gnomonlearn
Docs/RFC-0006 ClaimStatus regression condition

RFC-0006 ClaimStatus regression condition

Versioned draft snapshot

RFCs, reference catalogs, and evidence reflect Spec 3b2e3e7a114fb826b1445d39a48756b8fff629cd. Counts and pending/status notes belong to that snapshot. Read the dated project update for later progress.

Source repositories are private during staging. Links marked “repository access required” need access to the linked repository.

This is a trace and gap record from the pinned draft specification. It does not establish proof or implementation conformance.

One projection literal is restored under the adopted canonical ClaimStatus flow at Spec baseline 6bbd489aed90ffc10d2b9a0679a3acf62c633ee3. The owner choice is recorded in CLAIM_STATUS_MIGRATION.md. Authored RFC-0006 §7.1 requires governing-policy permission and an explicitly declared reversible regression. The unchanged Lean b22/b25 catalog rules reject non-reversible regressions; this repair establishes no new proof.

The sole maintained edit is claimText on claim/RFC-0006/sec-7/claimstatus-monotone-default. The old literal named policy permission alone; the new literal also requires the flow to declare the regression reversible: true. Every other property, IRI, canonical subject, state, and normative source is unchanged.

Actual results

CheckResult
Owning projection emissionExactly one old/new claimText triple
Combined RDF graph89,351 triples before and after
Owning reviewSubstantively re-reviewed; now matches
Other reviewsAll 6,961 messages and unrelated textual blocks unchanged
Review ledger6,962 reviewed; 2,306 mismatches; zero unreviewed
Displayed query resultsAll 38 multisets and columns unchanged
Complete relations without outer LIMITAll 14 unchanged

verify_adoption.py (repository access required) proves the exact one-literal source delta, source/Turtle equality, one removed/added triple, complete protobuf ledger equality except the owning hash/finding/verdict, and graph restoration. The complete node was compared with §7.1, canonical subject ownership, and Lean b22/b25. This is not fingerprint-only promotion.

The script reuses the existing query comparison functions and query-effects.proto (repository access required). It checks all 38 SELECT queries and removes the terminal outer LIMIT for all 14 limited queries. Bounds remain 300 seconds wall, 240 seconds CPU, 250,000 rows per relation, and an 8 MiB effect file.

Before the edit, the first execution captured before-claim-source.textproto (repository access required), before-claims.ttl (repository access required), before-review.textproto (repository access required), before-inputs.textproto (repository access required), and before-queries.textproto (repository access required). Query data includes complete multiset hashes, columns, and row counts. The final typed comparison (repository access required) binds current inputs and retained preimages separately.

Passing gates

The existing tools all returned zero:

The final comparison returned zero in ead139 (launch cae2ff, session 29447). The five final gates returned zero in fb30d3 (launch d8b7c7, session 22371). They used Python 3.14.7 and existing protoc. The comparison and final gates used retained Python dependencies from spec-site-approved-projections-preview-links-v1b-20261015/python-runtime. The initial script invoked the emitter with the same Python executable and existing installed dependencies. No installation or acquisition ran; no complete Python import-closure audit is claimed.

Preserved verifier failure

The first verifier returned one after the source/RDF/review edits at an incorrect equality comparison of blank-node identifiers from separate parses of unchanged ontology files. The exact named one-triple delta had passed. Its original bytes remain in verify_adoption.initial.py (repository access required) and its actual failure in adoption-verification.log (repository access required). The before-input catalog records that original verifier’s hash at its original path; the preserved initial file contains those exact bytes.

The corrected verifier binds unchanged RDF bytes and proves the sole changed source/Turtle triple without assuming separate parses assign identical anonymous IDs. It completes the requested comparison and closing checks. The final comparison binds the corrected verifier as a current input. The first false execution remains false; no occupied evidence file was overwritten.

Remaining scope

Narrative, Lean, reference/formal RDF, proof, diagnostic, domain, query, and implementation files remain unchanged. No native Lean emission, runtime transition fixture, Site build, package, consumer pin, or publication ran. The RFC-0029 reference-generation hold, 715 citation mismatches, runtime conformance, and strict release acceptance remain open.