research: From Model Output to Accepted State, owner-review packet

research/from-model-output-to-accepted-state/.gitattributes

@@ -0,0 +1,5 @@
+* -text
+
+evidence/blind-prompt/PROMPT.md whitespace=-blank-at-eof
+evidence/device-activation/BUILD-RECEIPT-000001.md whitespace=-blank-at-eof
+paper/From-Model-Output-to-Accepted-State-Owner-Review-v4-LinkedIn.md whitespace=-trailing-space

research/from-model-output-to-accepted-state/BOUNDARIES.md

@@ -0,0 +1,41 @@
+# Claim and reuse boundaries
+
+This file is load-bearing. A result copied from this packet should retain the relevant boundary.
+
+## What the checks establish
+
+- The files match the byte lengths and SHA-256 digests in `release-manifest.json`.
+- The retained JSON reports parse and remain byte-identical to the audited local outputs.
+- The paper outputs passed the local structure, visual, and privacy checks recorded in `evidence/RELEASE-BUILD-RECEIPT-000004.md`. The dependency-bearing validator used for those preparation checks is not included in this zero-dependency public packet.
+- The finite fixture results are exact for the frozen records, events, queries, candidate fields, equality rules, and implementations named in their reports.
+- `evidence/editorial-review/PUBLIC-DISPOSITION.md` records the owner's minimized editorial decisions and the supplied private source's byte identity. It does not authenticate the reviewer attribution or promote review assertions into evidence.
+
+## What the checks do not establish
+
+- scientific peer review, independent replication, mathematical novelty, or publication acceptance;
+- truth of source assertions, trusted time, authorship identity, causal validity, or authority;
+- production safety, security, privacy compliance, legal compliance, or fitness for a particular deployment;
+- a universal minimal state, universal proof canonicalizer, optimal policy, or complete literature review;
+- that a probability forecast is an outcome, that one resolved case establishes calibration, or that agent agreement is evidence of external truth;
+- that requested model labels in the frozen-oracle packet attest the runtime model, seed, sampling process, or independence of the responses;
+- that the private editorial review is evidence, authorship, peer review, source truth, or independent validation.
+- that retained-source IDs expose or reproduce the corresponding private bytes.
+
+## Vocabulary crosswalk
+
+- **Output**: a produced representation awaiting qualification. It is not yet an outcome.
+- **Outcome**: a later qualified observation about what occurred under a declared resolution rule.
+- **UNRESOLVED**: a condition result in the paper. It does not silently become PASS.
+- **HOLD**: a proposed operational disposition for unresolved required conditions.
+- **Frozen-oracle packet**: BP-001, whose oracle and semantic rubric were fixed before collection and withheld from responders. The legacy `evidence/blind-prompt/` path is retained for receipt continuity. The term does not imply blinded assignment or blinded assessment.
+- **BLOCK**: the execution result tested for BP-001's frozen `HOLD` value. BP-001 did not itself test the paper's `UNRESOLVED -> HOLD` crosswalk.
+- **Accepted state**: the projection produced by the declared reducer from accepted events under pinned policy. It is not the entire world state.
+
+## Publication boundary
+
+This directory is prepared for review by pull request. An open PR, branch, commit, or passing verifier does not authorize merge, release tagging, DOI registration, deployment, or a claim that the owner has accepted every semantic mapping.
+
+The private editorial review and its locator remain outside the public packet.
+Only the minimized owner disposition is authorized for this review surface.
+Absolute local-machine paths, attachment locators, workspace-only paths, raw review
+text, and private prompt responses are outside the publication boundary.

research/from-model-output-to-accepted-state/CHANGELOG.md

@@ -0,0 +1,42 @@
+# Owner-review revision lineage
+
+This changelog describes the public owner-review packet. Historical receipts remain
+unchanged; each later receipt names its predecessor.
+
+## 0.1.0-owner-review.1
+
+- Initial minimized public paper packet.
+- Added authored source, figures, bounded evidence, license, boundaries, manifest,
+  and separate Python and JavaScript release verifiers.
+- Receipt: `evidence/RELEASE-BUILD-RECEIPT-000001.md`.
+- PDF SHA-256: `1707ffab851bc963a2874a6303c1e2d2aa5c9db2e262343d921f9d7df839b4ca`.
+
+## 0.1.0-owner-review.2
+
+- Dispositioned private editorial feedback without publishing the raw review.
+- Calibrated the diagnostic-count, refusal-arm, response-packet, operator-threat,
+  non-stale-label, and source-location language.
+- Added LinkedIn document compatibility checks and a complete 35-page visual pass.
+- Receipt: `evidence/RELEASE-BUILD-RECEIPT-000002.md`.
+- PDF SHA-256: `019e372263176cb693e00a2be548c5f0dfb04c5027cb78b1f107070fc2e1afc4`.
+
+## 0.1.0-owner-review.3
+
+- Separated the auditability contribution from unsupported-claim accuracy results.
+- Added exploratory pairwise Fisher values with their clustering ceiling.
+- Added the exact standalone Lean query-quotient source and compile receipt.
+- Gave revised outputs a distinct v3 filename and repeated visual, technical,
+  privacy, and release-packet checks.
+- Receipt: `evidence/RELEASE-BUILD-RECEIPT-000003.md`.
+- PDF SHA-256: `26e7f35e5e4fb125ba4339d0179cccf662d5d0fcbf09d8e27693b3b74fb0767c`.
+
+## 0.1.0-owner-review.4
+
+- Standardized BP-001 as the frozen-oracle packet while retaining its legacy path
+  for receipt continuity.
+- Added the explicit Lean chronology and clean-room reproduction boundary.
+- Added packet-local line-ending protection and a clean-clone runbook.
+- Added a marker-blind claim register, separate author key, and external-review
+  template so demotion test 6 can be executed.
+- Added this in-document and packet-level revision lineage.
+- Final output digests are recorded in release receipt 000004.

research/from-model-output-to-accepted-state/CITATION.cff

@@ -0,0 +1,27 @@
+cff-version: 1.2.0
+message: "If you use or build on this owner-review release, cite the exact version or commit and retain its stated limitations."
+title: "From Model Output to Accepted State"
+type: report
+authors:
+  - family-names: Tiller
+    given-names: Jake
+version: "0.1.0-owner-review.4"
+date-released: 2026-08-15
+license: CC-BY-4.0
+repository-code: "https://github.com/JakeTOpenSource/Resilience-Ledger"
+url: "https://github.com/JakeTOpenSource/Resilience-Ledger/tree/main/research/from-model-output-to-accepted-state"
+abstract: >-
+  Owner-review draft of a typed boundary from probabilistic model output to
+  accepted state, with deterministic replay, finite query-sufficiency and
+  transition-refinement fixtures, and a separately audited probabilistic
+  forecast lane. The packet includes bounded local evidence and offline
+  integrity verification; it is not an independent validation or deployment
+  authorization.
+keywords:
+  - state transition protocol
+  - accepted state
+  - deterministic replay
+  - query sufficiency
+  - finite-state refinement
+  - probabilistic forecasts
+  - AI governance

research/from-model-output-to-accepted-state/LICENSE.txt

@@ -0,0 +1,20 @@
+From Model Output to Accepted State
+Copyright (c) 2026 Jake Tiller
+
+This release packet, including its manuscript, figures, documentation, evidence
+reports, and verification source, is licensed under the Creative Commons
+Attribution 4.0 International License (CC BY 4.0).
+
+You may share and adapt the material for any purpose, including commercially,
+provided that you give appropriate credit, provide a link to the license, and
+indicate whether changes were made.
+
+Human-readable summary: https://creativecommons.org/licenses/by/4.0/
+Full legal text: https://creativecommons.org/licenses/by/4.0/legalcode
+
+The material is provided AS IS, without warranties or conditions of any kind.
+It is independent educational research, not legal, compliance, security,
+medical, financial, or other professional advice. Names and trademarks of cited
+projects and organizations belong to their respective owners. This license does
+not relicense third-party works that are quoted, cited, or linked. Citation does
+not imply affiliation or endorsement.

research/from-model-output-to-accepted-state/README.md

@@ -0,0 +1,68 @@
+# From Model Output to Accepted State
+
+Status: **OWNER-REVIEW DRAFT**<br>
+Release packet: `0.1.0-owner-review.4`<br>
+Recorded: 2026-08-15
+
+This packet publishes a bounded draft, its readable outputs, the authored manuscript and figure source, minimized local evidence reports, and offline integrity checks. The dependency-bearing assembly helper remains outside this zero-dependency public boundary. This is a review surface, not a claim of publication acceptance, independent replication, production safety, or deployment authority.
+
+## Start here
+
+1. Read [`paper/From-Model-Output-to-Accepted-State-Owner-Review-v4.pdf`](paper/From-Model-Output-to-Accepted-State-Owner-Review-v4.pdf).
+2. Read [`BOUNDARIES.md`](BOUNDARIES.md) before reusing a result.
+3. Follow [`REPRODUCE.md`](REPRODUCE.md) for a clean-clone check.
+4. Run the packet verifier:
+
+   ```powershell
+   powershell -NoProfile -ExecutionPolicy Bypass -File .\tools\verify.ps1
+   ```
+
+The verifier is offline. It checks the release allowlist, raw byte lengths, SHA-256 digests, payload root, UTF-8 boundaries, JSON syntax, Python syntax, PDF signature, and selected privacy and credential patterns. The JavaScript and Python implementations must return identical canonical reports.
+
+The packet adds no package dependency or runtime network call. The owner workspace used Chrome and pypdf to render and inspect the retained PDF, but those optional build dependencies and their helper scripts are deliberately outside this public release boundary. The authored content, figures, styles, rendered outputs, evidence reports, and raw-byte verification floor are included.
+
+## What is in the packet
+
+| Path | Function |
+|---|---|
+| `paper/` | Tagged owner-review PDF and a LinkedIn-safe Markdown rendering. |
+| `content_*.py`, `figures.py`, `style.py`, `figures/` | Authored manuscript, figure source, rendered SVG figures, and print style. |
+| `evidence/device-activation/` | Frozen finite query-sufficiency result and its original local receipt. |
+| `evidence/transition-stable-quotient/` | Frozen future-stability refinement result and its original local receipt. |
+| `evidence/blind-prompt/` | Legacy receipt-stable locator for the frozen-oracle packet prompt and aggregate-only public summary. Raw responses, quote-bearing reports, per-response digests, and owner-private mappings are excluded. |
+| `evidence/editorial-review/` | Public minimized owner disposition of a private editorial review. The raw review and private locator are excluded. |
+| `evidence/lean-query-quotient/` | Exact standalone Lean source and append-only pinned compile receipt. The module is not imported by the package root and is not an upstream Mathlib contribution. |
+| `release-manifest.json` | Complete raw-byte allowlist for every packet file except the manifest itself. |
+| `claims.json`, `author-markers.json`, `reviewer-markers.template.json` | Marker-blind claims, the separate author key, and a reviewer template for demotion test 6. |
+| `REPRODUCE.md`, `CHANGELOG.md` | Clean-clone verification and append-only owner-review revision lineage. |
+| `tools/` | Deterministic manifest writer plus separate Python and JavaScript verifiers derived from one release contract. |
+
+## Bounded results
+
+- The paper separates candidate output, authorized action, observation, acceptance, and later outcome rather than treating them as one status.
+- In one synthetic 151-trace device fixture, all 1,023 nonempty subsets of ten declared candidate fields were checked for seven declared queries. Exactly one five-field subset was minimum by field count within that frozen model. It is not a universal device state or bit minimum.
+- In the same finite model, the full seven-query signature was already transition-stable at 33 classes. Removing `nextPermittedActions` produced 18 static classes that refined to the same 33 classes after one round. This is a local Moore/Myhill-Nerode-style result, not new automata theory.
+- Three frozen response configurations reproduced 27 of 27 exact answer fields in BP-001, the frozen-oracle packet. Six semantic functions were unambiguously unanimous; other semantic mappings remain bounded or unresolved as stated in the paper. Requested model labels were metadata, not runtime identity attestation.
+
+## Editorial review disposition
+
+[`evidence/editorial-review/PUBLIC-DISPOSITION.md`](evidence/editorial-review/PUBLIC-DISPOSITION.md)
+binds one private owner-supplied editorial review by byte count and SHA-256 and
+publishes only the owner's minimized dispositions. The reviewer attribution is
+owner-attested, not runtime-authenticated. The review is editorial assistance,
+not evidence, authorship, peer review, source truth, or independent validation.
+
+## Related-work credit
+
+With his permission, Jake Macdonald's OpenGoldenRatio (OGR) v0.1 is cited as parallel related work. Macdonald reviewed the bounded comparison and helped clarify the distinction between STP's governed path from candidate output to accepted state and OGR's containment of actor or agent relations. He contributed no code, data, experiments, or authorship to this release. See [`RELATED-WORK.md`](RELATED-WORK.md).
+
+## Reuse
+
+The packet is licensed under CC BY 4.0. Cite the exact version or commit you used, keep the status and limitations attached to extracted results, and identify modifications. A passing integrity check establishes byte consistency with this manifest; it does not establish that a claim is true or authorized for a new context.
+
+The release contains no absolute local-machine path, Codex attachment locator,
+workspace-only `work/...` locator, raw prompt response, private review text, account
+credential, email address, phone number, or local network address. Retained private
+inputs are identified only by bounded source IDs and digests.
+
+See repository for full list of sources and contributions.

research/from-model-output-to-accepted-state/RELATED-WORK.md

@@ -0,0 +1,23 @@
+# Related work and contribution boundary
+
+## OpenGoldenRatio
+
+Jake Macdonald granted permission to cite *OpenGoldenRatio (OGR) v0.1: Containment-First Multi-Agent Governance Protocol* as parallel related work and reviewed the comparison in this draft.
+
+The comparison retained in the paper is narrow:
+
+- STP follows the governed transformation from candidate output toward accepted state.
+- OGR focuses on containment and governed relationships among actors or agents, including whether a permitted action may propagate consequence.
+- Both separate proposal, evidence, permission, and consequence more carefully than a single undifferentiated approval state.
+
+Macdonald's contribution to this project was review and clarification of that comparison. He did not contribute code, data, experiments, implementation, or authorship, and OGR is not used as evidence that STP works. Neither work is described as deriving from, implementing, or subsuming the other.
+
+Citation:
+
+> J. Macdonald. *OpenGoldenRatio (OGR) v0.1: Containment-First Multi-Agent Governance Protocol*. Zenodo, 2026. https://doi.org/10.5281/zenodo.18969396
+
+Pinned demonstrator reviewed for the paper:
+
+`https://github.com/macess888-cmyk/open-golden-ratio-demo/commit/58450185582f4ecf1410b33f77e22d8d4b0441a2`
+
+See the paper's references and related-work section for the full bounded comparison.

research/from-model-output-to-accepted-state/REPRODUCE.md

@@ -0,0 +1,90 @@
+# Reproduce the release integrity check
+
+This procedure verifies the exact bytes of this owner-review packet. It does not
+rebuild the PDF, reproduce excluded inputs, independently replicate an experiment,
+or establish the truth, novelty, authorship, authority, safety, or fitness of a
+claim.
+
+## Requirements
+
+- Git
+- Python 3
+- Node.js
+
+No package installation or runtime network access is required after cloning.
+
+## Obtain a clean checkout
+
+Use the exact release commit or tag identified by the repository release or pull
+request.
+
+```sh
+git clone https://github.com/JakeTOpenSource/Resilience-Ledger.git
+cd Resilience-Ledger
+git checkout --detach <release-commit-or-tag>
+git status --porcelain=v1
+cd research/from-model-output-to-accepted-state
+```
+
+`git status --porcelain=v1` must print nothing. The repository-root and packet-local
+`.gitattributes` files disable line-ending conversion so byte comparisons do not
+produce a false failure on Windows. Do not remove that rule when re-homing the
+packet.
+
+## Windows
+
+```powershell
+powershell -NoProfile -ExecutionPolicy Bypass -File .\tools\verify.ps1
+```
+
+A successful result reports `VERIFY PASS`, cross-language parity, the checked file
+count, payload root, manifest SHA-256, and `status=PASS`.
+
+## macOS or Linux
+
+```sh
+python3 -B tools/update_claim_register.py --check || exit 1
+python_report="$(python3 tools/verify_release.py)" || exit 1
+node_report="$(node tools/verify-release.mjs)" || exit 1
+
+if [ "$python_report" != "$node_report" ]; then
+  printf '%s\n' "Cross-language canonical report mismatch" >&2
+  exit 1
+fi
+
+printf '%s\n' "$python_report"
+```
+
+If Python 3 is installed as `python`, substitute that executable name.
+
+## Claim-marker reassignment
+
+For demotion test 6, give an external reviewer only:
+
+- `claims.json`, which embeds the neutral four-marker policy;
+- `reviewer-markers.template.json`;
+- the packet files named by accessible `sources` records in `claims.json`; and
+- any public external source the reviewer retrieves and verifies against its
+  registered identity: commit and tree for a tree record, or commit, path, byte
+  length, and SHA-256 for a blob record.
+
+Copy `reviewer-markers.template.json` outside the checkout before filling it; editing
+the packet copy correctly breaks its byte manifest and clean-worktree check. Withhold
+`author-markers.json`, `content_a.py`, `content_b.py`, `content_c.py`, the PDF, and
+the LinkedIn companion until the reviewer seals the assignments. A retained or
+unavailable source stays unavailable and must be recorded that way; it may not be
+silently treated as reviewed. The separation is a procedural blind, not
+cryptographic secrecy after publication. After the reviewer seals assignments,
+compare by `claim_id`: a mismatch is `CONTESTED/HOLD`, a missing assignment is
+`INCOMPLETE`, and neither result can auto-promote a claim.
+
+## Boundaries
+
+Do not redirect verifier output into this packet: an added file correctly causes an
+allowlist failure. Do not run `tools/update_manifest.py` as a verification step; it
+rewrites the manifest and is a maintainer-only release operation. Do not run
+`tools/update_claim_register.py` during verification; use its `--check` option.
+
+After verification, `git status --porcelain=v1` should still print nothing. A pass
+establishes consistency with the committed manifest and checkout. The Git commit or
+release tag is the external identity anchor.