Public deployment and current known limits: [STATUS](../STATUS.md). # v0.4.2 — record-contract hardening September 23, 2026. This release updates the checked record contract to `0.1.2`. See [software verification](../VERIFICATION.md) and [service status](../STATUS.md). - Added a bounded pre-parse nesting guard shared by the local and hosted runtime paths. The existing post-parse semantic depth limit remains authoritative for the contract's value-depth rule. - Replaced the linear longest-prefix diagnostic with a logarithmic number of substring probes; membership semantics are unchanged and comparison cost still depends on input length. - Exposed declared strengths, the applied threshold and the registry threshold in measurement dependencies. Submitted values that differ from the registry are qualified as custom thresholds; no threshold is presented as calibrated. - Disclosed that `origin` and `accepted` are submitter declarations. Imports preserve them and do not authenticate identity or human review. - Disclosed raw engine heuristic annotations that can fire on bridge identifiers or quotation presence; these annotations are not observed evidence. - Added hardening and disclosure tests to the package allowlist. The original v0.4.1 archive and all earlier archives remain available under their original names. The software checks use bounded inputs; the reported crash-sized requests were not replayed. Deployment and public readback are separate from these checks. # v0.4.1 — typed records and worked-case entry point September 23, 2026. The public root now begins with fictional worked records so people and agents can review inputs before using the evaluator. - Added editable/importable `pfabc-typed-record/0.1` examples and a guided result view with human-readable outcome, evidence, assumptions and limits, plus full-report download and copy controls for agent handoff. - Added the stateless primary `POST /api/records/evaluate` route. It returns `pfabc-record-evaluation/0.1`, uses HTTP 200 for evaluated records and 422 for rejected records, and includes the raw `record_sha256` without persisting the record. Health exposes contract version `0.1.1` and the route path. - The contract wraps the unchanged PFABC engine `0.5.3+review1`. `record_ref` stays an opaque declaration, omitted observation scope is inherited and labelled, and accepted agent declarations remain unauthenticated. - The original `/api/evaluate` route, advanced engine and explicit workspaces keep their existing raw, declared semantics. No authentication, provider login, consciousness determination or science efficacy claim is added. - The release package includes the seven record-contract modules, contract specification, repair and transport tests, regression helpers and three public example records. Existing `v0.4.0-rc1` and `v0.4.0-rc2` archives are preserved; this release uses `pfabc-clarification-lab-v0.4.1.zip`. # v0.4.0-rc2 — public entry point and guided clarity - The public root opens the guided workbench; the local notebook remains at `index.html`. Missing static pages use the custom 404 response. - Corrected Guide's guided-workbench link and added direct navigation between Explore, Practice and the structured lab. - Disclosed evaluation assumptions remain editable, with less initial clutter for a reportability question. Guided results keep the exact evaluated claim beside its conditional status and translate next checks into readable wording; the unchanged technical recommendations and full envelope remain inspectable. - Added tab icons and a visible connecting state during structured-lab startup. - The original evaluator, theory registry, formal specification and limits are unchanged. This pass does not validate the framework or establish human benefit. - Automated-client filtering and the full browser pairing check remain open. The prior source archive remains available under its original download URL. # v0.4.0-rc1 — online evaluation and human-agent workspaces - Added a guided workbench with exact quotation matching, a visible constrained claim packet, explicit theory/mode choices, and the original Python evaluation. - Added a Cloudflare Python transport and opt-in isolated 30-minute workspaces. Ordinary evaluation remains stateless; shared inputs have separate owner and agent capabilities and an owner deletion control. - Retained the local CLI/server, conversation notebook, fictional practice, twenty original engine members, and formal mathematics without changes. - Added transport parity, negative-input, role/isolation and expiry checks. See current [verification](../VERIFICATION.md) and [status](../STATUS.md). - No model-provider login, hidden-state access, empirical validation, or human learning result is implied. Public deployment and GitHub are tracked separately. - Previous release archives remain preserved; this version has a new manifest. # v0.3.0-rc3 — understanding AI interactions September 7, 2026. Clarified the public purpose for developers and anyone using AI: examine the record, interpretations, evidence and open questions about an interaction. - Updated the homepage, reader guide, README and machine-readable introduction. - Added a neutral Request help email link using the existing public Arkhē contact. The link opens an email; no conversation or exercise is attached automatically. - Aligned visible package labels. The earlier v0.3.0-rc1 and rc2 archives are preserved. - Formal mathematics, native runtime behavior, conversation rules, example data and practice logic remain unchanged. No scientific or learning claim was added. --- # v0.3.0-rc2 — Apache-2.0 release licensing September 7, 2026. The owner selected the established ARSENAL licensing approach: standard Apache License 2.0 for the included source, engine data and project documentation, with separate project-identity and release-scope explanations. - Added the unchanged standard `LICENSE`, project `NOTICE`, `LICENSE_SCOPE.md` and `TRADEMARKS.md`; updated licensing references and contribution terms. - Made the license documents available through the local server's explicit file list. - Preserved the prior v0.3.0-rc1 ZIP. This candidate has its own manifest and archive. - No mathematical source, engine semantics, conversation rules or practice logic changed. - License selection is complete; public repository creation and deployment remain pending. The approach was checked against the published [ARSENAL license scope](https://github.com/AdemVessell/arsenal-first-drop/blob/36ec504782b2bda8ec16e678419f14de4da1da1f/LICENSE_SCOPE.md). The full license text is identical to ARSENAL's published `LICENSE` at that commit. --- # v0.3.0-rc1 — a worked exercise and optional native agent runtime September 7, 2026. This supersedes the RC2 website-only candidate below. - Added a positive-credit fictional practice route with a copyable agent brief. - Added a separate local structured-packet lab using the actual supplied Python engine, a CLI, loopback HTTP access, and a shared live evaluation view. - Added strict wrapper parsing, exact-input evaluation records, published local command documentation, and tests of failure and scientific-boundary cases. - Twenty engine source/data members are unchanged; see `native/UPSTREAM.json`. Historical archive adapters, private audits, study logs, and replays are excluded. - The formal v0.5.3 source and existing conversation/guide rules remain unchanged. - Added research context and explicit operating assumptions. Software checks do not validate the framework empirically or establish improved human reasoning. - Provider sign-in, a hosted engine service, MCP, and human pilot results are not part of this candidate. Chrome native interaction was verified; the managed in-app browser showed inconsistent native interactions after an initial success. The legacy conformance note about infinity in formal §3 is implementation history: the selected review runtime rejects non-finite values, consistent with its real-valued mathematical domain. No formal text was silently rewritten. --- # Release notes — website v0.2.0-rc2 Prepared September 7, 2026. Local preview and GitHub contribution package; publication and a reuse license are not established by this build. ## What changed - Clearer purpose, input labels, example navigation, and manual human/AI instructions. - A more readable editorial layout and touch controls, using local assets only. - Substantive initial HTML plus a separate static guide for readers without JavaScript. - Current README, AI proposal guide, glossary, proposed pilot, and bounded readiness report. - The complete supplied v0.5.3 formal specification, preserved byte-for-byte. - A copyable JSON export fallback when browser downloads are restricted. - Current-build software tests and an explicit package file list. Browser rules, migration rules, source-linked guide, and fictional example data remain the RC1 versions. The package version is not the session-schema version. The native Python engine and private self-audit materials are excluded. ## Provenance | Input | SHA-256 | |---|---| | Supplied PFABC_Clarification_Lab_v0.2_RC1.html | `61d456a8ff998771faad3e4cb26cd56e32c12a944b0001f9ac20882677ef661f` | | Supplied PFABC_Clarification_Lab_v0.1.zip | `b5cab343b01089f4029897076543edb6c7e0e24b57a9a5b998a6afcce6bd5b93` | | Preserved consolidated formal specification | `efcb293a4f9a722afa04a7357988fe9566e5465a7731e3fbdeb01c941601c739` | The specification was copied from `pfabc-clarity-prototype/vendor/pfabc-review1/PFABC_v0_5_3_Bridge_State_Semantics_KaTeX.md` inside the supplied v0.1 archive. The standalone RC1 was the implementation base. The older archive's README, test counts, Python adapter and private examples do not describe this successor and must not be substituted for current evidence. No original supplied file was overwritten. The local `reference/` folder is excluded from the release package. No private audit, saved conversation, credential, or historical native result is part of the package file list. ## Claim boundary The output is an experimental clarification tool and reproducible software contribution. It is not a consciousness diagnostic, authenticated evidence service, validated human intervention, or universal AI integration. See `../STATUS.md` for actual current checks and remaining work.