table_spec( + tmp_path, monkeypatch, kind="publisher_table", access="licensed" + ) + package_dir = tmp_path / "package" + package_dir.mkdir() + (package_dir / "source_package.yaml").write_text( + yaml.safe_dump( + { + "schema_version": "ledger.source_package.v1", + "package_id": "dwp-frs-parse-attempt", + "label": "Attempt to build a hash-only table entry", + "artifact": { + "source_name": spec.source_name, + "source_table": spec.source_table, + "resource_package": spec.resource_package, + "resource_directory": spec.resource_directory, + "manifest": "manifest.yaml", + "vintage": "2023_24", + "extracted_at": "2026-09-02", + "extraction_method": "none", + "parser": "delimited_text_full_rows", + "delimiter": "\t", + "artifact_year": 2023, + }, + "record_sets": [], + }, + sort_keys=False, + ) + ) + output_dir = tmp_path / "suite" + + with pytest.raises(ManifestAccessError): + build_source_suite(package_dir, output_dir, year=2023) + + assert not output_dir.exists() + + # -------------------------------------------------------------------------- # CLI # -------------------------------------------------------------------------- @@ -683,8 +2340,16 @@ def test_cli_register_artifact_round_trips(tmp_path, capsys): "restricted", "--doi", "10.5255/UKDA-SN-9422-1", - "--verified-at", - "2026-09-02", + "--hash-source", + "consumer_pin", + "--attested-by", + "PolicyEngine/microcosm", + "--pinned-from-repository", + "PolicyEngine/microcosm", + "--pinned-from-path", + "packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json", + "--pinned-from-commit", + FIXTURE_COMMIT, ] ) payload = json.loads(capsys.readouterr().out) @@ -692,11 +2357,14 @@ def test_cli_register_artifact_round_trips(tmp_path, capsys): assert exit_code == 0 assert payload["valid"] is True assert payload["access"] == "restricted" + assert payload["hash_source"] == "consumer_pin" assert payload["r2_location"] is None assert payload["registration"] == ( f"hmrc/hmrc-spi-public-use-tape-2022-23/2022/{FIXTURE_SHA}/put2223uk.tab" ) assert not (output_dir / "put2223uk.tab").exists() + entry = _manifest(output_dir)["files"][2022][0] + assert entry["pinned_from"]["commit"] == FIXTURE_COMMIT def test_cli_register_artifact_rejects_public_access(tmp_path): @@ -722,349 +2390,342 @@ def test_cli_register_artifact_rejects_public_access(tmp_path): "Public domain", "--access", "public", + "--hash-source", + "consumer_pin", + "--attested-by", + "x", ] ) -# -------------------------------------------------------------------------- -# Regressions: the byte boundary must not be escapable -# -------------------------------------------------------------------------- - - -def _licensed_manifest(package: Path, payload: dict) -> Path: - package.mkdir(parents=True, exist_ok=True) - (package / "manifest.yaml").write_text(yaml.safe_dump(payload)) - return package - - -def test_fetch_refuses_a_registration_recorded_under_another_vintage(tmp_path): - # The write target is a path in the package directory, not a year, so a - # registration under 2022 must still block a fetch requested for 2023. - package = _licensed_manifest( - tmp_path / "pkg", - { - "source_id": "dwp", - "package_id": "dwp-frs-2023-24", - "kind": "microdata_release", - "files": { - 2022: [ - { - "filename": "adult.tab", - "access": "licensed", - "licence": "UK Data Service End User Licence", - "vintage": "2022_23", - "sha256": FIXTURE_SHA, - "doi": "10.5255/UKDA-SN-9367-2", - "verified_at": "2026-09-02", - } - ] - }, - }, - ) - source = tmp_path / "adult.tab" - source.write_bytes(b"licensed microdata") - - with pytest.raises(ManifestAccessError, match="for 2022 as access='licensed'"): - fetch_source_artifact( - str(source), - source_id="dwp", - package_id="dwp-frs-2023-24", - year=2023, - output_dir=package, - filename="adult.tab", - licence="UK Data Service End User Licence", - access="public", - ) - - assert not (package / "adult.tab").exists() - - -def test_fetch_refuses_a_list_entry_in_a_manifest_without_a_kind(tmp_path): - # A missing or misspelled kind must not make the guard blind to the - # entries the manifest actually holds. - package = _licensed_manifest( - tmp_path / "pkg", - { - "source_id": "dwp", - "package_id": "dwp-frs-2023-24", - "files": { - 2023: [ - { - "filename": "adult.tab", - "access": "licensed", - "licence": "UK Data Service End User Licence", - "sha256": FIXTURE_SHA, - } - ] - }, - }, - ) - source = tmp_path / "adult.tab" - source.write_bytes(b"licensed microdata") - - with pytest.raises(ManifestAccessError, match="Its bytes must not enter"): - fetch_source_artifact( - str(source), - source_id="dwp", - package_id="dwp-frs-2023-24", - year=2023, - output_dir=package, - filename="adult.tab", - access="public", - ) - - assert not (package / "adult.tab").exists() - - -def test_fetch_never_drops_an_existing_registration(tmp_path): +def test_cli_refusals_print_an_error_and_exit_1(tmp_path, monkeypatch, capsys): output_dir = tmp_path / "pkg" _register(output_dir) - source = tmp_path / "other.csv" - source.write_bytes(b"a public table in the same package") + reads = _refuse_read(monkeypatch) - fetch_source_artifact( - str(source), - source_id="dwp", - package_id="dwp-frs-2023-24", - year=2023, - output_dir=output_dir, - filename="other.csv", - licence="Open Government Licence", - access="public", + exit_code = harness_main( + [ + "fetch-artifact", + "--url", + "https://publisher.example/adult.tab", + "--source-id", + "dwp", + "--package-id", + "dwp-frs-2023-24", + "--year", + "2023", + "--out-dir", + str(output_dir), + ] ) + captured = capsys.readouterr() - manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) - entries = manifest["files"][2023] - - assert [entry["filename"] for entry in entries] == ["adult.tab", "other.csv"] - assert entries[0]["access"] == "licensed" - assert entries[0]["sha256"] == FIXTURE_SHA - + assert exit_code == 1 + assert captured.out == "" + assert "Its bytes must not enter" in captured.err + assert reads == [] -def test_publish_raw_reports_a_violation_even_when_skipping(tmp_path, monkeypatch): - root = _hash_only_tree(tmp_path) - (root / "dwp" / "frs_2023_24" / "adult.tab").write_bytes(b"leaked bytes") - monkeypatch.setattr( - "chronicle.artifacts._upload_r2_object", - lambda *args, **kwargs: (_ for _ in ()).throw(AssertionError("no upload")), + exit_code = harness_main( + [ + "register-artifact", + "--source-id", + "dwp", + "--package-id", + "dwp-frs-2023-24", + "--year", + "2023", + "--out-dir", + str(output_dir), + "--filename", + "adult.tab", + "--sha256", + OTHER_SHA, + "--vintage", + "2023_24", + "--licence", + "UK Data Service End User Licence", + "--access", + "licensed", + "--doi", + "10.5255/UKDA-SN-9367-2", + "--hash-source", + "consumer_pin", + "--attested-by", + "PolicyEngine/microcosm", + "--pinned-from-repository", + "PolicyEngine/microcosm", + "--pinned-from-path", + "p", + "--pinned-from-commit", + FIXTURE_COMMIT, + ] ) - - report = publish_source_artifacts(root, skip_hash_only=True) - - # --skip-hash-only turns off the refusal, not the contract check. - assert not report.valid - assert "bytes_present_for_hash_only_entry" in report.entries[0].errors - - -def test_reregistering_the_current_pin_stays_idempotent_after_a_reissue(tmp_path): - output_dir = tmp_path / "pkg" - _register(output_dir, sha256=FIXTURE_SHA) - _register(output_dir, sha256="b" * 64, allow_reissue=True) - - report = _register(output_dir, sha256="b" * 64) - - assert report.replaced is True - entries = yaml.safe_load((output_dir / "manifest.yaml").read_text())["files"][2023] - assert [entry["sha256"] for entry in entries] == [FIXTURE_SHA, "b" * 64] - - -# -------------------------------------------------------------------------- -# Public microdata releases -# -------------------------------------------------------------------------- + captured = capsys.readouterr() + assert exit_code == 1 + assert "pass --allow-reissue" in captured.err -def test_fetch_can_declare_a_public_microdata_release(tmp_path): +def test_cli_fetch_archives_a_release_with_the_reviewed_identity( + tmp_path, monkeypatch, capsys +): output_dir = tmp_path / "pkg" - household = tmp_path / "csv_hus.zip" - household.write_bytes(b"public household pums") - person = tmp_path / "csv_pus.zip" - person.write_bytes(b"public person pums") - - for source in (household, person): - fetch_source_artifact( - str(source), - source_id="census_acs", - package_id="census-acs-pums-2022-1yr", - year=2022, - output_dir=output_dir, - licence="U.S. Census Bureau public-use file", - access="public", - kind="microdata_release", - ) - - manifest = yaml.safe_load((output_dir / "manifest.yaml").read_text()) - entries = manifest["files"][2022] - - # A release may hold several files under one vintage, and both are kept. - assert manifest["kind"] == "microdata_release" - assert [entry["filename"] for entry in entries] == ["csv_hus.zip", "csv_pus.zip"] - assert all(entry["access"] == "public" for entry in entries) - assert all(entry["licence"] for entry in entries) - - report = inventory_source_artifacts(output_dir) - assert report.valid - assert report.counts["hash_only_count"] == 0 + staging = tmp_path / "staging" + uploads = _record_uploads(monkeypatch) + _serve(monkeypatch, PUBLIC_BYTES) + argv = [ + "fetch-artifact", + "--url", + "https://www2.census.gov/programs-surveys/cps/datasets/2023/march/asecpub23csv.zip", + "--source-id", + "census_cps", + "--package-id", + "census-cps-asec-2023", + "--year", + "2023", + "--out-dir", + str(output_dir), + "--publisher", + "U.S. Census Bureau", + "--vintage", + "2023 ASEC / 2022 income reference year", + "--access", + "public", + "--licence", + "US-Government-Work", + "--kind", + "microdata_release", + "--expected-sha256", + PUBLIC_SHA, + "--expected-size-bytes", + str(len(PUBLIC_BYTES)), + "--licence-evidence-issuer", + EVIDENCE["issuer"], + "--licence-evidence-scope", + EVIDENCE["scope"], + "--licence-evidence-url", + EVIDENCE["url"], + "--staging-dir", + str(staging), + "--upload-r2", + ] + assert harness_main(argv) == 0 + payload = json.loads(capsys.readouterr().out) + entry = _manifest(output_dir)["files"][2023][0] -def test_fetch_rejects_an_unknown_manifest_kind(tmp_path): - source = tmp_path / "table.xlsx" - source.write_bytes(b"publisher table") + assert payload["valid"] is True + assert payload["sha256"] == PUBLIC_SHA + assert entry["filename"] == "asecpub23csv.zip" + assert entry["vintage"] == "2023 ASEC / 2022 income reference year" + assert entry["licence_evidence"]["sha256"] == PUBLIC_SHA + assert uploads[0][0].endswith("/asecpub23csv.zip") - with pytest.raises(ManifestAccessError, match="Unknown manifest kind"): - fetch_source_artifact( - str(source), - source_id="irs_soi", - package_id="soi-table-1-2", - year=2023, - output_dir=tmp_path / "pkg", - kind="microdata_rows", - ) + _serve(monkeypatch, b"a reissue") + assert harness_main(argv) == 1 + captured = capsys.readouterr() + assert PUBLIC_SHA in captured.err + assert captured.out == "" # -------------------------------------------------------------------------- -# The generated fetch plan +# The committed registrations # -------------------------------------------------------------------------- - -def test_planned_fetch_commands_parse_against_the_real_cli(): - """Every command `plan` prints must be runnable as printed.""" - import argparse - import contextlib - import shlex - import subprocess - import sys - - result = subprocess.run( - [ - sys.executable, - str(REPO_ROOT / "scripts" / "register_microdata_releases.py"), - "--microcosm-root", - str(MICROCOSM_ROOT), - "plan", - ], - capture_output=True, - text=True, - cwd=REPO_ROOT, - ) - if result.returncode != 0: - pytest.skip(f"No readable microcosm checkout: {result.stderr.strip()[:120]}") - - commands = [ - line - for line in result.stdout.splitlines() - if line.startswith("uv run chronicle") +#: Every committed hash-only pin, keyed by registration identity. Values are +#: (sha256, size_bytes, vintage, access, licence); the checksums are the +#: reviewed consumer pins and any change here is a change of identity. +GOLDEN_PINS: dict[tuple[str, str, int, str], tuple[str, int, str, str, str]] = { + ("dwp", "dwp-frs-2023-24", 2023, "accounts.tab"): ( + "c5e31932bfd06087f835d2c83c0984c85a93409bf5ef85b699cb0958abcba1ea", + 1807921, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "adult.tab"): ( + "e09f9647d03585c81a528636028b2ed495f8f1fbcf64c5e7b4fe521b67367e06", + 35323384, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "benefits.tab"): ( + "ff30d054cc659bcf23b44c492d98cfd701c0bfdb63e8e9aa9769b490ba9d636b", + 4460292, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "benunit.tab"): ( + "88946815eace8561516d5cbb442c27e319c1e90abc381fb2338f0126e3b9e05b", + 21213867, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "child.tab"): ( + "b5dc84fe8b002ee925e61fae23fed27b11537af9fb174f1d07d9cc1748b9702e", + 2913156, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "chldcare.tab"): ( + "566e0ebca1d5e2f3e424e556c91f4cb583d17dadfdfa59feb3841eda7e5976a3", + 273837, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "extchild.tab"): ( + "8d358d7ee66ee4a7ceab87b4f24fbbf21ac86dc038dc7831e51fb271f96a57ec", + 18677, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "househol.tab"): ( + "5fd26b8b675f33b3b30c9ac789a18da17de734790f77e00ded287d1c3a187b30", + 12387117, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "job.tab"): ( + "88b77ffe06865f029f713bb1d55ff12bdea8a1234de5bc293e72458fe64f3a74", + 10934873, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "maint.tab"): ( + "f2dc924eb5a51b0c357791693d15b431327dc39c6421011efb313d88bf839695", + 15440, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "mortgage.tab"): ( + "ce36b477d67837c469608a0d68f7ef269ac04758974235f1157d2f6b92cdbfdc", + 631783, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "oddjob.tab"): ( + "b4ba3dd3151f73a01422983c60514a3e38458ddfa4fb33ae4ed0326873406305", + 5165, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "penprov.tab"): ( + "ee001461c40306ec24b38b2881e1774121114266a2ee449d606cd0a811c37731", + 522313, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("dwp", "dwp-frs-2023-24", 2023, "pension.tab"): ( + "150d6fad1fce81254fb7aea1526fbb00b63d4027d6e2ac4c26bb90aea3127eb7", + 1225838, + "2023_24", + "licensed", + "UK Data Service End User Licence", + ), + ("hmrc", "hmrc-spi-public-use-tape-2022-23", 2022, "put2223uk.tab"): ( + "5ef829461060c91a2a47be59ad541d9b519fc3976d66ca80d4920f711bb96f66", + 141323762, + "2022-23", + "restricted", + "UK Data Service End User Licence (study SN 9422)", + ), +} + +CONSUMER_PIN_COMMITS = { + "packages/microcosm-build/src/microcosm/build/uk/source_stages.json": ( + "2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77" + ), + "packages/microcosm-build/src/microcosm/build/uk/hmrc_income_source_stages.json": ( + "de7451bd19ca46d2967e73cdf393908d29e72542" + ), +} + + +def _committed_entries(package: Path) -> list[tuple[dict, int, dict]]: + manifest = yaml.safe_load((package / "manifest.yaml").read_text()) + assert manifest["kind"] == "microdata_release" + return [ + (manifest, year, entry) + for year, entries in manifest["files"].items() + for entry in entries ] - assert commands - - captured: dict[str, argparse.ArgumentParser] = {} - real_parse = argparse.ArgumentParser.parse_args - - def capture(self, args=None, namespace=None): - captured["parser"] = self - raise SystemExit(0) - - argparse.ArgumentParser.parse_args = capture - try: - with contextlib.suppress(SystemExit): - harness_main(["--help"]) - finally: - argparse.ArgumentParser.parse_args = real_parse - - parser = captured["parser"] - for command in commands: - argv = shlex.split(command)[3:] - namespace = parser.parse_args(argv) - assert namespace.command == "fetch-artifact" - assert namespace.access == "public" - assert namespace.licence - assert namespace.kind == "microdata_release" - - -def test_plan_never_prints_a_fabricated_url(): - """A release Microcosm does not pin a URL for prints TODO, not a guess.""" - import subprocess - import sys - - result = subprocess.run( - [ - sys.executable, - str(REPO_ROOT / "scripts" / "register_microdata_releases.py"), - "--microcosm-root", - str(MICROCOSM_ROOT), - "plan", - ], - capture_output=True, - text=True, - cwd=REPO_ROOT, - ) - if result.returncode != 0: - pytest.skip("No readable microcosm checkout") - - assert "TODO_PUBLISHER_URL" in result.stdout -# -------------------------------------------------------------------------- -# The committed registrations -# -------------------------------------------------------------------------- - +def test_committed_pins_match_the_golden_mapping_exactly(): + committed = { + (manifest["source_id"], manifest["package_id"], year, entry["filename"]): ( + entry["sha256"], + entry["size_bytes"], + entry["vintage"], + entry["access"], + entry["licence"], + ) + for package in (FRS_PACKAGE, SPI_PACKAGE) + for manifest, year, entry in _committed_entries(package) + } -def _committed_entries(package: Path) -> list[dict]: - manifest = yaml.safe_load((package / "manifest.yaml").read_text()) - assert manifest["kind"] == "microdata_release" - return [entry for entries in manifest["files"].values() for entry in entries] + assert committed == GOLDEN_PINS + assert len(GOLDEN_PINS) == 15 def test_committed_frs_registration_covers_every_pinned_tab(): - entries = _committed_entries(FRS_PACKAGE) - - assert [entry["filename"] for entry in entries] == [ - "accounts.tab", - "adult.tab", - "benefits.tab", - "benunit.tab", - "child.tab", - "chldcare.tab", - "extchild.tab", - "househol.tab", - "job.tab", - "maint.tab", - "mortgage.tab", - "oddjob.tab", - "penprov.tab", - "pension.tab", - ] + filenames = [entry["filename"] for _m, _y, entry in _committed_entries(FRS_PACKAGE)] + assert filenames == sorted( + key[3] for key in GOLDEN_PINS if key[1] == "dwp-frs-2023-24" + ) @pytest.mark.parametrize("package", [FRS_PACKAGE, SPI_PACKAGE]) -def test_committed_registrations_are_identity_only(package): +def test_committed_registrations_are_consumer_pins_without_bytes(package): entries = _committed_entries(package) assert entries - for entry in entries: + for _manifest_payload, _year, entry in entries: assert entry["access"] in {"licensed", "restricted"} - assert entry["licence"] - assert entry["vintage"] - assert entry["verified_at"] - assert len(entry["sha256"]) == 64 - assert int(entry["size_bytes"]) > 0 + assert entry["hash_source"] == "consumer_pin" + assert entry["attested_by"] == "PolicyEngine/microcosm" + assert entry["pinned_from"]["repository"] == "PolicyEngine/microcosm" + assert ( + entry["pinned_from"]["commit"] + == (CONSUMER_PIN_COMMITS[entry["pinned_from"]["path"]]) + ) + assert "verified_at" not in entry assert "storage" not in entry # No bytes accompany a hash-only registration. assert not (package / entry["filename"]).exists() - - -@pytest.mark.parametrize("package", [FRS_PACKAGE, SPI_PACKAGE]) -def test_committed_registrations_hold_no_microdata_bytes(package): assert sorted(path.name for path in package.iterdir()) == ["manifest.yaml"] -def test_committed_registrations_pass_inventory(): +def test_committed_registrations_pass_inventory_with_the_golden_identities(): report = inventory_source_artifacts(REPO_ROOT / "db" / "data") hash_only = [entry for entry in report.entries if entry.hash_only] assert report.valid - assert len(hash_only) == 15 assert all(entry.valid and not entry.exists for entry in hash_only) assert all(entry.r2 is None for entry in hash_only) + identities = { + ( + yaml.safe_load(Path(entry.manifest_path).read_text())["source_id"], + yaml.safe_load(Path(entry.manifest_path).read_text())["package_id"], + int(entry.year), + entry.filename, + ): (entry.sha256_expected, entry.size_bytes) + for entry in hash_only + } + assert identities == { + key: (value[0], value[1]) for key, value in GOLDEN_PINS.items() + } def test_no_committed_registration_computes_an_r2_key(): From d88373da778732090fc60a1f27bd189d25038dbd Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 02:19:04 -0400 Subject: [PATCH 39/42] Document the explicit kind, the evidence rule, attestation and staging The harness doc's Hash-Only Registrations section now states the manifest kind rule and its frozen list, the closed hash_source set and the attester fields each carries, bare case-folded filenames and one vintage key, the strict pre-fetch validation, the byte-reader refusal, and the public-release rule: an allowlisted licence with artifact-bound evidence, --expected-sha256, staging outside the repository, and the no-tracked-bytes guard. The README quick-start and the storage note follow. Co-Authored-By: Claude Fable 5.1 --- README.md | 14 +++- docs/agent-source-package-harness.md | 107 ++++++++++++++++++++++----- docs/storage-architecture.md | 10 +++ 3 files changed, 111 insertions(+), 20 deletions(-) diff --git a/README.md b/README.md index 08dc05d0..237bee9e 100644 --- a/README.md +++ b/README.md @@ -386,7 +386,8 @@ uv run chronicle inventory-artifacts --root db/data uv run chronicle publish-raw --root db/data # Register a licensed or restricted release by identity alone. No bytes are -# fetched, stored, or uploaded, and no R2 key is recorded: +# fetched, stored, or uploaded, and no R2 key is recorded. The checksum is the +# consumer's: name the attester and where the pin was read from: uv run chronicle register-artifact \ --source-id dwp \ --package-id dwp-frs-2023-24 \ @@ -399,7 +400,16 @@ uv run chronicle register-artifact \ --licence "UK Data Service End User Licence" \ --access licensed \ --doi 10.5255/UKDA-SN-9367-2 \ - --verified-at 2026-09-02 + --hash-source consumer_pin \ + --attested-by PolicyEngine/microcosm \ + --pinned-from-repository PolicyEngine/microcosm \ + --pinned-from-path packages/microcosm-build/src/microcosm/build/uk/source_stages.json \ + --pinned-from-commit 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 + +# A public microdata release is archived only against a reviewed checksum and +# licence evidence binding the file to an allowlisted term; its bytes are +# staged outside the repository and uploaded from there. See +# docs/agent-source-package-harness.md, "Hash-Only Registrations". ``` To coordinate broad PE source migration without jumping straight to semantic diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 2cdd1579..20585611 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -46,19 +46,40 @@ emit a one-time deprecation warning naming the `CHRONICLE_` variable to set instead; see "Environment Variable Rename Window" in [`docs/storage-architecture.md`](storage-architecture.md#environment-variable-rename-window). +## Manifest Kinds + +Every manifest declares `kind`: `publisher_table` (one file per vintage, +parsed by a source package) or `microdata_release` (a registered release, +never parsed). The manifests that predate this rule are frozen, byte for +byte, in `chronicle/grandfathered_manifests.py` and read as publisher tables +only while they still match the freeze; `fetch-artifact` writes `kind` onto +any manifest it touches, so a frozen manifest leaves the freeze the first time +it is modified. A kindless manifest outside that list is an error at every +entry point — `fetch-artifact`, `publish-raw`, `inventory-artifacts`, +`validate-package` and the source-package byte reader — and never a publisher +table by default. A manifest's kind is fixed once declared: `fetch-artifact +--kind` must match it, and a conflicting kind is refused before anything is +read. + ## Hash-Only Registrations Not every raw artifact a build starts from may be redistributed. Every manifest file entry carries an `access` class from a closed set — `public`, -`licensed`, or `restricted` — and a `licence` naming the publisher's terms as -an identifier or URL. `public` is inferred when an entry omits `access`, and -`fetch-artifact` now writes the class explicitly onto every entry it touches. -Both fields are required on a `kind: microdata_release` manifest. +`licensed`, or `restricted` — and a `licence` naming the publisher's terms. +`public` is inferred when a publisher-table entry omits `access`, and +`fetch-artifact` writes the class explicitly onto every entry it touches. +Both fields are required on a `kind: microdata_release` manifest, as is the +attestation of who asserts the checksum: `hash_source` is one of +`chronicle_fetch` (`attested_by: chronicle`, `verified_at` = fetch date), +`consumer_attested` (`attested_by` = the consumer, `attestation_evidence`, +`verified_at`) or `consumer_pin` (`attested_by` = the consumer, `pinned_from` += repository, path and commit, and no `verified_at`). Only `public` bytes enter a Chronicle store. A `licensed` or `restricted` artifact is registered *hash-only*: the manifest records the checksum, size, -vintage, licence, access route, and verification timestamp, and no bytes are -fetched, written, or uploaded. Agents should register one with: +vintage, licence, access route, and attestation, and no bytes are fetched, +written, or uploaded. Chronicle never fetched the bytes, so a hash-only +checksum is always the consumer's. Agents should register one with: ```bash uv run chronicle register-artifact \ @@ -73,28 +94,58 @@ uv run chronicle register-artifact \ --licence "UK Data Service End User Licence" \ --access licensed \ --doi 10.5255/UKDA-SN-9367-2 \ - --verified-at 2026-09-02 + --hash-source consumer_pin \ + --attested-by PolicyEngine/microcosm \ + --pinned-from-repository PolicyEngine/microcosm \ + --pinned-from-path packages/microcosm-build/src/microcosm/build/uk/source_stages.json \ + --pinned-from-commit 2fb2e2f8a99c37725bd6e7a15ff4c2595c912b77 ``` Agents should never invent a checksum to satisfy the command: `--sha256` must be a lowercase 64-character digest taken from a reviewed pin, and a release whose checksum nobody has published is a blocker to record, not a value to guess. The command refuses `--access public`, refuses bytes sitting beside the -manifest, and refuses to write into a `publisher_table` manifest. +manifest, refuses to write into a `publisher_table` manifest, and refuses to +reclassify a filename the manifest already holds as `public` — an archived +release's object must be removed, with its `storage` record, before the entry +can become hash-only. + +Filenames are bare names compared case-insensitively: `./adult.tab`, +`ADULT.TAB` and `sub/../adult.tab` are the file `adult.tab`, and a manifest +never holds one path under two access classes. `2023` and `'2023'` are one +vintage key; a manifest that records both is refused rather than read through +one of them. Every command validates the whole manifest with the codes +`inventory-artifacts` reports before reading, writing, or uploading anything. The other commands enforce the same boundary from their side. `fetch-artifact` refuses a `licensed` or `restricted` access class before reading anything, and -refuses to pull bytes over an entry already registered hash-only. +refuses to pull bytes over an entry already registered hash-only — including +when the filename is inferred from the URL, which happens before the read. `publish-raw` refuses such an entry without reading or uploading its bytes; pass `--skip-hash-only` to publish a tree that deliberately mixes both kinds. `inventory-artifacts` treats a hash-only entry with no local file as valid — the absent bytes are the correct state — and reports an error if the bytes -appear. +appear. The source-package byte reader refuses a hash-only entry before +consulting the package tree, the content-addressed cache, or the publisher, +whatever manifest kind it sits under; `validate-package` reports it as +`hash_only_artifact_not_parseable`, and `build-suite` refuses before creating +its output directory. A `public` microdata release is different: its bytes are redistributable, so -it is acquired with `fetch-artifact` like any other public artifact, and it -does get an R2 key. Pass `--kind microdata_release` so the manifest still -declares what it is: +they are archived — but only with artifact-bound evidence. Being downloadable +is not a licence. The `licence` must be one of the allowlisted terms in +`chronicle/licences.py` (`US-Government-Work`, `OGL-UK-3.0`, `CC0-1.0`, +`CC-BY-4.0`), and the entry carries `licence_evidence` binding this file to +that term: the issuer, the licence identifier, a scope statement, a durable +evidence URL, and the reviewed SHA-256. The fetch therefore takes +`--expected-sha256` and refuses, before writing or uploading, bytes that hash +differently; `--record-revision` does not override that. Release bytes are +staged in a transient directory outside the repository +(`$CHRONICLE_MICRODATA_STAGING_DIR`, default +`~/.cache/policyengine-chronicle/microdata-staging`) and uploaded from there; +a file of that name beside the manifest is refused as tracked microdata bytes, +and the test suite guards that no release package tracks anything but its +manifest. Pass `--kind microdata_release` so the manifest declares what it is: ```bash uv run chronicle fetch-artifact \ @@ -103,17 +154,29 @@ uv run chronicle fetch-artifact \ --package-id census-acs-pums-2022-1yr \ --year 2022 \ --out-dir db/data/census/acs_pums_2022_1yr \ + --publisher "U.S. Census Bureau" \ + --vintage 2022 \ --access public \ - --licence "U.S. Census Bureau public-use file" \ + --licence US-Government-Work \ --kind microdata_release \ + --expected-sha256 \ + --licence-evidence-issuer "U.S. Census Bureau" \ + --licence-evidence-scope "Public-use file of a federal agency; 17 U.S.C. §105" \ + --licence-evidence-url \ --upload-r2 ``` +The entry records `hash_source: chronicle_fetch`, `attested_by: chronicle` +and the fetch date. A public release without licence evidence is classed +`licensed` and registered hash-only instead. + Because several files can share one vintage, a `kind: microdata_release` -manifest may give `files[year]` as a list of entries rather than a single +manifest gives `files[year]` as a list of entries rather than a single mapping — the ACS household and person files above land side by side. A list under any other manifest kind is an error. A fetch replaces only the entry for -its own filename, so acquiring a second file never drops the first. +its own filename, in place, so acquiring a second file never drops the first, +and re-fetching different bytes for one of them is a publisher revision under +the same guard as any other entry. No source package parses a microdata release. `validate-package` fails with `microdata_release_not_parseable` if a package spec points at one, and no @@ -122,8 +185,16 @@ manifest-level identity; see `docs/adr-chronicle-raw-microdata-identity.md`. `scripts/register_microdata_releases.py` drives both halves from a read-only PolicyEngine/microcosm checkout: `emit` writes the hash-only manifests from -Microcosm's reviewed pins, and `plan` prints the `fetch-artifact` commands to -run for public releases from a networked machine. +Microcosm's reviewed pins as `consumer_pin` registrations (recording the +consumer manifest's path and the last commit that changed it, or the commit +given with `--microcosm-commit [PATH=]COMMIT`), and `plan` prints the +`fetch-artifact` commands to run for public releases from a networked machine +with every reviewed identity as an argument. Anything Microcosm does not pin +prints as a `TODO` the command refuses to run with. The suite exercises the +script against `tests/fixtures/microcosm`, a synthetic snapshot of the +consumer manifests; re-deriving the catalogue means re-snapshotting that +fixture and regenerating `tests/fixtures/microcosm/golden_plan.json` in the +same change. For broad PE source migration, generate the agent queue from the manifest before assigning work: diff --git a/docs/storage-architecture.md b/docs/storage-architecture.md index 07f50420..d2c33848 100644 --- a/docs/storage-architecture.md +++ b/docs/storage-architecture.md @@ -103,6 +103,16 @@ and no object is written to either bucket — so the DWP Family Resources Survey tabs, which would otherwise route to `raw/uk/dwp/...`, occupy no key. See `docs/adr-chronicle-raw-microdata-identity.md`. +A public release's key exists only once the fetch has verified the bytes +against the reviewed `--expected-sha256` its `licence_evidence` covers. Its +bytes never enter the repository: they are staged under +`$CHRONICLE_MICRODATA_STAGING_DIR` (default +`~/.cache/policyengine-chronicle/microdata-staging`), keyed like the R2 object, +and uploaded from there. The source-artifact cache under +`~/.cache/policyengine-chronicle/source-artifacts` is a Chronicle store too: +the byte reader refuses a licensed or restricted entry before it would read, +fetch into, or serve from that cache. + Derived artifacts are reproducible and may be replaced by a new build, but a specific `{build_id}` path should be immutable once published. From baf944985f469b86a1a225f54c5a47aaeccbe456 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 03:03:54 -0400 Subject: [PATCH 40/42] Reconcile the rebase onto PR #226's gate round PR #226's gate round taught fetch-artifact to refuse a stray default manifest, a non-mapping files block and a same-bytes rename, to record into a fresh mapping when files: is an explicit null, and to report an object a preserved bucket already holds as skipped. Rebasing this branch over it left three seams, each reproduced failing-first on the rebased tree: - RawArtifactPublishEntry.to_dict listed "skipped" twice (ruff F601). The hash-only skip now shares #226's skipped/uploaded vocabulary behind HASH_ONLY_SKIP_PREFIX, so publish-raw counts both skipped_count and hash_only_refused_count without conflating them. - #226's explicit-null files: cases were refused by the explicit-kind rule: a kindless manifest holding source_id and a bare files: line has no entry that could be read as a publisher table. has_file_entries() says so, and manifest_kind treats an entry-less manifest like an absent one: the command writing its first entry declares the kind. A declared kind stays fixed (register-artifact and fetch-artifact --kind still refuse to reclassify one), and register-artifact records into a fresh mapping on an explicit null just as fetch does. - test_emit_needs_a_commit_it_can_read_or_be_told read a Chronicle commit once the fixture was committed here; it now copies the fixture outside any repository and into one where the manifests are untracked, and refuses in both. _upsert_manifest's repeated revision guard now calls the same _assert_recorded_identity_holds_these_bytes as the preflight, so #226's rename rule (identical bytes under another name, --record-revision does not apply) holds at the write as well; a direct call is pinned by test_the_manifest_write_refuses_a_same_bytes_rename_by_itself. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 37 ++--- chronicle/registration.py | 43 +++++- docs/agent-source-package-harness.md | 8 +- tests/test_chronicle_microdata_catalogue.py | 73 +++++++-- .../test_chronicle_microdata_registration.py | 138 ++++++++++++++++++ 5 files changed, 265 insertions(+), 34 deletions(-) diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 7b73e25e..27fd5d4a 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -45,6 +45,7 @@ ManifestKindError, bare_filename, filename_key, + has_file_entries, is_bare_filename, is_hash_only, iter_file_specs, @@ -562,7 +563,6 @@ def to_dict(self) -> dict[str, Any]: "size_bytes": self.size_bytes, "r2_location": (self.r2_location.to_dict() if self.r2_location else None), "upload": self.upload.to_dict() if self.upload else None, - "skipped": self.skipped, "errors": list(self.errors), } @@ -1754,7 +1754,12 @@ def _resolve_manifest_kind( ) from exc if requested is None: return stored - if existing_manifest and requested != stored: + if requested != stored and ( + existing_manifest.get("kind") is not None or has_file_entries(existing_manifest) + ): + # A declared kind is fixed, and a frozen kindless manifest with + # entries is a publisher table. A manifest with neither is declared + # by this fetch. raise ManifestAccessError( f"{manifest_path} is a {stored} manifest; refusing to fetch into it " f"as a {requested}. A manifest's kind is fixed once declared: " @@ -2384,22 +2389,20 @@ def _upsert_manifest( size_bytes=size_bytes, ) new_r2 = r2_location.to_dict() if r2_location is not None else None + # Different bytes under the same vintage, or the same bytes under another + # name: fetch_source_artifact refuses both before the read; the guard is + # repeated here so no caller can reach a false-provenance write. + _assert_recorded_identity_holds_these_bytes( + identity, + manifest_path=manifest_path, + year=key, + filename=filename, + sha256=sha256, + size_bytes=size_bytes, + r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), + record_revision=record_revision, + ) holds = identity is not None and identity.holds(sha256=sha256, filename=filename) - if identity is not None and not holds and not record_revision: - # Different bytes under the same vintage. The guard in - # fetch_source_artifact refuses this without --record-revision; repeat - # the check here so no caller can reach a false-provenance write. - raise SourceArtifactRevisionError( - _revision_error_message( - manifest_path=manifest_path, - year=key, - filename=filename, - identity=identity, - sha256=sha256, - size_bytes=size_bytes, - r2_bucket=(new_r2 or {}).get("bucket") or default_r2_raw_bucket(), - ) - ) if holds and identity.r2 is not None: # A recorded storage.r2 block for these exact bytes is historical # truth: archived witness records pin raw R2 URLs by hash. Re-fetching diff --git a/chronicle/registration.py b/chronicle/registration.py index 179d5189..a698c243 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -163,14 +163,19 @@ def manifest_kind( """Return a manifest's declared kind. An absent or empty manifest has the default kind: the command creating it - declares one. A manifest with content must declare ``kind`` itself, unless - ``manifest_path`` names a file frozen kindless before the rule and its - bytes still match the freeze. + declares one. So does a manifest that declares no file entry (a bare + ``files:`` line or an empty mapping): there is nothing in it that could be + read as a publisher table, and the command writing its first entry + declares the kind. A manifest with entries must declare ``kind`` itself, + unless ``manifest_path`` names a file frozen kindless before the rule and + its bytes still match the freeze. """ if not isinstance(manifest, Mapping) or not manifest: return DEFAULT_MANIFEST_KIND declared = manifest.get("kind") if declared is None: + if not has_file_entries(manifest): + return DEFAULT_MANIFEST_KIND if manifest_path is not None and is_grandfathered_manifest(manifest_path): return PUBLISHER_TABLE_KIND where = str(manifest_path) if manifest_path is not None else "Manifest" @@ -390,6 +395,25 @@ def iter_manifest_entries( yield key, None, spec +def has_file_entries(manifest: Mapping[str, Any] | None) -> bool: + """Whether a manifest declares any file entry at all. + + A ``files`` block that is absent, an explicit null (a bare ``files:`` + line) or an empty mapping declares nothing; a vintage key holding an empty + list declares nothing either. A ``files`` value that is not a mapping is + content Chronicle cannot read, and counts as entries so that the kind rule + and the ``files_not_a_mapping`` refusal both fire on it. + """ + if not isinstance(manifest, Mapping): + return False + files = manifest.get("files") + if files is None: + return False + if not isinstance(files, Mapping): + return True + return any(True for _entry in iter_manifest_entries(manifest)) + + def registration_id( *, source_id: str, @@ -847,7 +871,12 @@ def register_hash_only_artifact( existing_kind = manifest_kind(payload, manifest_path=manifest_path) except ManifestAccessError as exc: raise HashOnlyRegistrationError(str(exc)) from exc - if payload and existing_kind != MICRODATA_RELEASE_KIND: + if existing_kind != MICRODATA_RELEASE_KIND and ( + payload.get("kind") is not None or has_file_entries(payload) + ): + # A declared kind is fixed, and a frozen kindless manifest with + # entries is a publisher table. A manifest with neither is declared + # by this write. raise HashOnlyRegistrationError( f"{manifest_path} is a {existing_kind} manifest; hash-only " "registrations belong in a kind: microdata_release manifest." @@ -911,7 +940,10 @@ def register_hash_only_artifact( payload.setdefault("source_page", source_page) if table: payload.setdefault("table", table) - payload.setdefault("files", {}) + if payload.get("files") is None: + # setdefault keeps an explicit null (a bare ``files:`` line); the + # entry below needs a mapping to record into. + payload["files"] = {} entries = _existing_entries(payload["files"], key) wanted = filename_key(artifact_name) @@ -1196,6 +1228,7 @@ def _dedupe(values: Iterable[str]) -> list[str]: "bare_filename", "entry_access", "filename_key", + "has_file_entries", "is_bare_filename", "is_hash_only", "is_microdata_release", diff --git a/docs/agent-source-package-harness.md b/docs/agent-source-package-harness.md index 20585611..909ed14c 100644 --- a/docs/agent-source-package-harness.md +++ b/docs/agent-source-package-harness.md @@ -57,9 +57,11 @@ any manifest it touches, so a frozen manifest leaves the freeze the first time it is modified. A kindless manifest outside that list is an error at every entry point — `fetch-artifact`, `publish-raw`, `inventory-artifacts`, `validate-package` and the source-package byte reader — and never a publisher -table by default. A manifest's kind is fixed once declared: `fetch-artifact ---kind` must match it, and a conflicting kind is refused before anything is -read. +table by default. A manifest that declares no file entry yet (a bare +`files:` line or an empty mapping) has nothing to classify, and the command +writing its first entry declares the kind. A manifest's kind is fixed once +declared: `fetch-artifact --kind` must match it, and a conflicting kind is +refused before anything is read. ## Hash-Only Registrations diff --git a/tests/test_chronicle_microdata_catalogue.py b/tests/test_chronicle_microdata_catalogue.py index 9f9d8213..dc8d17c6 100644 --- a/tests/test_chronicle_microdata_catalogue.py +++ b/tests/test_chronicle_microdata_catalogue.py @@ -262,35 +262,90 @@ def test_emit_refuses_a_pin_that_drifted_from_the_committed_one(tmp_path, capsys assert target.read_bytes() == FRS_MANIFEST.read_bytes() +def _fixture_copy(destination: Path) -> Path: + for path in FIXTURE_ROOT.rglob("*.json"): + copy = destination / path.relative_to(FIXTURE_ROOT) + copy.parent.mkdir(parents=True, exist_ok=True) + copy.write_bytes(path.read_bytes()) + return destination + + +def _git(repo: Path, *args: str) -> str: + return subprocess.run( + ["git", "-C", str(repo), *args], capture_output=True, text=True, check=True + ).stdout.strip() + + def test_emit_needs_a_commit_it_can_read_or_be_told(tmp_path, capsys): + # The fixture inside this repository is committed, so a run against it + # would read a Chronicle commit as if it were the consumer's. Outside any + # repository there is no commit to read; inside one whose manifests are + # untracked there is none either. Both refuse before writing. + outside = _fixture_copy(tmp_path / "outside-git") + assert not (outside / ".git").exists() + untracked = _fixture_copy(tmp_path / "untracked") + _git(untracked, "init", "-q") + _git(untracked, "config", "user.email", "t@example.com") + _git(untracked, "config", "user.name", "t") + (untracked / "README").write_text("nothing pinned here") + _git(untracked, "add", "README") + _git(untracked, "commit", "-q", "-m", "unrelated") + + for checkout in (outside, untracked): + exit_code, _out, err = _run( + [ + "--microcosm-root", + str(checkout), + "--root", + str(tmp_path / "data"), + "emit", + ], + capsys, + ) + assert exit_code == 1 + assert "--microcosm-commit" in err + assert not (tmp_path / "data").exists() + exit_code, _out, err = _run( [ "--microcosm-root", - str(FIXTURE_ROOT), + str(outside), "--root", str(tmp_path / "data"), "emit", + "--microcosm-commit", + "not-a-commit", ], capsys, ) - assert exit_code == 1 - assert "pass --microcosm-commit" in err + assert exit_code == 2 + assert "40-hex commit" in err assert not (tmp_path / "data").exists() - exit_code, _out, err = _run( + +def test_emit_from_a_checkout_outside_git_registers_with_the_given_commit( + tmp_path, capsys +): + outside = _fixture_copy(tmp_path / "outside-git") + + exit_code, out, err = _run( [ "--microcosm-root", - str(FIXTURE_ROOT), + str(outside), "--root", str(tmp_path / "data"), + "--json", "emit", - "--microcosm-commit", - "not-a-commit", + *PIN_COMMIT_ARGS, ], capsys, ) - assert exit_code == 2 - assert "40-hex commit" in err + + assert exit_code == 0, err + assert len(json.loads(out)["registrations"]) == 15 + assert (tmp_path / "data" / "dwp/frs_2023_24/manifest.yaml").read_bytes() == ( + FRS_MANIFEST.read_bytes() + ) def test_pin_commit_reads_the_last_commit_that_changed_the_file(tmp_path): diff --git a/tests/test_chronicle_microdata_registration.py b/tests/test_chronicle_microdata_registration.py index 53797e35..ecef4e86 100644 --- a/tests/test_chronicle_microdata_registration.py +++ b/tests/test_chronicle_microdata_registration.py @@ -29,6 +29,8 @@ ExpectedArtifactIdentityError, MalformedManifestError, SourceArtifactRevisionError, + _expected_identity, + _upsert_manifest, fetch_source_artifact, inventory_source_artifacts, microdata_staging_path, @@ -47,6 +49,7 @@ bare_filename, entry_access, filename_key, + has_file_entries, is_bare_filename, is_hash_only, is_microdata_release, @@ -267,6 +270,21 @@ def test_manifest_kind_is_explicit_except_for_an_absent_manifest(): manifest_kind({"files": {2023: {"filename": "table.xlsx"}}}) +def test_a_manifest_with_no_entries_has_nothing_to_classify(): + # A bare ``files:`` line or an empty mapping declares no entry that could + # be misread as a publisher table, so the manifest reads like an absent + # one and the command writing its first entry declares the kind. + assert has_file_entries({"source_id": "irs_soi", "files": None}) is False + assert has_file_entries({"source_id": "irs_soi", "files": {}}) is False + assert has_file_entries({"files": {2023: {"filename": "t.xlsx"}}}) is True + assert has_file_entries({"files": {2023: []}}) is False + assert has_file_entries({"files": ["not", "a", "mapping"]}) is True + assert manifest_kind({"source_id": "irs_soi", "files": None}) == "publisher_table" + assert manifest_kind({"source_id": "irs_soi", "files": {}}) == "publisher_table" + with pytest.raises(ManifestAccessError, match="declares no kind"): + manifest_kind({"source_id": "irs_soi", "files": {2023: {"filename": "t"}}}) + + def test_hash_sources_are_the_closed_contract_set(): assert HASH_SOURCES == ("chronicle_fetch", "consumer_attested", "consumer_pin") @@ -769,6 +787,39 @@ def test_register_refuses_a_kindless_manifest_with_content(tmp_path): assert (output_dir / "manifest.yaml").read_bytes() == original +def _entryless_manifest(tmp_path: Path, **fields: object) -> Path: + output_dir = tmp_path / "pkg" + output_dir.mkdir() + payload = {"source_id": "dwp", "package_id": "dwp-frs-2023-24", **fields} + (output_dir / "manifest.yaml").write_text( + yaml.safe_dump(payload, sort_keys=False).replace("files: null", "files:") + ) + return output_dir + + +@pytest.mark.parametrize("files", [None, {}], ids=["explicit-null", "empty-mapping"]) +def test_register_declares_the_kind_of_an_entryless_kindless_manifest(tmp_path, files): + output_dir = _entryless_manifest(tmp_path, files=files) + + report = _register(output_dir) + manifest = _manifest(output_dir) + + assert report.valid + assert manifest["kind"] == "microdata_release" + assert manifest["files"][2023][0]["filename"] == "adult.tab" + + +def test_register_never_reclassifies_a_declared_publisher_table(tmp_path): + # An explicit kind is fixed even when the manifest holds no entry yet. + output_dir = _entryless_manifest(tmp_path, kind="publisher_table", files={}) + original = (output_dir / "manifest.yaml").read_bytes() + + with pytest.raises(HashOnlyRegistrationError, match="publisher_table manifest"): + _register(output_dir) + + assert (output_dir / "manifest.yaml").read_bytes() == original + + def test_register_refuses_a_manifest_for_a_different_source(tmp_path): output_dir = tmp_path / "pkg" _register(output_dir) @@ -1283,6 +1334,55 @@ def test_fetch_refuses_a_kind_that_conflicts_with_the_manifest(tmp_path, monkeyp assert not (path / "codebook.pdf").exists() +@pytest.mark.parametrize("files", [None, {}], ids=["explicit-null", "empty-mapping"]) +def test_fetch_declares_the_requested_kind_on_an_entryless_kindless_manifest( + tmp_path, monkeypatch, files +): + output_dir = _entryless_manifest(tmp_path, files=files) + _serve(monkeypatch, PUBLIC_BYTES) + + report = _fetch_release( + output_dir, + staging_dir=tmp_path / "staging", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + licence="OGL-UK-3.0", + publisher="Department for Work and Pensions", + vintage="2023_24", + licence_evidence={**EVIDENCE, "issuer": "DWP"}, + ) + manifest = _manifest(output_dir) + + assert report.valid + assert manifest["kind"] == "microdata_release" + assert [entry["filename"] for entry in manifest["files"][2023]] == ["csv_hus.zip"] + + +def test_fetch_never_reclassifies_a_declared_kind_without_entries( + tmp_path, monkeypatch +): + output_dir = _entryless_manifest(tmp_path, kind="publisher_table", files={}) + original = (output_dir / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + + with pytest.raises(ManifestAccessError, match="is a publisher_table manifest"): + _fetch_release( + output_dir, + staging_dir=tmp_path / "staging", + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + licence="OGL-UK-3.0", + publisher="Department for Work and Pensions", + vintage="2023_24", + licence_evidence={**EVIDENCE, "issuer": "DWP"}, + ) + + assert reads == [] + assert (output_dir / "manifest.yaml").read_bytes() == original + + def test_fetch_refuses_a_stored_unknown_kind_even_with_an_explicit_kind( tmp_path, monkeypatch ): @@ -1766,6 +1866,44 @@ def test_a_second_file_never_turns_a_publisher_table_vintage_into_a_list( assert revised["filename"] == "22in05ira_rev.xlsx" +def test_the_manifest_write_refuses_a_same_bytes_rename_by_itself(tmp_path): + # PR #226's rule: identical bytes under another filename are a rename, + # not a revision, so --record-revision does not apply. The write path + # repeats the guard so no caller reaches a false-provenance write. + output_dir = tmp_path / "pkg" + (tmp_path / "22in05ira.xlsx").write_bytes(b"IRA table 5") + _fetch_table(tmp_path / "22in05ira.xlsx", output_dir, year=2022) + original = (output_dir / "manifest.yaml").read_bytes() + + for record_revision in (False, True): + with pytest.raises(SourceArtifactRevisionError, match="rename is not"): + _upsert_manifest( + output_dir / "manifest.yaml", + source_id="irs_soi", + package_id="soi-table-1-2", + dataset="irs_soi_soi-table-1-2", + source_page=None, + table=None, + publisher=None, + year=2022, + filename="table-5.xlsx", + source_url="https://publisher.example/table-5.xlsx", + sha256=hashlib.sha256(b"IRA table 5").hexdigest(), + size_bytes=len(b"IRA table 5"), + fetched_at="2026-09-04T00:00:00+00:00", + access="public", + licence=None, + kind="publisher_table", + vintage=None, + licence_evidence=None, + expected=_expected_identity(None, None), + r2_location=None, + record_revision=record_revision, + ) + + assert (output_dir / "manifest.yaml").read_bytes() == original + + def test_a_second_file_over_an_unidentified_table_entry_is_refused( tmp_path, monkeypatch ): From c0d9d74dcdfa2fa7e477e4ffee8d09a8c519a333 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 05:34:27 -0400 Subject: [PATCH 41/42] Hold the byte boundary across every manifest a package directory keeps PR #226's --manifest lets a fetch record into one of several manifests in a publisher directory. The hash-only guard read only that manifest, so a release registered licensed in manifest.yaml could be fetched, published through a sibling manifest_.yaml, parsed by a source package that reads through the sibling, or reclassified by register-artifact, all under the very name the registration says must never hold bytes. Reproduced failing-first in tests/test_chronicle_package_directory.py (18 tests). The boundary is now the file in the directory, not the manifest naming it: - fetch-artifact reads every manifest the directory keeps before the publisher is touched and refuses a name or a digest any of them registers hash-only (the reviewed pin's digest before the read, the served bytes' digest before anything is written), refuses to overwrite a file a sibling records as other bytes, and refuses a manifest whose source_id/package_id differ from the fetch's. An artifact may not be named like a manifest, and --manifest must name manifest.yaml or manifest_.yaml. - publish-raw and inventory-artifacts report filename_collision_across_ manifests / sha256_collision_across_manifests (a name or digest public in one manifest and hash-only in another, or public under two digests) and publish nothing under such a directory; the tracked shape of one public file recorded by two manifests as the same bytes is accepted. validate_manifest_files gains sha256_collision within one manifest. - register-artifact takes --manifest with the same stray-default refusal as fetch, refuses a manifest-named filename, and refuses an identity (by name or digest) any manifest in the directory holds public or archived. - The source-package byte reader refuses a file a sibling manifest registers hash-only, by name or digest, before any store is consulted. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 205 +++++- chronicle/harness.py | 13 + chronicle/registration.py | 220 ++++++- chronicle/source_package.py | 51 ++ .../test_chronicle_microdata_registration.py | 4 +- tests/test_chronicle_package_directory.py | 591 ++++++++++++++++++ 6 files changed, 1063 insertions(+), 21 deletions(-) create mode 100644 tests/test_chronicle_package_directory.py diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 27fd5d4a..70fc3249 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -40,24 +40,30 @@ MANIFEST_KINDS, MICRODATA_RELEASE_KIND, AmbiguousVintageKeyError, + ArtifactFilenameError, ListSpecRejected, ManifestAccessError, ManifestKindError, bare_filename, filename_key, has_file_entries, + hash_only_registrations, is_bare_filename, is_hash_only, + is_manifest_filename, + iter_directory_entries, iter_file_specs, iter_manifest_entries, manifest_kind as normalize_manifest_kind, normalize_access, + package_manifest_paths, recorded_r2, resolve_vintage_key, safe_entry_access, safe_manifest_kind, validate_file_entry, validate_manifest_files, + validate_package_directory, ) @@ -138,9 +144,116 @@ def _manifest_path(output: Path, manifest_filename: str) -> Path: "Manifest must name a file inside the package directory, not " f"{manifest_filename!r}." ) + if not is_manifest_filename(name): + raise ManifestNameError( + f"Manifest must be named {DEFAULT_MANIFEST_FILENAME} or " + f"manifest_.yaml, not {manifest_filename!r}: the sweeps " + "address a package's manifests by those names, and a manifest " + "under any other name is invisible to them." + ) return output / name +def _package_manifests( + output: Path, + manifest_path: Path, + existing_manifest: dict[str, Any], +) -> dict[str, dict[str, Any]]: + """Return every manifest the package directory keeps, by path. + + The byte boundary is the file in the directory, not the manifest that + records it: a name or a digest registered hash-only in any manifest there + must not be fetched, published, or reclassified through another. A + sibling manifest Chronicle cannot read is a refusal, because the boundary + cannot be decided without it. + """ + manifests: dict[str, dict[str, Any]] = {str(manifest_path): existing_manifest} + for path in package_manifest_paths(output): + if path == manifest_path or filename_key(path.name) == filename_key( + manifest_path.name + ): + continue + manifests[str(path)] = _read_manifest(path) + return manifests + + +def _assert_no_hash_only_bytes( + manifests: Mapping[str, dict[str, Any]], + *, + sha256: str, + filename: str, + what: str, +) -> None: + """Refuse to archive bytes any manifest in the directory registers hash-only. + + The same bytes under another name are the same gated artifact. + """ + for name, key, entry in hash_only_registrations(manifests, sha256=sha256): + raise ManifestAccessError( + f"{name} registers {entry.get('filename')!r} for {key!r} as " + f"access={safe_entry_access(entry)!r} with sha256={sha256}; {what} " + "are those exact bytes. A gated artifact is not archived under " + f"another name ({filename!r}); keep the hash-only registration." + ) + + +def _assert_siblings_record_these_bytes( + manifests: Mapping[str, dict[str, Any]], + *, + manifest_path: Path, + filename: str, + sha256: str, +) -> None: + """Refuse to overwrite a file another manifest records as other bytes. + + Two manifests may record one public file only as the same bytes; a fetch + of different bytes under that name would silently rewrite what the other + manifest describes. A revision is recorded through the manifest that + holds the entry. + """ + wanted = filename_key(filename) + for name, key, _index, entry in iter_directory_entries(manifests): + if name == str(manifest_path) or not isinstance(entry, dict): + continue + recorded_name = entry.get("filename") + if recorded_name is None or filename_key(recorded_name) != wanted: + continue + recorded = entry.get("sha256") + recorded = recorded.strip() if isinstance(recorded, str) else None + if recorded and recorded != sha256: + raise ManifestAccessError( + f"{name} records {recorded_name!r} for {key!r} as " + f"sha256={recorded}; this fetch would write sha256={sha256} to " + "the same file. One file in a package directory has one record " + "of its bytes: revise it through the manifest that records it " + f"(fetch-artifact --manifest {Path(name).name} --record-revision)." + ) + + +def _assert_manifest_identifies( + existing_manifest: dict[str, Any], + manifest_path: Path, + *, + source_id: str, + package_id: str, +) -> None: + """Refuse to fetch into a manifest that identifies another package. + + The R2 key and the registration identity are built from the fetch's + identifiers; recording the entry under a manifest that declares others + would leave the two disagreeing about which package the bytes belong to. + """ + for field, value in (("source_id", source_id), ("package_id", package_id)): + declared = existing_manifest.get(field) + declared = declared.strip() if isinstance(declared, str) else declared + if declared not in (None, "") and str(declared) != value: + raise ManifestAccessError( + f"{manifest_path} declares {field}={declared!r}; refusing to " + f"fetch {field}={value!r} into it. Fetch into the package the " + "manifest identifies, or into that package's own directory." + ) + + def _sibling_manifests(output: Path) -> list[str]: """Return the ``manifest_*.yaml`` files a package directory keeps.""" if not output.is_dir(): @@ -781,14 +894,19 @@ def fetch_source_artifact( # resolved -- and any alias of a registered name refused -- before the # publisher is read. Reading first and refusing afterwards would pull # gated bytes to decide their name. + what = ( + "--filename" if filename is not None else "The filename inferred from the URL" + ) artifact_filename = bare_filename( filename if filename is not None else _infer_artifact_filename(source_url), - what=( - "--filename" - if filename is not None - else "The filename inferred from the URL" - ), + what=what, ) + if is_manifest_filename(artifact_filename): + raise ArtifactFilenameError( + f"{what} {artifact_filename!r} is a manifest name. An artifact may " + "not be named like a manifest, which it would overwrite; pass " + "--filename with the publisher's name for the bytes." + ) expected = _expected_identity(expected_sha256, expected_size_bytes) resolved_r2_prefix = resolve_r2_prefix( prefix=r2_prefix, @@ -805,10 +923,32 @@ def fetch_source_artifact( _refuse_a_stray_default_manifest(output, manifest_path) existing_manifest = _read_manifest(manifest_path) _manifest_files(existing_manifest, manifest_path) - # The byte boundary is checked first: overwriting a hash-only registration - # with bytes is the more serious refusal, and its message is the one the - # caller needs, not a prompt about the manifest's kind or licence. - _assert_no_hash_only_entry(existing_manifest, manifest_path, artifact_filename) + # The byte boundary is checked first, across every manifest the directory + # keeps: overwriting a hash-only registration with bytes is the more + # serious refusal, and its message is the one the caller needs, not a + # prompt about the manifest's kind or licence. + manifests = _package_manifests(output, manifest_path, existing_manifest) + for sibling_path, sibling in manifests.items(): + _assert_no_hash_only_entry(sibling, Path(sibling_path), artifact_filename) + if expected.sha256: + _assert_no_hash_only_bytes( + manifests, + sha256=expected.sha256, + filename=artifact_filename, + what="the reviewed pin's bytes", + ) + _assert_siblings_record_these_bytes( + manifests, + manifest_path=manifest_path, + filename=artifact_filename, + sha256=expected.sha256, + ) + _assert_manifest_identifies( + existing_manifest, + manifest_path, + source_id=source_id, + package_id=package_id, + ) manifest_kind_value = _resolve_manifest_kind( existing_manifest, manifest_path=manifest_path, @@ -902,6 +1042,18 @@ def fetch_source_artifact( r2_bucket=r2_bucket, record_revision=record_revision, ) + _assert_no_hash_only_bytes( + manifests, + sha256=sha256, + filename=artifact_filename, + what=f"the bytes served by {source_url}", + ) + _assert_siblings_record_these_bytes( + manifests, + manifest_path=manifest_path, + filename=artifact_filename, + sha256=sha256, + ) if release: # Public microdata never lands in the package tree: it is staged in an @@ -1189,10 +1341,20 @@ def publish_source_artifacts( kind, kind_error = safe_manifest_kind(manifest, manifest_path=manifest_path) manifest_errors = [kind_error] if kind_error else [] manifest_errors.extend(validate_manifest_files(manifest)) + try: + manifest_errors.extend( + validate_package_directory( + _package_manifests(manifest_path.parent, manifest_path, manifest) + ) + ) + except (OSError, MalformedManifestError) as exc: + errors.append(f"Could not read a manifest beside {manifest_path}: {exc}") + continue if manifest_errors: - # Validate, then touch: a manifest Chronicle cannot classify or - # whose entries collide is reported and left alone; publishing any - # entry under it could ship bytes through the wrong record. + # Validate, then touch: a manifest Chronicle cannot classify, whose + # entries collide, or whose directory's other manifests disagree + # with it is reported and left alone; publishing any entry under + # it could ship bytes through the wrong record. errors.extend(f"{code}: {manifest_path}" for code in manifest_errors) continue updated = False @@ -1320,6 +1482,15 @@ def inventory_source_artifacts( errors.extend( f"{code}: {manifest_path}" for code in validate_manifest_files(manifest) ) + try: + errors.extend( + f"{code}: {manifest_path}" + for code in validate_package_directory( + _package_manifests(manifest_path.parent, manifest_path, manifest) + ) + ) + except (OSError, MalformedManifestError) as exc: + errors.append(f"Could not read a manifest beside {manifest_path}: {exc}") for year, spec in files.items(): for file_spec in iter_file_specs(spec, kind=kind): entries.append( @@ -2355,7 +2526,15 @@ def _upsert_manifest( kind = _resolve_manifest_kind( payload, manifest_path=manifest_path, requested_kind=kind ) - _assert_no_hash_only_entry(payload, manifest_path, filename) + manifests = _package_manifests(manifest_path.parent, manifest_path, payload) + for sibling_path, sibling in manifests.items(): + _assert_no_hash_only_entry(sibling, Path(sibling_path), filename) + _assert_no_hash_only_bytes( + manifests, sha256=sha256, filename=filename, what="the fetched bytes" + ) + _assert_siblings_record_these_bytes( + manifests, manifest_path=manifest_path, filename=filename, sha256=sha256 + ) payload.setdefault("source_id", source_id) payload.setdefault("package_id", package_id) payload = _with_declared_kind(payload, kind) diff --git a/chronicle/harness.py b/chronicle/harness.py index b450bfa0..9e295609 100644 --- a/chronicle/harness.py +++ b/chronicle/harness.py @@ -429,6 +429,7 @@ def register_artifact_file( fetched_at: str | None = None, notes: str | None = None, allow_reissue: bool = False, + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, ) -> ArtifactRegistrationReport: """Register a licensed or restricted artifact by identity, without bytes.""" return register_hash_only_artifact( @@ -458,6 +459,7 @@ def register_artifact_file( fetched_at=fetched_at, notes=notes, allow_reissue=allow_reissue, + manifest_filename=manifest_filename, ) @@ -1189,6 +1191,16 @@ def main(argv: list[str] | None = None) -> int: required=True, help="Directory where manifest.yaml should live. No bytes are written.", ) + registration_parser.add_argument( + "--manifest", + default=DEFAULT_MANIFEST_FILENAME, + help=( + "Manifest filename inside --out-dir to record into, for a " + "directory that keeps one manifest_.yaml per package. " + "Must be manifest.yaml or manifest_.yaml; the default " + "name is refused beside named manifests." + ), + ) registration_parser.add_argument( "--filename", required=True, @@ -1821,6 +1833,7 @@ def main(argv: list[str] | None = None) -> int: fetched_at=args.fetched_at, notes=args.notes, allow_reissue=args.allow_reissue, + manifest_filename=args.manifest, ) except (HashOnlyRegistrationError, ManifestAccessError) as error: print(f"error: {error}", file=sys.stderr) diff --git a/chronicle/registration.py b/chronicle/registration.py index a698c243..87efca40 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -62,6 +62,9 @@ #: Access class inferred for a publisher-table entry that does not declare one. DEFAULT_ACCESS = ACCESS_PUBLIC +#: The manifest a package directory keeps unless it feeds several packages. +DEFAULT_MANIFEST_FILENAME = "manifest.yaml" + PUBLISHER_TABLE_KIND = "publisher_table" MICRODATA_RELEASE_KIND = "microdata_release" #: The closed set of manifest kinds. @@ -338,6 +341,122 @@ def filename_key(value: Any) -> str: return Path(str(value)).name.casefold() +#: The names a package directory's manifests may carry: ``manifest.yaml`` or +#: ``manifest_.yaml`` (``.yml`` accepted), matched case-insensitively +#: because the directory is as often as not on a case-insensitive filesystem. +_MANIFEST_FILENAME_RE = re.compile(r"^manifest(?:_[^/\\]+)?\.ya?ml$", re.IGNORECASE) + + +def is_manifest_filename(value: Any) -> bool: + """Whether ``value`` is a name a package manifest may carry. + + The sweeps address manifests by name (``manifest.yaml`` by default, + ``manifest_.yaml`` for a directory that feeds several source + packages), so a manifest under any other name is invisible to them, and + an artifact under one of these names would overwrite a manifest. + """ + return is_bare_filename(value) and bool(_MANIFEST_FILENAME_RE.match(str(value))) + + +def package_manifest_paths(package_dir: Path) -> list[Path]: + """Return every manifest file a package directory keeps, sorted by name.""" + directory = Path(package_dir) + if not directory.is_dir(): + return [] + return sorted( + path + for path in directory.iterdir() + if path.is_file() and is_manifest_filename(path.name) + ) + + +def iter_directory_entries( + manifests: Mapping[str, Mapping[str, Any] | None], +) -> Iterator[tuple[str, Any, int | None, Any]]: + """Yield ``(manifest_name, key, index, entry)`` across a directory's manifests.""" + for name, manifest in manifests.items(): + for key, index, entry in iter_manifest_entries(manifest): + yield name, key, index, entry + + +def hash_only_registrations( + manifests: Mapping[str, Mapping[str, Any] | None], + *, + filename: Any = None, + sha256: Any = None, +) -> list[tuple[str, Any, Mapping[str, Any]]]: + """Return the hash-only entries across ``manifests`` matching a name or digest. + + A match by case-folded bare filename is the same path in the package + directory; a match by digest is the same bytes under another name. Both + identify the gated artifact, whichever manifest registers it. An entry + whose access class cannot be read is treated as hash-only (never public). + """ + wanted_name = filename_key(filename) if filename else None + wanted_digest = _text(sha256) + matches: list[tuple[str, Any, Mapping[str, Any]]] = [] + for name, key, _index, entry in iter_directory_entries(manifests): + if not isinstance(entry, Mapping): + continue + if not is_hash_only(safe_entry_access(entry)): + continue + entry_name = entry.get("filename") + same_name = ( + wanted_name is not None + and entry_name is not None + and filename_key(entry_name) == wanted_name + ) + same_bytes = wanted_digest is not None and _text(entry.get("sha256")) == ( + wanted_digest + ) + if same_name or same_bytes: + matches.append((name, key, entry)) + return matches + + +def validate_package_directory( + manifests: Mapping[str, Mapping[str, Any] | None], +) -> tuple[str, ...]: + """Return the collision codes across every manifest a directory keeps. + + Two manifests may record one public file only as the same bytes (the + tracked shape: ``manifest.yaml`` beside ``manifest_.yaml`` both + recording one publisher zip with one digest). A name held public in one + manifest and hash-only in another, public under two digests, or a digest + held public in one and hash-only in another, is one file that two records + disagree about, and no command may act through either record. + """ + by_name: dict[str, list[tuple[str, bool, str]]] = {} + by_digest: dict[str, list[tuple[str, bool]]] = {} + for name, _key, _index, entry in iter_directory_entries(manifests): + if not isinstance(entry, Mapping): + continue + hash_only = is_hash_only(safe_entry_access(entry)) + digest = _text(entry.get("sha256")) or "" + filename = entry.get("filename") + if filename is not None: + by_name.setdefault(filename_key(filename), []).append( + (name, hash_only, digest) + ) + if digest: + by_digest.setdefault(digest, []).append((name, hash_only)) + + errors: list[str] = [] + for key, records in by_name.items(): + if len({name for name, _hash_only, _digest in records}) < 2: + continue + classes = {hash_only for _name, hash_only, _digest in records} + digests = {digest for _name, _hash_only, digest in records} + if len(classes) > 1 or (not classes.pop() and len(digests) > 1): + errors.append(f"filename_collision_across_manifests:{key}") + for digest, records in by_digest.items(): + if len({name for name, _hash_only in records}) < 2: + continue + if len({hash_only for _name, hash_only in records}) > 1: + errors.append(f"sha256_collision_across_manifests:{digest}") + return tuple(_dedupe(errors)) + + def vintage_key_forms(year: Any) -> tuple[Any, ...]: """Return the key spellings that address the same vintage as ``year``. @@ -481,6 +600,22 @@ def validate_manifest_files(manifest: Mapping[str, Any] | None) -> tuple[str, .. errors.append(f"non_canonical_filename:{filename}") by_name.setdefault(filename_key(filename), []).append((key, entry)) + by_digest: dict[str, set[bool]] = {} + for _key, _index, entry in iter_manifest_entries(manifest): + if not isinstance(entry, Mapping): + continue + digest = _text(entry.get("sha256")) + if digest: + by_digest.setdefault(digest, set()).add( + is_hash_only(safe_entry_access(entry)) + ) + for digest, classes in by_digest.items(): + if len(classes) > 1: + # The same bytes are the same artifact whatever name they carry: + # a digest is not both bytes Chronicle holds and bytes it must + # never hold. + errors.append(f"sha256_collision:{digest}") + for name, entries in by_name.items(): if len(entries) < 2: continue @@ -534,6 +669,8 @@ def validate_file_entry( filename = spec.get("filename") if filename is not None and not is_bare_filename(filename): errors.append(f"non_canonical_filename:{filename}") + elif filename is not None and is_manifest_filename(filename): + errors.append(f"manifest_named_filename:{filename}") declared_access = spec.get("access") if declared_access is None: @@ -812,6 +949,7 @@ def register_hash_only_artifact( fetched_at: str | None = None, notes: str | None = None, allow_reissue: bool = False, + manifest_filename: str = DEFAULT_MANIFEST_FILENAME, ) -> ArtifactRegistrationReport: """Register a licensed or restricted artifact by identity, without bytes. @@ -820,6 +958,11 @@ def register_hash_only_artifact( asserts the checksum. No bytes are read, written, or uploaded, and no R2 key is recorded. Every refusal below happens before the manifest is touched. + + ``manifest_filename`` names the manifest inside ``output_dir`` the entry + belongs to, exactly as ``fetch-artifact --manifest`` does: a directory + that keeps ``manifest_.yaml`` files and no ``manifest.yaml`` is + refused the default name rather than given a stray third manifest. """ access_class = normalize_access(access) if stores_bytes(access_class): @@ -846,6 +989,11 @@ def register_hash_only_artifact( raise HashOnlyRegistrationError( f"Registration filename must be a bare filename; got {filename!r}." ) + if is_manifest_filename(filename): + raise HashOnlyRegistrationError( + f"Registration filename {filename!r} is a manifest name; an " + "artifact may not be named like a manifest." + ) artifact_name = str(filename) provenance = _hash_only_attestation( hash_source=hash_source, @@ -865,8 +1013,16 @@ def register_hash_only_artifact( "must not live in a Chronicle store." ) - manifest_path = output / "manifest.yaml" + manifest_path = _registration_manifest_path(output, manifest_filename) payload = _load_manifest(manifest_path) + # Every manifest the directory keeps takes part in the identity checks: + # the boundary is the file in the directory, not the manifest naming it. + siblings = { + str(path): _load_manifest(path) + for path in package_manifest_paths(output) + if path != manifest_path + and filename_key(path.name) != filename_key(manifest_path.name) + } try: existing_kind = manifest_kind(payload, manifest_path=manifest_path) except ManifestAccessError as exc: @@ -922,7 +1078,13 @@ def register_hash_only_artifact( f"{', '.join(manifest_errors)}. Fix it by hand before registering " "into it." ) - _assert_no_archived_identity(payload, manifest_path, artifact_name, access_class) + _assert_no_archived_identity( + payload, manifest_path, artifact_name, access_class, sha256=checksum + ) + for sibling_path, sibling in siblings.items(): + _assert_no_archived_identity( + sibling, Path(sibling_path), artifact_name, access_class, sha256=checksum + ) try: vintage_key = resolve_vintage_key(files or {}, year) @@ -1084,31 +1246,39 @@ def _assert_no_archived_identity( manifest_path: Path, artifact_name: str, access_class: str, + *, + sha256: str | None = None, ) -> None: - """Refuse to register hash-only a filename the manifest holds as public. + """Refuse to register hash-only an identity the manifest holds as public. A public entry may have been archived: its object sits in the raw bucket under ``storage.r2`` (or in ``storage.previous_r2`` once revised). Replacing that entry with a hash-only one would leave the bytes in a Chronicle store with nothing recording them, and inventory would report the tree clean. The transition is refused until the public entry, and the - object it names, have been explicitly removed. + object it names, have been explicitly removed. The identity is matched by + case-folded bare filename and by digest: the same bytes archived under + another name are the same artifact. """ wanted = filename_key(artifact_name) for key, _index, existing in iter_manifest_entries(payload): if not isinstance(existing, Mapping): continue - if filename_key(existing.get("filename")) != wanted: + existing_name = existing.get("filename") + same_name = existing_name is not None and filename_key(existing_name) == wanted + same_bytes = sha256 is not None and _text(existing.get("sha256")) == sha256 + if not (same_name or same_bytes): continue recorded = [ str(block.get("uri") or block.get("key") or block) for block in (recorded_r2(existing), *recorded_previous_r2(existing)) if isinstance(block, Mapping) ] + how = "" if same_name else f" (the same bytes as {artifact_name!r})" if recorded: raise HashOnlyRegistrationError( f"{manifest_path} records the R2 object(s) {recorded} for " - f"{existing.get('filename')!r} ({key!r}, " + f"{existing_name!r}{how} ({key!r}, " f"access={safe_entry_access(existing)!r}). Registering it " f"{access_class} would leave those bytes in a Chronicle store " "with nothing recording them. Remove the object and its " @@ -1117,13 +1287,43 @@ def _assert_no_archived_identity( ) if not is_hash_only(safe_entry_access(existing)): raise HashOnlyRegistrationError( - f"{manifest_path} already registers {existing.get('filename')!r} " + f"{manifest_path} already registers {existing_name!r}{how} " f"({key!r}) as access={safe_entry_access(existing)!r}. A change " f"of access class to {access_class!r} is an explicit decision: " "remove the public entry by hand, then register the release." ) +def _registration_manifest_path(output: Path, manifest_filename: Any) -> Path: + """Return the manifest a registration records into, refusing a stray one. + + Mirrors ``fetch-artifact --manifest``: the name is a bare manifest name + inside the package directory, and the default name is refused beside a + package's named manifests (PolicyEngine/chronicle#225), so a + registration never creates a manifest no sweep or package reads. + """ + name = str(manifest_filename).strip() if manifest_filename is not None else "" + if not is_manifest_filename(name): + raise HashOnlyRegistrationError( + f"--manifest must name {DEFAULT_MANIFEST_FILENAME} or " + f"manifest_.yaml inside the package directory, not " + f"{manifest_filename!r}." + ) + manifest_path = output / name + if name == DEFAULT_MANIFEST_FILENAME and not manifest_path.exists(): + siblings = [ + path.name for path in package_manifest_paths(output) if path.name != name + ] + if siblings: + raise HashOnlyRegistrationError( + f"{output} keeps {', '.join(siblings)} and no " + f"{DEFAULT_MANIFEST_FILENAME}; pass --manifest to name the " + "manifest this registration records into rather than creating " + f"{DEFAULT_MANIFEST_FILENAME} beside them." + ) + return manifest_path + + def _assert_manifest_identity( payload: Mapping[str, Any], manifest_path: Path, @@ -1210,6 +1410,7 @@ def _dedupe(values: Iterable[str]) -> list[str]: "ArtifactRegistrationReport", "CHRONICLE_ATTESTER", "DEFAULT_ACCESS", + "DEFAULT_MANIFEST_FILENAME", "DEFAULT_MANIFEST_KIND", "HASH_ONLY_HASH_SOURCES", "HASH_SOURCES", @@ -1229,14 +1430,18 @@ def _dedupe(values: Iterable[str]) -> list[str]: "entry_access", "filename_key", "has_file_entries", + "hash_only_registrations", "is_bare_filename", "is_hash_only", + "is_manifest_filename", "is_microdata_release", + "iter_directory_entries", "iter_file_specs", "iter_manifest_entries", "manifest_kind", "normalize_access", "normalize_hash_source", + "package_manifest_paths", "recorded_previous_r2", "recorded_r2", "records_r2_object", @@ -1249,5 +1454,6 @@ def _dedupe(values: Iterable[str]) -> list[str]: "strict_entry_access", "validate_file_entry", "validate_manifest_files", + "validate_package_directory", "vintage_key_forms", ] diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 390fa9e7..74b83fdc 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -39,8 +39,10 @@ ManifestKindError, MicrodataReleaseNotParseableError, entry_access, + hash_only_registrations, is_bare_filename, is_hash_only, + is_manifest_filename, is_microdata_release, resolve_vintage_key, ) @@ -902,8 +904,57 @@ def assert_parseable(self, year: int) -> dict[str, Any]: manifest = self.manifest_payload() spec = _year_mapping(manifest["files"], self.artifact_year or year) _assert_entry_bytes_readable(spec) + self._assert_no_sibling_hash_only_registration(spec) return spec + def _assert_no_sibling_hash_only_registration(self, spec: Any) -> None: + """Refuse a file another manifest in the directory registers hash-only. + + The boundary is the file in the package directory, not the manifest + that names it: a name or a digest registered ``licensed`` or + ``restricted`` in a sibling manifest is identity only, whichever + manifest this package reads through. A sibling Chronicle cannot read + is a refusal too, because the boundary cannot be decided without it. + """ + if not isinstance(spec, dict): + return + directory = files(self.resource_package).joinpath(self.resource_directory) + siblings: dict[str, dict[str, Any]] = {} + for item in directory.iterdir(): + if item.name == self.manifest or not is_manifest_filename(item.name): + continue + if not item.is_file(): + continue + try: + with item.open("r", encoding="utf-8") as file: + payload = yaml.safe_load(file) or {} + except (OSError, yaml.YAMLError) as exc: + raise ManifestAccessError( + f"{self.resource_directory}/{item.name} cannot be read " + f"({exc}), so whether it registers " + f"{spec.get('filename')!r} hash-only cannot be decided; " + "fix the manifest before parsing beside it." + ) from exc + if not isinstance(payload, dict): + raise ManifestAccessError( + f"{self.resource_directory}/{item.name} is not a YAML " + "mapping, so whether it registers " + f"{spec.get('filename')!r} hash-only cannot be decided." + ) + siblings[item.name] = payload + for name, key, entry in hash_only_registrations( + siblings, filename=spec.get("filename"), sha256=spec.get("sha256") + ): + raise ManifestAccessError( + f"{self.resource_directory}/{name} registers " + f"{entry.get('filename')!r} for {key!r} as " + f"access={entry.get('access')!r}: the same file, or the same " + f"bytes, as {spec.get('filename')!r}. A licensed or restricted " + "artifact is identity only, so no source package reads, caches, " + "fetches, or parses it through another manifest " + "(docs/adr-chronicle-raw-microdata-identity.md)." + ) + def _artifact_content( self, year: int, diff --git a/tests/test_chronicle_microdata_registration.py b/tests/test_chronicle_microdata_registration.py index ecef4e86..928e0054 100644 --- a/tests/test_chronicle_microdata_registration.py +++ b/tests/test_chronicle_microdata_registration.py @@ -1209,7 +1209,9 @@ def test_fetch_into_a_microdata_release_manifest_requires_the_evidence( "licence": "OGL-UK-3.0", "publisher": "Department for Work and Pensions", "vintage": "2023_24", - "expected_sha256": FIXTURE_SHA, + # Not the digest the manifest registers hash-only: those bytes are + # refused as the gated artifact whatever name they arrive under. + "expected_sha256": OTHER_SHA, "licence_evidence": {**EVIDENCE, "issuer": "DWP"}, "staging_dir": tmp_path / "staging", } diff --git a/tests/test_chronicle_package_directory.py b/tests/test_chronicle_package_directory.py new file mode 100644 index 00000000..c7197f32 --- /dev/null +++ b/tests/test_chronicle_package_directory.py @@ -0,0 +1,591 @@ +"""The byte boundary spans every manifest a package directory keeps. + +A publisher directory may keep several manifests -- ``manifest.yaml`` beside +``manifest_.yaml`` files -- and ``fetch-artifact --manifest`` (PR +#226) selects which one a fetch records into. The boundary is the file in the +directory, not the manifest: a name or a digest registered hash-only in any +manifest there must not be fetched, published, parsed, or reclassified +through another. An artifact may not be named like a manifest, a fetch may +not record into a manifest that identifies another package, and +``register-artifact`` addresses a named manifest exactly as a fetch does. +""" + +from __future__ import annotations + +import hashlib +from pathlib import Path + +import pytest +import yaml + +from chronicle.artifacts import ( + AmbiguousManifestError, + inventory_source_artifacts, + publish_source_artifacts, +) +from chronicle.harness import main as harness_main +from chronicle.registration import ( + ArtifactFilenameError, + HashOnlyRegistrationError, + ManifestAccessError, + is_manifest_filename, + register_hash_only_artifact, + validate_file_entry, + validate_package_directory, +) +from chronicle.source_package import SourceArtifactSpec +from tests.test_chronicle_microdata_registration import ( + ATTESTED, + EVIDENCE, + FIXTURE_SHA, + LICENSED_BYTES, + PUBLIC_BYTES, + PUBLIC_SHA, + _attested_entry, + _fetch_release, + _fetch_table, + _forbid_uploads, + _isolated_reader, + _record_uploads, + _refuse_read, + _register, + _serve, +) + + +REPO_ROOT = Path(__file__).resolve().parents[1] +LICENSED_SHA = hashlib.sha256(LICENSED_BYTES).hexdigest() + + +def _write(path: Path, payload: dict) -> bytes: + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(yaml.safe_dump(payload, sort_keys=False)) + return path.read_bytes() + + +def _table_manifest(**fields: object) -> dict: + payload = { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "publisher_table", + "files": {}, + } + payload.update(fields) + return payload + + +def _public_table_entry(filename: str, content: bytes) -> dict: + return { + "filename": filename, + "source_url": f"https://publisher.example/{filename}", + "access": "public", + "sha256": hashlib.sha256(content).hexdigest(), + "size_bytes": len(content), + } + + +def _hash_only_manifest(**entry: object) -> dict: + return { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "microdata_release", + "files": {2023: [_attested_entry(**entry)]}, + } + + +def _snapshot(package: Path) -> dict[str, bytes]: + return {path.name: path.read_bytes() for path in sorted(package.iterdir())} + + +# -------------------------------------------------------------------------- +# fetch-artifact +# -------------------------------------------------------------------------- + + +@pytest.mark.parametrize( + ("target", "sibling"), + [ + ("manifest_codebook.yaml", "manifest.yaml"), + ("manifest.yaml", "manifest_release.yaml"), + ("manifest_a.yaml", "manifest_b.yml"), + ], +) +def test_fetch_refuses_a_name_a_sibling_manifest_registers_hash_only( + tmp_path, monkeypatch, target, sibling +): + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + _write(package / sibling, _hash_only_manifest()) + # The target exists so the stray-default rule is not what refuses. + _write(package / target, _table_manifest()) + before = _snapshot(package) + reads = _refuse_read(monkeypatch) + _forbid_uploads(monkeypatch) + + with pytest.raises(ManifestAccessError, match=sibling) as refused: + _fetch_table( + tmp_path / "adult.tab", + package, + source_id="dwp", + package_id="dwp-frs-2023-24", + filename="adult.tab", + manifest_filename=target, + upload_r2=True, + ) + + assert "access='licensed'" in str(refused.value) + assert reads == [] + assert _snapshot(package) == before + + +def test_fetch_refuses_bytes_a_sibling_manifest_registers_hash_only( + tmp_path, monkeypatch +): + """The same bytes under another name are the same gated artifact.""" + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + _write(package / "manifest.yaml", _hash_only_manifest(sha256=LICENSED_SHA)) + _write(package / "manifest_tables.yaml", _table_manifest()) + before = _snapshot(package) + + # Pinned to the gated digest: refused before the read. + reads = _refuse_read(monkeypatch) + with pytest.raises(ManifestAccessError, match="manifest.yaml"): + _fetch_table( + tmp_path / "other.tab", + package, + source_id="dwp", + package_id="dwp-frs-2023-24", + filename="other.tab", + manifest_filename="manifest_tables.yaml", + expected_sha256=LICENSED_SHA, + ) + assert reads == [] + + # Not pinned: the served bytes turn out to be the gated ones, and are + # refused before anything is written or uploaded. + _serve(monkeypatch, LICENSED_BYTES) + _forbid_uploads(monkeypatch) + with pytest.raises(ManifestAccessError, match="manifest.yaml"): + _fetch_table( + tmp_path / "other.tab", + package, + source_id="dwp", + package_id="dwp-frs-2023-24", + filename="other.tab", + manifest_filename="manifest_tables.yaml", + upload_r2=True, + ) + assert _snapshot(package) == before + + +def test_fetch_refuses_bytes_its_own_manifest_registers_hash_only( + tmp_path, monkeypatch +): + """A release manifest never archives a digest it registers hash-only.""" + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + _write(package / "manifest.yaml", _hash_only_manifest(sha256=LICENSED_SHA)) + before = _snapshot(package) + _serve(monkeypatch, LICENSED_BYTES) + _forbid_uploads(monkeypatch) + + with pytest.raises(ManifestAccessError, match="adult.tab"): + _fetch_release( + package, + staging_dir=tmp_path / "staging", + filename="codebook.pdf", + content=LICENSED_BYTES, + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + vintage="2023_24", + licence="OGL-UK-3.0", + publisher="Department for Work and Pensions", + licence_evidence={**EVIDENCE, "issuer": "DWP"}, + upload_r2=True, + ) + + assert _snapshot(package) == before + assert not (tmp_path / "staging").exists() + + +def test_fetch_refuses_to_overwrite_a_file_a_sibling_manifest_records( + tmp_path, monkeypatch +): + """Two manifests may record one public file only as the same bytes.""" + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + original = b"the table both manifests record" + (package / "table.ods").parent.mkdir(parents=True) + (package / "table.ods").write_bytes(original) + _write( + package / "manifest_source_package.yaml", + _table_manifest(files={2024: _public_table_entry("table.ods", original)}), + ) + _write(package / "manifest.yaml", _table_manifest()) + before = _snapshot(package) + _serve(monkeypatch, b"a different table") + + with pytest.raises(ManifestAccessError, match="manifest_source_package.yaml"): + _fetch_table( + tmp_path / "table.ods", + package, + source_id="dwp", + package_id="dwp-frs-2023-24", + filename="table.ods", + ) + + assert _snapshot(package) == before + + # The same bytes are the same file: recorded twice, overwritten never. + _serve(monkeypatch, original) + report = _fetch_table( + tmp_path / "table.ods", + package, + source_id="dwp", + package_id="dwp-frs-2023-24", + filename="table.ods", + ) + assert report.valid + assert (package / "table.ods").read_bytes() == original + + +@pytest.mark.parametrize( + "name", + ["manifest.yaml", "MANIFEST.YAML", "manifest.yml", "manifest_tables.yaml"], +) +def test_an_artifact_may_not_be_named_like_a_manifest(tmp_path, monkeypatch, name): + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + reads = _refuse_read(monkeypatch) + + with pytest.raises(ArtifactFilenameError, match="manifest"): + _fetch_table( + tmp_path / name, + package, + source_id="dwp", + package_id="dwp-frs-2023-24", + filename=name, + ) + with pytest.raises(ArtifactFilenameError, match="manifest"): + _fetch_table( + tmp_path / "publisher" / name, + package, + source_id="dwp", + package_id="dwp-frs-2023-24", + ) + with pytest.raises(HashOnlyRegistrationError, match="manifest"): + _register(package, filename=name) + + assert reads == [] + assert not package.exists() + assert is_manifest_filename(name) + assert f"manifest_named_filename:{name}" in validate_file_entry( + _attested_entry(filename=name), + kind="microdata_release", + manifest={}, + local_file_exists=False, + ) + + +def test_fetch_refuses_a_manifest_that_identifies_another_package( + tmp_path, monkeypatch +): + package = tmp_path / "db" / "data" / "irs_soi" / "table_1_2" + original = _write( + package / "manifest.yaml", + { + "source_id": "irs_soi", + "package_id": "soi-table-1-2", + "kind": "publisher_table", + }, + ) + reads = _refuse_read(monkeypatch) + + with pytest.raises(ManifestAccessError, match="package_id='soi-table-1-2'"): + _fetch_table(tmp_path / "t.xlsx", package, package_id="soi-table-9") + with pytest.raises(ManifestAccessError, match="source_id='irs_soi'"): + _fetch_table(tmp_path / "t.xlsx", package, source_id="irs") + + assert reads == [] + assert (package / "manifest.yaml").read_bytes() == original + + +# -------------------------------------------------------------------------- +# publish-raw and inventory-artifacts +# -------------------------------------------------------------------------- + + +def _mixed_directory(tmp_path: Path) -> Path: + """manifest.yaml registers adult.tab hash-only; a sibling table manifest + records the same name, and another file with the same bytes, as public.""" + root = tmp_path / "data" + package = root / "dwp" / "frs_2023_24" + _write(package / "manifest.yaml", _hash_only_manifest(sha256=LICENSED_SHA)) + (package / "adult.tab").write_bytes(LICENSED_BYTES) + (package / "extract.tab").write_bytes(LICENSED_BYTES) + (package / "table.ods").write_bytes(b"a public table") + _write( + package / "manifest_tables.yaml", + _table_manifest( + files={ + 2023: _public_table_entry("adult.tab", LICENSED_BYTES), + 2022: _public_table_entry("extract.tab", LICENSED_BYTES), + 2021: _public_table_entry("table.ods", b"a public table"), + } + ), + ) + return root + + +def test_publish_raw_never_uploads_what_a_sibling_manifest_registers_hash_only( + tmp_path, monkeypatch +): + root = _mixed_directory(tmp_path) + package = root / "dwp" / "frs_2023_24" + before = _snapshot(package) + uploads = _record_uploads(monkeypatch) + + report = publish_source_artifacts(root, manifest_filename="manifest_tables.yaml") + + assert not report.valid + assert uploads == [] + assert report.entries == () + assert any( + error.startswith("filename_collision_across_manifests:adult.tab") + for error in report.errors + ) + assert any( + error.startswith(f"sha256_collision_across_manifests:{LICENSED_SHA}") + for error in report.errors + ) + assert _snapshot(package) == before + + +def test_inventory_reports_collisions_across_manifests(tmp_path): + root = _mixed_directory(tmp_path) + + report = inventory_source_artifacts(root, manifest_filename="manifest_tables.yaml") + + assert not report.valid + codes = {error.split(": ")[0] for error in report.errors} + assert "filename_collision_across_manifests:adult.tab" in codes + assert f"sha256_collision_across_manifests:{LICENSED_SHA}" in codes + # table.ods is recorded once and reported only for itself. + assert not any("table.ods" in code for code in codes) + + +def test_two_manifests_may_record_one_public_file_as_the_same_bytes(tmp_path): + """The tracked shape: manifest.yaml and manifest_.yaml both + record one publisher file with one digest (db/data/usda_snap/...).""" + root = tmp_path / "data" + package = root / "usda_snap" / "fy69_to_current" + content = b"snap zip" + (package).mkdir(parents=True) + (package / "snap.zip").write_bytes(content) + entry = _public_table_entry("snap.zip", content) + _write(package / "manifest.yaml", _table_manifest(files={2024: entry})) + _write(package / "manifest_fy2025.yaml", _table_manifest(files={2025: entry})) + + assert ( + validate_package_directory( + { + "manifest.yaml": yaml.safe_load( + (package / "manifest.yaml").read_text() + ), + "manifest_fy2025.yaml": yaml.safe_load( + (package / "manifest_fy2025.yaml").read_text() + ), + } + ) + == () + ) + assert inventory_source_artifacts(root).valid + assert inventory_source_artifacts( + root, manifest_filename="manifest_fy2025.yaml" + ).valid + + +def test_the_committed_tree_has_no_collisions_across_manifests(): + report = inventory_source_artifacts(REPO_ROOT / "db" / "data") + assert not [error for error in report.errors if "across_manifests" in error] + + +# -------------------------------------------------------------------------- +# register-artifact +# -------------------------------------------------------------------------- + + +def test_register_refuses_an_identity_a_sibling_manifest_holds_public(tmp_path): + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + archived = _public_table_entry("adult.tab", PUBLIC_BYTES) + archived["storage"] = { + "r2": { + "provider": "r2", + "bucket": "ledger-raw", + "key": f"raw/uk/dwp/dwp-frs-2023-24/2023/{PUBLIC_SHA}/adult.tab", + "uri": f"r2://ledger-raw/raw/uk/dwp/dwp-frs-2023-24/2023/{PUBLIC_SHA}/adult.tab", + } + } + _write( + package / "manifest_tables.yaml", + _table_manifest( + files={ + 2023: archived, + 2022: _public_table_entry("extract.tab", b"unarchived public bytes"), + } + ), + ) + _write( + package / "manifest.yaml", {"source_id": "dwp", "package_id": "dwp-frs-2023-24"} + ) + before = _snapshot(package) + + # By name: the sibling archived adult.tab. + with pytest.raises(HashOnlyRegistrationError, match="manifest_tables.yaml"): + _register(package) + # By digest: the sibling archived these bytes under another name. + with pytest.raises(HashOnlyRegistrationError, match="manifest_tables.yaml"): + _register(package, filename="other.tab", sha256=PUBLIC_SHA) + # A public record with no object is still an explicit class change. + with pytest.raises(HashOnlyRegistrationError, match="manifest_tables.yaml"): + _register(package, filename="extract.tab") + + assert _snapshot(package) == before + + +def test_register_refuses_a_stray_default_manifest_beside_named_ones(tmp_path): + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + _write(package / "manifest_tables.yaml", _table_manifest()) + before = _snapshot(package) + + with pytest.raises(HashOnlyRegistrationError, match="--manifest"): + _register(package) + + assert _snapshot(package) == before + + +def test_register_targets_the_named_manifest(tmp_path, capsys): + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + _write(package / "manifest_tables.yaml", _table_manifest()) + + report = _register(package, manifest_filename="manifest_release.yaml") + + assert report.valid + assert report.manifest_path.endswith("manifest_release.yaml") + written = yaml.safe_load((package / "manifest_release.yaml").read_text()) + assert written["kind"] == "microdata_release" + assert [entry["filename"] for entry in written["files"][2023]] == ["adult.tab"] + assert not (package / "manifest.yaml").exists() + + with pytest.raises(HashOnlyRegistrationError, match="manifest"): + _register(package, manifest_filename="../manifest.yaml") + with pytest.raises(HashOnlyRegistrationError, match="manifest"): + _register(package, manifest_filename="adult.tab") + + argv = [ + "register-artifact", + "--source-id", + "dwp", + "--package-id", + "dwp-frs-2023-24", + "--year", + "2023", + "--out-dir", + str(package), + "--manifest", + "manifest_release.yaml", + "--filename", + "benefits.tab", + "--sha256", + FIXTURE_SHA, + "--vintage", + "2023_24", + "--licence", + "UK Data Service End User Licence", + "--access", + "licensed", + "--doi", + "10.5255/UKDA-SN-9367-2", + "--hash-source", + "consumer_attested", + "--attested-by", + ATTESTED["attested_by"], + "--attestation-evidence", + ATTESTED["attestation_evidence"], + "--verified-at", + ATTESTED["verified_at"], + ] + assert harness_main(argv) == 0 + capsys.readouterr() + written = yaml.safe_load((package / "manifest_release.yaml").read_text()) + assert [entry["filename"] for entry in written["files"][2023]] == [ + "adult.tab", + "benefits.tab", + ] + + +# -------------------------------------------------------------------------- +# The source-package byte reader +# -------------------------------------------------------------------------- + + +def test_byte_reader_refuses_a_file_a_sibling_manifest_registers_hash_only( + tmp_path, monkeypatch +): + import sys + import uuid + + _isolated_reader(tmp_path, monkeypatch) + package_name = f"chronicle_test_{uuid.uuid4().hex}" + resource_dir = tmp_path / "pkgroot" / package_name / "data" / "dwp" / "frs" + resource_dir.mkdir(parents=True) + _write( + resource_dir / "manifest.yaml", + _table_manifest(files={2023: _public_table_entry("adult.tab", LICENSED_BYTES)}), + ) + _write( + resource_dir / "manifest_release.yaml", + _hash_only_manifest(sha256=LICENSED_SHA), + ) + (resource_dir / "adult.tab").write_bytes(LICENSED_BYTES) + monkeypatch.syspath_prepend(str(tmp_path / "pkgroot")) + monkeypatch.delitem(sys.modules, package_name, raising=False) + spec = SourceArtifactSpec( + source_name="dwp", + source_table="Family Resources Survey", + resource_package=package_name, + resource_directory="data/dwp/frs", + manifest="manifest.yaml", + vintage="2023_24", + extracted_at="2026-09-02", + extraction_method="none", + parser="delimited_text_full_rows", + delimiter="\t", + artifact_year=2023, + ) + + with pytest.raises(ManifestAccessError, match="manifest_release.yaml"): + spec.assert_parseable(2023) + with pytest.raises(ManifestAccessError, match="manifest_release.yaml"): + spec.build_source_rows(2023) + + +def test_the_stray_default_manifest_rule_reaches_register_before_any_write(tmp_path): + package = tmp_path / "db" / "data" / "irs_soi" / "ira_contributions" + _write(package / "manifest_roth_source_package.yaml", _table_manifest()) + _write(package / "manifest_traditional_source_package.yaml", _table_manifest()) + before = _snapshot(package) + + with pytest.raises((HashOnlyRegistrationError, AmbiguousManifestError)): + register_hash_only_artifact( + source_id="dwp", + package_id="dwp-frs-2023-24", + year=2023, + output_dir=package, + filename="adult.tab", + sha256=FIXTURE_SHA, + licence="UK Data Service End User Licence", + access="licensed", + vintage="2023_24", + doi="10.5255/UKDA-SN-9367-2", + **ATTESTED, + ) + + assert _snapshot(package) == before From 7f9bfe6edefaaf2e0550c2194afc35f905f63ee6 Mon Sep 17 00:00:00 2001 From: Max Ghenis Date: Fri, 4 Sep 2026 06:55:55 -0400 Subject: [PATCH 42/42] Read manifests strictly and validate every entry field the byte boundary relies on A manifest is the record the byte boundary is decided from, so the reader now refuses what it cannot represent faithfully and the commands validate what they used to act on silently: - StrictManifestLoader refuses a document with duplicate mapping keys (PyYAML keeps the last of two equal keys, so `files:` recorded twice, or a vintage as `2023` and again as `2_023`, would read as one entry and the shadowed entry would be dropped by the next write). - The filename comparison key is NFC-normalised as well as case-folded, so the composed and decomposed spellings of an accented name are one file, as they are on the filesystems these commands run on. - validate_file_entry reports a nameless or malformed entry, a misspelled field (`Access` is not `access`, and an entry whose access class sits under the wrong key is not public by omission), a storage block or a chronicle_fetch attestation on a hash-only entry, a consumer attestation signed by Chronicle, and evidence whose URL has no host; a public release may hold one filename under different vintages but not under two digests in one vintage. - register-artifact validates every existing entry of the manifest before recording into it: an invalid entry is never replaced or reclassified in passing, and the refusal names the same codes inventory-artifacts reports. Covered by tests/test_chronicle_manifest_reading.py; the lane that authored this (fix-227-r2, resumed after lane limits) left it uncommitted, tests green, and it is committed here unchanged. Co-Authored-By: Claude Fable 5.1 --- chronicle/artifacts.py | 3 +- chronicle/licences.py | 18 +- chronicle/registration.py | 164 ++++++++- chronicle/source_package.py | 18 +- tests/test_chronicle_manifest_reading.py | 438 +++++++++++++++++++++++ 5 files changed, 622 insertions(+), 19 deletions(-) create mode 100644 tests/test_chronicle_manifest_reading.py diff --git a/chronicle/artifacts.py b/chronicle/artifacts.py index 70fc3249..320612b4 100644 --- a/chronicle/artifacts.py +++ b/chronicle/artifacts.py @@ -54,6 +54,7 @@ iter_directory_entries, iter_file_specs, iter_manifest_entries, + load_manifest_document, manifest_kind as normalize_manifest_kind, normalize_access, package_manifest_paths, @@ -1791,7 +1792,7 @@ def _read_manifest(manifest_path: Path) -> dict[str, Any]: if not manifest_path.exists(): return {} try: - payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) + payload = load_manifest_document(manifest_path.read_text(encoding="utf-8")) except yaml.YAMLError as exc: raise MalformedManifestError( f"{manifest_path} is not valid YAML: {exc}" diff --git a/chronicle/licences.py b/chronicle/licences.py index a3679ba7..c14c76c0 100644 --- a/chronicle/licences.py +++ b/chronicle/licences.py @@ -18,11 +18,13 @@ from dataclasses import dataclass from types import MappingProxyType from typing import Any +from urllib.parse import urlparse __all__ = [ "LICENCE_EVIDENCE_FIELDS", "REDISTRIBUTABLE_LICENCES", "RedistributableLicence", + "is_durable_url", "is_redistributable_licence", "licence_evidence_errors", ] @@ -132,6 +134,20 @@ def licence_evidence_errors( errors.append(f"licence_not_redistributable:{evidence['licence'].strip()}") if evidence["sha256"].strip() != str(sha256 or "").strip(): errors.append("licence_evidence_sha256_mismatch") - if not evidence["url"].strip().startswith(("http://", "https://")): + if not is_durable_url(evidence["url"]): errors.append("licence_evidence_url_not_durable") return errors + + +def is_durable_url(value: Any) -> bool: + """Whether ``value`` is an http(s) URL with a host and no whitespace. + + A bare scheme, a padded string, or a URL with a space is not a location + a reviewer can follow back to the publisher's statement. + """ + if not isinstance(value, str) or value != value.strip() or not value: + return False + if any(character.isspace() for character in value): + return False + parsed = urlparse(value) + return parsed.scheme in ("http", "https") and bool(parsed.netloc) diff --git a/chronicle/registration.py b/chronicle/registration.py index 87efca40..9172f9f4 100644 --- a/chronicle/registration.py +++ b/chronicle/registration.py @@ -47,6 +47,7 @@ from pathlib import Path import re from typing import Any +import unicodedata import yaml @@ -118,6 +119,63 @@ ) +#: Every field a manifest file entry may carry, for the misspelling check. +_ENTRY_FIELDS: frozenset[str] = frozenset( + {*_REGISTRATION_FIELD_ORDER, "storage", "size_bytes", "source_url"} +) + + +class StrictManifestLoader(yaml.SafeLoader): + """A YAML loader that refuses a mapping with duplicate keys. + + PyYAML keeps the last of two equal keys, so ``files:`` recorded twice, or + a vintage recorded as ``2023`` and again as ``2_023`` (the same integer), + would read as one entry and the shadowed entry would be dropped by the + next write. A manifest is the record the byte boundary is decided from, + so a document the loader cannot represent faithfully is malformed. + """ + + def construct_mapping(self, node: Any, deep: bool = False) -> dict[Any, Any]: + if not isinstance(node, yaml.MappingNode): + raise yaml.constructor.ConstructorError( + None, + None, + f"expected a mapping node, but found {node.id}", + node.start_mark, + ) + self.flatten_mapping(node) + mapping: dict[Any, Any] = {} + for key_node, value_node in node.value: + key = self.construct_object(key_node, deep=deep) + try: + hash(key) + except TypeError as exc: + raise yaml.constructor.ConstructorError( + "while constructing a mapping", + node.start_mark, + f"found unhashable key ({exc})", + key_node.start_mark, + ) from exc + if key in mapping: + raise yaml.constructor.ConstructorError( + "while constructing a mapping", + node.start_mark, + f"found duplicate key {key!r}", + key_node.start_mark, + ) + mapping[key] = self.construct_object(value_node, deep=deep) + return mapping + + +def load_manifest_document(text: str) -> Any: + """Parse a manifest document, refusing duplicate keys. + + Raises :class:`yaml.YAMLError` (a ``ConstructorError`` naming the + duplicate key) for a document YAML would otherwise silently collapse. + """ + return yaml.load(text, Loader=StrictManifestLoader) # noqa: S506 + + class ManifestAccessError(ValueError): """Raised when a manifest declares an unusable access class or kind.""" @@ -333,12 +391,13 @@ def bare_filename(value: Any, *, what: str = "filename") -> str: def filename_key(value: Any) -> str: """Return the comparison key for a filename. - Case-folded, because the package directories these commands run in are as - often as not on a case-insensitive filesystem, where ``ADULT.TAB`` and - ``adult.tab`` are one file. Treating them as one artifact path is the safe - rule everywhere. + Case-folded and Unicode-normalised (NFC), because the package directories + these commands run in are as often as not on a filesystem that folds + both: ``ADULT.TAB`` and ``adult.tab`` are one file, and so are the + composed and decomposed spellings of an accented name. Treating them as + one artifact path is the safe rule everywhere. """ - return Path(str(value)).name.casefold() + return unicodedata.normalize("NFC", Path(str(value)).name).casefold() #: The names a package directory's manifests may carry: ``manifest.yaml`` or @@ -588,6 +647,7 @@ def validate_manifest_files(manifest: Mapping[str, Any] | None) -> tuple[str, .. if other != key and other in files: errors.append(f"duplicate_vintage_key:{key}") + release = manifest.get("kind") == MICRODATA_RELEASE_KIND # Same file, one package directory: group every entry by its resolved name. by_name: dict[str, list[tuple[Any, Mapping[str, Any]]]] = {} for key, _index, entry in iter_manifest_entries(manifest): @@ -637,6 +697,15 @@ def validate_manifest_files(manifest: Mapping[str, Any] | None) -> tuple[str, .. # Several vintages, or an explicit reissue, may register the same # filename with different bytes: no file exists to collide. continue + if release: + # A public release's bytes are staged content-addressed outside + # the tree, so one filename may hold different bytes under + # different vintages; under one vintage a revision is recorded in + # storage.previous_r2, not as a second entry. + for digests in seen.values(): + if len(digests) > 1: + errors.append(f"filename_collision:{name}") + continue digests = {_text(entry.get("sha256")) or "" for _key, entry in entries} if len(digests) > 1: # Public entries share one path in the tree and one current object @@ -663,13 +732,25 @@ def validate_file_entry( if isinstance(spec, ListSpecRejected): return ("list_file_spec_requires_microdata_release_kind",) if not isinstance(spec, Mapping): - return () + return ("malformed_file_spec",) errors: list[str] = [] + for field in spec: + if ( + isinstance(field, str) + and field not in _ENTRY_FIELDS + and field.strip().casefold() in _ENTRY_FIELDS + ): + # A field the writer meant but the reader would ignore: ``Access`` + # is not ``access``, and an entry whose access class sits under + # the wrong key is not public by omission. + errors.append(f"misspelled_field:{field}") filename = spec.get("filename") - if filename is not None and not is_bare_filename(filename): + if not _text(filename): + errors.append("missing_filename") + elif not is_bare_filename(filename): errors.append(f"non_canonical_filename:{filename}") - elif filename is not None and is_manifest_filename(filename): + elif is_manifest_filename(filename): errors.append(f"manifest_named_filename:{filename}") declared_access = spec.get("access") @@ -733,6 +814,16 @@ def _hash_only_entry_errors( errors.append("r2_location_for_hash_only_entry") if recorded_previous_r2(spec): errors.append("r2_history_for_hash_only_entry") + storage = spec.get("storage") + if storage is not None and storage != {}: + # No Chronicle store holds these bytes, so there is nothing a storage + # block could truthfully record, whatever its shape. + errors.append("storage_for_hash_only_entry") + hash_source = _text(spec.get("hash_source")) + if hash_source in HASH_SOURCES and hash_source not in HASH_ONLY_HASH_SOURCES: + # Chronicle never fetched a hash-only entry's bytes: its checksum is + # always the consumer's. + errors.append(f"hash_source_not_allowed_for_hash_only_entry:{hash_source}") return errors @@ -792,9 +883,17 @@ def _attestation_errors(spec: Mapping[str, Any]) -> list[str]: errors.extend(_pinned_from_errors(spec.get("pinned_from"))) if verified_at: errors.append("verified_at_forbidden_for_consumer_pin") + if hash_source != HASH_SOURCE_CHRONICLE_FETCH and _is_chronicle(attested_by): + # A consumer's checksum is attested by the consumer; Chronicle only + # attests what it fetched and hashed itself. + errors.append("attested_by_chronicle_for_consumer_hash_source") return errors +def _is_chronicle(attester: str | None) -> bool: + return bool(attester) and attester.strip().casefold() == CHRONICLE_ATTESTER + + def _pinned_from_errors(pinned_from: Any) -> list[str]: if pinned_from is None: return ["missing_pinned_from"] @@ -1071,12 +1170,30 @@ def register_hash_only_artifact( f"{type(files).__name__}. Chronicle will not write into a manifest " "it cannot read." ) - manifest_errors = validate_manifest_files(payload) + manifest_errors = list(validate_manifest_files(payload)) + for existing_key, _index, existing in iter_manifest_entries(payload): + existing_name = ( + existing.get("filename") if isinstance(existing, Mapping) else None + ) + exists = ( + is_bare_filename(existing_name) and (output / str(existing_name)).exists() + ) + manifest_errors.extend( + f"{existing_key!r}/{existing_name}: {code}" + for code in validate_file_entry( + existing, + kind=existing_kind, + manifest=payload, + local_file_exists=exists, + ) + ) if manifest_errors: + # An invalid entry is never replaced or reclassified in passing: the + # registration would carry the defect forward or conceal it. raise HashOnlyRegistrationError( - f"{manifest_path} is not a valid manifest: " - f"{', '.join(manifest_errors)}. Fix it by hand before registering " - "into it." + f"{manifest_path} is not a valid {existing_kind} manifest: " + f"{'; '.join(manifest_errors)}. Fix it by hand before registering " + "into it; inventory-artifacts reports the same codes." ) _assert_no_archived_identity( payload, manifest_path, artifact_name, access_class, sha256=checksum @@ -1202,6 +1319,13 @@ def _hash_only_attestation( f"A {source} registration must name the consumer that attests the " "checksum; pass --attested-by." ) + if _is_chronicle(attester): + raise HashOnlyRegistrationError( + f"A {source} registration is attested by the consumer whose pin it " + f"transcribes, never by {CHRONICLE_ATTESTER!r}: Chronicle holds no " + "bytes for it and verified nothing. Pass --attested-by with the " + "consumer's name." + ) fields: dict[str, Any] = {"hash_source": source, "attested_by": attester} if source == HASH_SOURCE_CONSUMER_ATTESTED: if not _text(attestation_evidence): @@ -1372,10 +1496,20 @@ def _existing_entries(files: Any, key: Any) -> list[Any]: def _load_manifest(manifest_path: Path) -> dict[str, Any]: - """Load a manifest mapping, or an empty mapping when absent.""" + """Load a manifest mapping, or an empty mapping when absent. + + Read strictly: a document with duplicate keys, or one that is not a + mapping, is a refusal rather than something to record into. + """ if not manifest_path.exists(): return {} - payload = yaml.safe_load(manifest_path.read_text(encoding="utf-8")) or {} + try: + payload = load_manifest_document(manifest_path.read_text(encoding="utf-8")) + except yaml.YAMLError as exc: + raise HashOnlyRegistrationError( + f"{manifest_path} is not valid YAML: {exc}" + ) from exc + payload = payload or {} if not isinstance(payload, dict): raise HashOnlyRegistrationError(f"Manifest must be a mapping: {manifest_path}") return payload @@ -1420,6 +1554,8 @@ def _dedupe(values: Iterable[str]) -> list[str]: "HashOnlyRegistrationError", "ListSpecRejected", "MANIFEST_KINDS", + "StrictManifestLoader", + "load_manifest_document", "MICRODATA_RELEASE_KIND", "ManifestAccessError", "ManifestKindError", diff --git a/chronicle/source_package.py b/chronicle/source_package.py index 74b83fdc..e365f4ab 100644 --- a/chronicle/source_package.py +++ b/chronicle/source_package.py @@ -44,6 +44,7 @@ is_hash_only, is_manifest_filename, is_microdata_release, + load_manifest_document, resolve_vintage_key, ) from chronicle.sources.cells import ( @@ -870,9 +871,20 @@ def manifest_resource(self) -> Any: ) def manifest_payload(self) -> dict[str, Any]: - """Load the artifact manifest this package spec points at.""" + """Load the artifact manifest this package spec points at, strictly. + + A document with duplicate keys is refused rather than read through + whichever entry YAML kept. + """ with self.manifest_resource().open("r", encoding="utf-8") as file: - return yaml.safe_load(file) or {} + text = file.read() + try: + payload = load_manifest_document(text) + except yaml.YAMLError as exc: + raise ValueError( + f"{self.resource_directory}/{self.manifest} is not valid YAML: {exc}" + ) from exc + return payload or {} def assert_parseable_manifest(self) -> None: """Refuse to parse a manifest that registers a microdata release. @@ -927,7 +939,7 @@ def _assert_no_sibling_hash_only_registration(self, spec: Any) -> None: continue try: with item.open("r", encoding="utf-8") as file: - payload = yaml.safe_load(file) or {} + payload = load_manifest_document(file.read()) or {} except (OSError, yaml.YAMLError) as exc: raise ManifestAccessError( f"{self.resource_directory}/{item.name} cannot be read " diff --git a/tests/test_chronicle_manifest_reading.py b/tests/test_chronicle_manifest_reading.py new file mode 100644 index 00000000..6a1f56ce --- /dev/null +++ b/tests/test_chronicle_manifest_reading.py @@ -0,0 +1,438 @@ +"""Manifests are read strictly and every entry field the sweeps rely on is validated. + +A manifest is the record the byte boundary is decided from, so the reader +refuses what it cannot represent faithfully (a document whose duplicate keys +YAML would silently collapse), the comparison key for filenames covers every +spelling a filesystem folds together (case and Unicode normalisation), and +``validate_file_entry`` reports the shapes the commands used to act on +silently: a nameless or malformed entry, a misspelled field, a hash-only +entry carrying a storage block or a ``chronicle_fetch`` attestation, a +consumer attestation signed by Chronicle, and evidence whose URL has no host. +""" + +from __future__ import annotations + +import hashlib +import unicodedata +from pathlib import Path + +import pytest +import yaml + +from chronicle.artifacts import ( + MalformedManifestError, + _read_manifest, + inventory_source_artifacts, + publish_source_artifacts, +) +from chronicle.licences import licence_evidence_errors +from chronicle.registration import ( + HashOnlyRegistrationError, + ManifestAccessError, + filename_key, + validate_file_entry, + validate_manifest_files, +) +from chronicle.source_package import SourceArtifactSpec +from tests.test_chronicle_microdata_registration import ( + ATTESTED, + EVIDENCE, + FIXTURE_SHA, + LICENSED_BYTES, + OTHER_SHA, + PINNED, + PUBLIC_SHA, + _attested_entry, + _fetch_table, + _forbid_uploads, + _public_release_entry, + _refuse_read, + _register, +) + + +NFC_NAME = unicodedata.normalize("NFC", "adúlt.tab") +NFD_NAME = unicodedata.normalize("NFD", "adúlt.tab") +LICENSED_SHA = hashlib.sha256(LICENSED_BYTES).hexdigest() + + +def _write(path: Path, text: str) -> bytes: + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(text) + return path.read_bytes() + + +def _release_manifest(**fields: object) -> dict: + payload = { + "source_id": "dwp", + "package_id": "dwp-frs-2023-24", + "kind": "microdata_release", + "files": {}, + } + payload.update(fields) + return payload + + +# -------------------------------------------------------------------------- +# Filename keys fold Unicode normalisation as well as case +# -------------------------------------------------------------------------- + + +def test_filename_key_folds_unicode_normalisation(): + assert NFC_NAME != NFD_NAME + assert filename_key(NFC_NAME) == filename_key(NFD_NAME) + assert filename_key(NFD_NAME.upper()) == filename_key(NFC_NAME) + + +def test_a_normalisation_alias_is_a_filename_collision(): + manifest = _release_manifest( + files={ + 2023: [ + _attested_entry(filename=NFC_NAME), + {**_public_release_entry(filename=NFD_NAME)}, + ] + } + ) + codes = validate_manifest_files(manifest) + assert any(code.startswith("filename_collision:") for code in codes), codes + + +def test_fetch_refuses_a_normalisation_alias_of_a_hash_only_name(tmp_path, monkeypatch): + package = tmp_path / "db" / "data" / "dwp" / "frs_2023_24" + _register(package, filename=NFC_NAME) + original = (package / "manifest.yaml").read_bytes() + reads = _refuse_read(monkeypatch) + _forbid_uploads(monkeypatch) + + with pytest.raises(ManifestAccessError, match="access='licensed'"): + _fetch_table( + tmp_path / NFD_NAME, + package, + source_id="dwp", + package_id="dwp-frs-2023-24", + filename=NFD_NAME, + upload_r2=True, + ) + + assert reads == [] + assert (package / "manifest.yaml").read_bytes() == original + assert sorted(path.name for path in package.iterdir()) == ["manifest.yaml"] + + +# -------------------------------------------------------------------------- +# Entry validation vocabulary +# -------------------------------------------------------------------------- + + +def test_a_nameless_or_malformed_entry_is_an_error_everywhere(): + nameless = _attested_entry(filename=None) + assert "missing_filename" in validate_file_entry( + nameless, kind="microdata_release", manifest={}, local_file_exists=False + ) + assert "missing_filename" in validate_file_entry( + {"source_url": "x"}, + kind="publisher_table", + manifest={}, + local_file_exists=False, + ) + for malformed in ("adult.tab", 3, None, ["adult.tab"]): + assert validate_file_entry( + malformed, kind="publisher_table", manifest={}, local_file_exists=False + ) == ("malformed_file_spec",) + + +@pytest.mark.parametrize( + ("field", "code"), + [ + ("Access", "misspelled_field:Access"), + ("ACCESS", "misspelled_field:ACCESS"), + ("Sha256", "misspelled_field:Sha256"), + ("Licence", "misspelled_field:Licence"), + ("Storage", "misspelled_field:Storage"), + ("Hash_Source", "misspelled_field:Hash_Source"), + (" access", "misspelled_field: access"), + ], +) +def test_a_misspelled_field_is_reported_not_ignored(field, code): + entry = _attested_entry() + entry[field] = entry.pop("access") if field.strip().lower() == "access" else "x" + codes = validate_file_entry( + entry, kind="microdata_release", manifest={}, local_file_exists=False + ) + assert code in codes, codes + + +@pytest.mark.parametrize( + "storage", + [ + {"r2": "r2://ledger-raw/raw/x"}, + {"previous_r2": "r2://ledger-raw/raw/x"}, + {"r2": {"uri": "r2://ledger-raw/raw/x"}}, + {"previous_r2": [{"uri": "r2://ledger-raw/raw/x"}]}, + {"anything": 1}, + "r2://ledger-raw/raw/x", + ["r2://ledger-raw/raw/x"], + ], +) +def test_a_hash_only_entry_may_carry_no_storage_block_of_any_shape(storage): + codes = validate_file_entry( + _attested_entry(storage=storage), + kind="microdata_release", + manifest={}, + local_file_exists=False, + ) + assert "storage_for_hash_only_entry" in codes, codes + + +def test_a_hash_only_entry_is_never_a_chronicle_fetch(): + entry = _attested_entry( + hash_source="chronicle_fetch", + attested_by="chronicle", + verified_at="2026-09-03", + attestation_evidence=None, + ) + codes = validate_file_entry( + entry, kind="microdata_release", manifest={}, local_file_exists=False + ) + assert "hash_source_not_allowed_for_hash_only_entry:chronicle_fetch" in codes + + +@pytest.mark.parametrize("attestation", [ATTESTED, PINNED]) +def test_a_consumer_attestation_is_not_signed_by_chronicle(attestation): + entry = _attested_entry(**{**attestation, "attested_by": "chronicle"}) + codes = validate_file_entry( + entry, kind="microdata_release", manifest={}, local_file_exists=False + ) + assert "attested_by_chronicle_for_consumer_hash_source" in codes, codes + + +def test_register_refuses_a_consumer_attestation_signed_by_chronicle(tmp_path): + package = tmp_path / "pkg" + with pytest.raises(HashOnlyRegistrationError, match="chronicle"): + _register(package, attested_by="chronicle") + with pytest.raises(HashOnlyRegistrationError, match="chronicle"): + _register(package, **{**PINNED, "attested_by": "Chronicle"}) + assert not package.exists() + + +def test_release_public_digests_are_keyed_per_vintage(): + """Public release bytes are staged content-addressed outside the tree, so + one filename may hold different bytes under different vintages; under one + vintage a revision is history, not a second entry.""" + per_vintage = _release_manifest( + files={ + 2022: [_public_release_entry()], + 2023: [ + _public_release_entry( + sha256=OTHER_SHA, + vintage="2023", + licence_evidence={ + **EVIDENCE, + "licence": "US-Government-Work", + "sha256": OTHER_SHA, + }, + ) + ], + } + ) + assert validate_manifest_files(per_vintage) == () + + same_vintage = _release_manifest( + files={ + 2022: [ + _public_release_entry(), + _public_release_entry( + sha256=OTHER_SHA, + licence_evidence={ + **EVIDENCE, + "licence": "US-Government-Work", + "sha256": OTHER_SHA, + }, + ), + ] + } + ) + assert "filename_collision:csv_hus.zip" in validate_manifest_files(same_vintage) + + table = { + "source_id": "irs_soi", + "package_id": "soi-table-1-2", + "kind": "publisher_table", + "files": { + 2022: {"filename": "t.xlsx", "sha256": PUBLIC_SHA}, + 2023: {"filename": "t.xlsx", "sha256": OTHER_SHA}, + }, + } + assert "filename_collision:t.xlsx" in validate_manifest_files(table) + + +@pytest.mark.parametrize( + "url", + [ + "https://", + "http://", + "https:// evidence.example", + " https://evidence.example", + "https://evidence.example/a b", + "ftp://evidence.example/x", + "evidence.example/x", + ], +) +def test_an_evidence_url_needs_a_scheme_and_a_host_and_no_whitespace(url): + evidence = { + **EVIDENCE, + "licence": "US-Government-Work", + "sha256": PUBLIC_SHA, + "url": url, + } + assert "licence_evidence_url_not_durable" in licence_evidence_errors( + evidence, licence="US-Government-Work", sha256=PUBLIC_SHA + ) + + +def test_an_evidence_url_with_a_host_is_durable(): + evidence = {**EVIDENCE, "licence": "US-Government-Work", "sha256": PUBLIC_SHA} + assert ( + licence_evidence_errors( + evidence, licence="US-Government-Work", sha256=PUBLIC_SHA + ) + == [] + ) + + +# -------------------------------------------------------------------------- +# register-artifact validates the manifest it records into +# -------------------------------------------------------------------------- + + +@pytest.mark.parametrize( + "existing", + [ + {"filename": "other.tab", "access": "Public", "sha256": OTHER_SHA}, + {"filename": "adult.tab", "access": "open", "sha256": FIXTURE_SHA}, + {**_attested_entry(filename="other.tab", storage="r2://ledger-raw/x")}, + {**_attested_entry(filename="other.tab", hash_source="chronicle_fetch")}, + ], + ids=[ + "title-case-access", + "unknown-access-same-name", + "string-storage", + "chronicle-fetch", + ], +) +def test_register_refuses_an_invalid_existing_entry_instead_of_replacing_it( + tmp_path, existing +): + package = tmp_path / "pkg" + original = _write( + package / "manifest.yaml", + yaml.safe_dump(_release_manifest(files={2023: [existing]}), sort_keys=False), + ) + + with pytest.raises(HashOnlyRegistrationError, match="not a valid"): + _register(package) + + assert (package / "manifest.yaml").read_bytes() == original + + +def test_register_reports_unreadable_yaml_as_a_refusal(tmp_path): + package = tmp_path / "pkg" + original = _write(package / "manifest.yaml", "files: [\n") + + with pytest.raises(HashOnlyRegistrationError, match="not valid YAML"): + _register(package) + + assert (package / "manifest.yaml").read_bytes() == original + + +# -------------------------------------------------------------------------- +# Duplicate keys are a malformed manifest, never silently collapsed +# -------------------------------------------------------------------------- + + +DUPLICATE_DOCUMENTS = { + "duplicate-vintage": ( + "source_id: irs_soi\npackage_id: soi-table-1-2\nkind: publisher_table\n" + "files:\n 2023:\n filename: a.xlsx\n 2023:\n filename: b.xlsx\n" + ), + "int-alias-vintage": ( + "source_id: irs_soi\npackage_id: soi-table-1-2\nkind: publisher_table\n" + "files:\n 2023:\n filename: a.xlsx\n 2_023:\n filename: b.xlsx\n" + ), + "float-alias-vintage": ( + "source_id: irs_soi\npackage_id: soi-table-1-2\nkind: publisher_table\n" + "files:\n 2023:\n filename: a.xlsx\n 2023.0:\n filename: b.xlsx\n" + ), + "duplicate-files-block": ( + "source_id: irs_soi\npackage_id: soi-table-1-2\nkind: publisher_table\n" + "files:\n 2023:\n filename: a.xlsx\nfiles:\n 2024:\n filename: b.xlsx\n" + ), + "duplicate-entry-field": ( + "source_id: irs_soi\npackage_id: soi-table-1-2\nkind: publisher_table\n" + "files:\n 2023:\n filename: a.xlsx\n access: public\n access: licensed\n" + ), +} + + +@pytest.mark.parametrize( + "document", DUPLICATE_DOCUMENTS.values(), ids=DUPLICATE_DOCUMENTS +) +def test_a_manifest_with_duplicate_keys_is_refused_by_every_reader( + tmp_path, monkeypatch, document +): + package = tmp_path / "db" / "data" / "irs_soi" / "table_1_2" + original = _write(package / "manifest.yaml", document) + (package / "a.xlsx").write_bytes(b"a") + (package / "b.xlsx").write_bytes(b"b") + + with pytest.raises(MalformedManifestError, match="duplicate key"): + _read_manifest(package / "manifest.yaml") + + reads = _refuse_read(monkeypatch) + _forbid_uploads(monkeypatch) + with pytest.raises(MalformedManifestError, match="duplicate key"): + _fetch_table(tmp_path / "c.xlsx", package, year=2025, upload_r2=True) + assert reads == [] + + inventory = inventory_source_artifacts(tmp_path / "db" / "data") + assert not inventory.valid + assert inventory.entries == () + assert any("duplicate key" in error for error in inventory.errors) + + published = publish_source_artifacts(tmp_path / "db" / "data") + assert not published.valid + assert published.entries == () + assert any("duplicate key" in error for error in published.errors) + + with pytest.raises(HashOnlyRegistrationError, match="duplicate key"): + _register(package, source_id="irs_soi", package_id="soi-table-1-2") + + assert (package / "manifest.yaml").read_bytes() == original + + +def test_the_byte_reader_refuses_a_manifest_with_duplicate_keys(tmp_path, monkeypatch): + import sys + import uuid + + package_name = f"chronicle_test_{uuid.uuid4().hex}" + resource_dir = tmp_path / "pkgroot" / package_name / "data" / "irs_soi" / "t" + _write(resource_dir / "manifest.yaml", DUPLICATE_DOCUMENTS["duplicate-vintage"]) + (resource_dir / "a.xlsx").write_bytes(b"a") + (resource_dir / "b.xlsx").write_bytes(b"b") + monkeypatch.syspath_prepend(str(tmp_path / "pkgroot")) + monkeypatch.delitem(sys.modules, package_name, raising=False) + spec = SourceArtifactSpec( + source_name="irs_soi", + source_table="Table 1.2", + resource_package=package_name, + resource_directory="data/irs_soi/t", + manifest="manifest.yaml", + vintage="2023", + extracted_at="2026-09-02", + extraction_method="none", + parser="delimited_text_full_rows", + artifact_year=2023, + ) + + with pytest.raises(ValueError, match="duplicate key"): + spec.assert_parseable(2023)