Bind the Chronicle transport and energy facts on the UK national surface (#890 PR-T)

[juaristi22]

Target-binding half of #890 (PR-T of the two-PR plan in repos/uk-890-implementation-plan.md; the spine half, PR-S, follows on main). Everything here binds pinned Chronicle facts and lets the solver calibrate — the #790 ruling — and nothing in it changes an imputed value.

Commits

1. Re-pin the Chronicle feed to c6f9361 (chronicle Add state-level SNAP benefit hard targets from USDA FNS administrative data #255 + UK firm generator: net-VAT liability missing the 0.20 rate factor (inherited from firm-microsim-paper pre-fix) #258: 138,847 rows) as one declaration for both grains. Round 2 renamed uk/national_chronicle_feed.json to uk/chronicle_feed.json and made the local census, local references and local validation levels restate that pin instead of carrying their own (the base already includes Bind UK census households from Chronicle and map NI Data Zones with NISRA's published lookup (#887) #900, which had moved both to ec7169b). No existing reference on either surface resolves a different fact or value: national 415/7/8 unchanged before the new bindings, local 20,430 active / 1,586 no-fact / 513 signed deferred / 1 signed compile error unchanged; only the feed labels move.
2. Vendor pinned facts into build resources. tools/vendor_uk_ledger_facts.py + uk/ledger_fact_vendor_selections.json copy the facts the stages and diagnostics need into nine per-concern resources, verbatim, with the feed identity; a test regenerates them byte-identical from the pin. This is the doctrine "fixtures of publisher facts come from Chronicle" applied to build resources and retires the chronicle_candidate anchors one concern at a time (PR-S consumes them).
3. Region leg. The generator pins region-only contract targets at region level; the compiler validates region pins against the nine English region codes. The two London bus rows bind (fares GBP 1,347.4m, support GBP 1,130.2m); England outside London is derived; the UK rows stay signed out (Wales fare revenue unpublished; UK support replaced by four heterogeneous country legs).
4. BUS0415 alignment, declared. The four dft.bus_fare_receipts.* contract rows declare uprating_index: dft.local_bus_fares_index; the generator applies it while authoring (membership records the applied value with index, factor, value_before_uprating) and the runtime applies the same applier keyed on the declaration, so membership, parity receipts and the run manifest carry one value (England 3,417,388,656 → 3,612,036,036, ×1.05696; London ×1.0). Net support declares no index and is never aligned. The index series is read from the vendored dft_bus_value_anchors.json, refused if its feed identity differs from the pin, so the production factors reproduce in CI without the licensed feed. At target period 2023 (production parity receipt) the same alignment gives England ×0.99368 and London ×0.99769, below one because the fare cap held index growth inside calendar 2023 under the fiscal-year mean.
5. New targets (registry scope, unmapped declarations): obr.fuel_duties_cars (cars receipts) on the engine's cars-only fuel_duty, bound at FY2024-25 (period_value: 2024, GBP 14.4bn, held to 2025 like the CGT rows) per María's ruling that a fiscal-year source binds the base data's FY2024-25 or CY2025 and never a later projection; every row of the vehicle-type release is the April 2024 vintage, so this is that vintage's figure for the base year (its all-vehicle FY2024-25 total overshoots the March 2026 outturn by 1.4%). obr.fuel_duties is signed out and its Swap the microcosm-built UK spine into the national build and retire the incumbent feed #757 measure exclusion retired. Open for María: the rest of the obr family still binds FY2025-26 projections at period 2025 by the inherited uk-data policy; orr.rail_government_support (7270, GBP 21.6bn, GB); ons.household_energy_expenditure (COICOP 04.5, CY2025 GBP 43.4bn); Scotland passenger revenue and government support; Wales public support (local-authority route); NI Translink receipts; NI public transport support as a joint bus-plus-rail cell.
6. Baseline receipts, no exclusion. On the newest gated national spine (spine-p) every new row fits inside the 25% bound after calibration. In round 1 the cars fuel-duty row sat at −25.6% against the FY2025-26 projection and carried a dated exclusion; against FY2024-25 it solves to −16.7% (the frame's calibrated cars duty is GBP 12.0bn either way, so the change is the target, not the frame), the round-2 run blocked on that exclusion being stale, and the register is back to empty as on main (Part G of the receipts). The receipts record what the fit costs today: the London bus rows, the Scotland and Northern Ireland legs and the ORR rail row are closed by weight stretch (73–85% of London bus mass and 71% of rail support mass sit on households stretched more than 5×), which is exactly what PR-S's value-side work removes. ONS household energy expenditure already sits on the frame at design weights.

Round 2 (Vahid's review)

• Should-fix 1 (alignment as a declaration): done as commit 4 describes; test_childcare_and_bus_references_compile_with_declared_provenance now pins the applied values and checks them against the membership.
• Should-fix 2 (consumers): the register now separates consumers (present readers; a test imports each module and checks it names the file) from planned_consumers (PR-S). One resource has a present reader, the BUS0415 series.
• Should-fix 3 (adjudication): moot, the entry is retired (Part G); the plan of record stays outside the tree by repo convention and its rulings are summarised in this body.
• Should-fix 4 (CI): the separate local feed default is gone with the pin unification; the alignment factors, the vendored headers and the membership feed identity are all checked feed-free in CI now; provisioning the pinned feed in CI so the regeneration tests run there is a follow-up (predates this PR).
• Questions: the 2023 factors and the OBR vintage are in the body above; covered_by_country_legs is now replaced_by_heterogeneous_country_legs with the scope of each leg spelled out. Nits: the Part F placeholder is gone.
• María's rulings: one Chronicle pin (commit 1) and FY2024-25 for the cars fuel duty (commit 5); the baseline was re-run on that surface (Part G) and the fuel exclusion retired.

Rebase onto main (after #903, #894, #855)

Main's schema-8 hierarchy now requires a catalog category_id and a label on every contract target: the eight #890 targets declare them (four new providers, six new categories, the cars fuel duty under obr.efo_receipts), and the committed references carry the contract-derived hierarchy lookups the generator emits. What the rebase exposes and does not fix: main's hierarchy completion needs dimension labels on every selected Chronicle fact (chronicle#261, María's), which no artifact carries, so on main and on this branch alike the generator and compile_uk_target_registry return nothing against the pinned feed (0 / 424 unsupported). The reference surfaces here are the round-2 regenerations plus those lookups; the feed-backed regeneration tests skip in CI and are blocked locally until #261 or the interim Microcosm fallback that issue announces. The nine English regions join Microcosm's UK geography label catalog (UK_GEOGRAPHY_ID_TO_LABEL): the London fact carries no geography name and the catalog had only country codes, so the region leg would otherwise be refused at compile once labels arrive. Vendored resources still regenerate byte-identical. Receipts Part H.

Surface

Contract 236 → 244 targets (registry 202 → 210); active references 415 → 424; 7 deferred; signed out 8 → 7. National runtime compile 424 / 0 unsupported on the pinned feed. Compile-parity receipts regenerate with the new rows as ledger-only entries.

Receipts

experiments/890-fuel-bus-energy-receipts.md; licensed evidence under data/ukds/acceptance/890-fuel-bus/.

Open rulings carried in the plan

• Rail: 7270 broad total (GBP 21.6bn, bound here) versus 7271 all-sources operational (GBP 11.9bn, of which DfT GBP 9.5bn). The frame carries GBP 6.1bn of rail support at design weights, so 7270 is closed by a ×3.6 stretch on the 14% of households with rail use; a value-side treatment or a dated exclusion may be the better posture until then.
• The NI joint public-transport cell versus a declared bus share; ONS 07.2.2 as a diagnostic; the outside-London reason id wording.

Verification

Rebased head 3bb29277 on main 3094bfe8. Local lanes on that tree: uk 2,089 tests, 4 failures, all four the chronicle#261-blocked feed-backed cases (test_committed_surfaces_regenerate_from_pinned_feed[national], [local], test_childcare_and_bus_references_compile_with_declared_provenance, test_bus0415_alignment_on_the_pinned_feed, which skip in CI and fail on main the same way), everything else green (the pre-existing eFRS parity failure is gone, retired on main by #903); shared-spec 1,903 tests, 0 failures. tools/ci_test_groups.py --verify, census_uk_local_targets --check, census_uk_data_target_parity --check, vendor_uk_ledger_facts --check and ruff clean. Round 2 (pre-rebase, 3fdede45) baseline runs pr-t-round2-spine-p and pr-t-round2b-spine-p are in receipts Part G.

Not in this PR

Every imputed value is unchanged: the car-count predictor, donor uprating, bus incidence, the bus rakes and the energy repricing are PR-S. No dense candidate run.

🤖 Generated with Claude Code

[vahid-ahmadi]

Review at a256c880 (Claude Code, high effort; worktree checkout, the six commits read in order; without the engine or the pinned feed: 382 passed, 7 skipped, 0 failed across the twelve touched suites; ruff check and ruff format --check clean on all 16 changed files; both census tools, the coverage manifest, the spec-engine coverage check and ci_test_groups --verify current; CI in progress with no failures so far. Draft, so a comment rather than a review.)

The binding work reproduces. The re-pin commit on its own moves only the membership feed label, so "no existing reference resolves a different fact or value" holds for it. All nine vendored resources carry the new feed identity with matching row counts. Every new resolved value rebuilds from the vendored rows: ONS 04.5 at £43.427bn, Scotland 391m and 499m, Wales 62.4m + 69.0m, DfI 49.6m + 100.5m, NITHC 61.8m + 49.9m, and the BUS0415 factors (England 1.056957928802589, London 1.0) from the four quarter-end means. All five measurement variables exist on the engine at household entity, the two-term value_expression is supported, the region roster and the two region asserts fail closed, and the London derived member is a signed derived_partition_member exclusion rather than a computed target, which is the right shape. The fuel-duty deferral is a dated register entry, not prose, and the receipts are honest that the terminal battery fails without it.

Should-fix

1. The first applied rescale on the UK surface is a runtime post-pass, not a declaration, and the surface now carries two values for one target. align_dft_bus_fare_receipts_to_period (uk_runtime/ledger_targets.py:364-424) runs inside compile_uk_target_registry, keyed on the dft_local_bus family and the fares concept string, and multiplies spec.value by the factor. LedgerTargetReference already has an uprating_index field that flows to spec metadata; the reference does not use it. So target_references.json shows the two rescaled fares rows and the two lag-only support rows in identical shape (uprating_from_period: "2024", no index), uprating_holds in the membership lists all four as the same 2024 → 2025 hold, the membership records England fares at the raw 3,417,388,656 while both parity receipts and the runtime compile record 3,612,036,036, and test_childcare_and_bus_references_compile_with_declared_provenance still pins the unaligned value on the generic compiler. #874 established that a hold is a declared lag with values unchanged; this PR makes the same field trigger a value change for one concept by code. The compiled spec metadata is auditable (index, factor, months, means, record ids, pre-alignment value all present), so the fix is declaration rather than logic: put uprating_index on the two references, let the generator apply or record it, and make membership and parity agree on one value.

2. The vendored register's consumers field is not true yet. ledger_fact_vendor_selections.json lists uk_runtime.ledger_targets as a consumer of three resources, but _fares_index_series reads the feed rows and nothing outside ledger_fact_vendoring.py and its tests loads any vendored file. Nine files, 1.7 MB, no runtime reader; every consumer is PR-S. Not a second source of truth today, but the register says it is. Either mark them prospective or wire the one reader this PR does have.

3. The reviewed exclusion's adjudication is untracked. obr.fuel_duties_cars@2025 cites repos/uk-890-implementation-plan.md (approved 2026-09-10) and a run id; the body and receipts also cite data/ukds/acceptance/890-fuel-bus/. Neither path is in the tree, and neither is gitignored. Every other register entry cites a GitHub comment id. Same ask as on #903: commit the plan or cite something a reader can open.

4. "Regenerates byte-identical under test" runs nowhere in CI. The workflow sets no pinned-feed path, so the vendored regeneration, the reference regeneration and the BUS0415 pinned-feed tests skip in CI as they do here. The local-surface regeneration test also now defaults to a new .codex-work/consumer_facts_uk_local.jsonl, so on a machine with only the national feed it flips silently from run to skip.

Questions

• The production-2023 parity receipt moves two rows the body does not mention: England fares 3,278,976,692 → 3,258,254,371 (factor 0.99368) and a new London row at 0.99769 of the FY2023-24 fact. Both are the alignment at 2023 with factors below one from the fare cap. Worth a line, since the body gives only the 2025 factors.
• The OBR cars fuel-duty rows are all vintage april_2024, so the FY2025 £15.9bn bound at period 2025 is a projection two and a half years old. allow_source_projection makes that consistent with the previous obr.fuel_duties row; the vintage should be visible in the body.
• dft.bus_net_support.uk is signed out as covered_by_country_legs, but the legs are heterogeneous — NITHC is a bus-plus-rail cell excluding the £53.2m other revenue funding, Wales is the local-authority route only. The notes disclose it; "covered" overstates.

Nits

Receipts Part F still carries the placeholder "(filled from … once the run completes)". The two BUS0415 windows share the March 2025 quarter-end, which the basis string declares.

#903 overlap

Same finding as posted there: five of the seven shared files auto-merge, two conflict (test_uk_measure_simulation.py counts — merged truth 47 / 35 with both census markers dropped — and the UK spec sha). Land #903 first and rebase this; also re-run census_uk_data_target_parity --check after, since #903 touches that register.

Could not verify

Regeneration from the pinned feed (no feed here, skipped in CI too), the spine-p baseline runs and the acceptance evidence, and the plan-of-record rulings.

---

Nothing blocks. 1 is the one that matters: it is the first value the UK surface changes rather than declares, and it should be visible on the reference, not only in the run manifest. 2 and 3 are register accuracy; 4 is a CI gap that predates this PR but bites harder now.

[juaristi22]

Round 2 at 3fdede45 (all four should-fixes, the three questions and the nits; two of María's rulings landed at the same time).

1. Alignment as a declaration. The four dft.bus_fare_receipts.* contract rows now declare uprating_index: dft.local_bus_fares_index. The generator carries it onto the reference and applies the UK applier while authoring, so target_reference_membership.json records the applied value with index, factor and value_before_uprating on the hold (England 3,417,388,656 → 3,612,036,036; London ×1). The runtime applies the same applier keyed on the declaration (apply_declared_uk_uprating; an index no applier implements refuses the compile), so membership, both parity receipts and the run manifest carry one value. Net support declares nothing and holds unchanged. test_childcare_and_bus_references_compile_with_declared_provenance pins the applied values and checks them against the membership.

2. consumers. Split into consumers (present readers) and planned_consumers (PR-S). The BUS0415 series is now read from the vendored dft_bus_value_anchors.json (refused if its feed identity differs from the pin), which makes uk_runtime.ledger_targets a real reader of that one resource; the other eight list their readers under planned_consumers. A test imports every present consumer and checks its source names the file.

3. Adjudication. Moot: the entry is retired. With the cars fuel duty bound at FY2024-25 (below) the row solves to −16.7%, the round-2 run blocked on the exclusion being stale, and the register is back to empty as on main (receipts Part G). The plan of record lives outside the tree by repo convention (plans are not committed), same as #903; its rulings are summarised in the body.

4. CI. The separate local feed default is gone: María's ruling is that the national and local surfaces share one pin, so uk/national_chronicle_feed.json became uk/chronicle_feed.json and the local census, local references and validation levels restate it (both surfaces regenerate byte-identical from c6f9361 apart from the feed label; local values unchanged). What now runs feed-free in CI: the BUS0415 factors from the vendored series (test_bus0415_alignment_from_the_vendored_series), the vendored headers against the pin, and the membership feed identity. Provisioning the licensed feed in CI so the regeneration tests run there is a follow-up that predates this PR.

Questions. 2023 factors (England 0.99368, London 0.99769, below one from the fare cap) and the april_2024 vintage are in the body and receipts. covered_by_country_legs → replaced_by_heterogeneous_country_legs, with each leg's scope spelled out in the rationale (Wales LA-route only, NITHC bus-plus-rail excluding the GBP 53.2m other revenue funding).

Also in this round (María's ruling). The cars fuel duty binds at FY2024-25 (period_value: 2024, GBP 14.4bn, held to 2025 like the CGT rows), not the FY2025-26 projection: fiscal-year sources bind the base data's FY2024-25 or CY2025. The baseline was re-run on the round-2 surface (Part G): every new row inside the 25% bound, loss 0.01123, 95.45% within 10%, ESS 10,528; the cars fuel duty goes −41.0% → −16.7% with the frame's calibrated duty at GBP 12.0bn either way, so the change is the target, not the frame.

#903 overlap. Agreed on the order; this PR rebases after #903 lands and census_uk_data_target_parity --check is re-run then (it is current at this head).

Nits: the Part F placeholder is gone.

Verification is in the body: shared-spec 1,888/0; uk 2,050 with the known eFRS parity failure plus one I introduced and fixed in the same commit (the validation-levels loader now reads the shared pin, and that file's drift test patches json.loads for the whole call; the pin is read once at import, 28 tests re-run green). UK spec digest re-pinned.