gnomonlearn
Docs/RFC-0006 Claim wrapper proposal

RFC-0006 Claim wrapper proposal

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.

Status: AI wrapper rule approved for the first release; authored source aligned, runtime validation pending. Owners: semantic core and compiler. Nothing is deployed, so no stored-field migration is required.

Source conflict

RFC-0006 §3 invariant 3 says a Claim<T> wrapper MUST accompany any field that may be produced by an AI actor. §10 repeats MUST for every field modified by an AI actor and names source, confidence, and timestamp preservation. The Lean normative statements b2 and b23 also use must. By contrast, §4.5 says any field produced by an AI actor, extracted from a document, or inferred from rules SHOULD use Claim<T>. The latter sentence weakens the AI case without distinguishing the three producers.

Approved first-release wording: AI-produced or AI-modified fields MUST use Claim<T>. Document extraction and rule inference SHOULD use it unless another normative contract requires the wrapper. This retains the existing AI safety invariant while preserving the advisory rule for other sources. RFC-0006 §4.5 now states the AI MUST and the other-source SHOULD explicitly. Its Stages 0 and 1 are import/preflight states that cannot pass AI publication, and Lean defines M06520 as an error. Runtime compiler and storage enforcement still need validation. The portable tools/check_claim_wrapper_contract.py gate now checks the approved modal split and M06520/M06530 severity across RFC-0006, draft RFC-0055, the Lean diagnostic catalog, the typed diagnostic inventory, and the relevant RDF claims. It is included in check_release.py; it checks specification consistency rather than runtime behavior. Section 4.A now limits M06530 to unannotated JSON during import preflight; the conditional diagnostic relationship in RDF still needs correction. The earlier owner choice of seven Claim fields with conditional requirements remains in force.

Draft RFC-0055 §1 now uses the same modal split. Its unsupported two-minor- version diagnostic remap has been replaced with a first-release alignment contract; M06520 retains responsibility for unwrapped AI fields.

The source-bound review also found projection issues distinct from this wording: semantic-version change claims attach release impact to facet, entity, or Claim nodes rather than the particular schema edit; a declared-kind restriction points only to entity; and a flow-transition removal claim says breaking only when instances occupy the source state while §11’s compatibility table classifies every removal as MAJOR. Each needs its own exact relationship or condition before publication.

Acceptance

The approved rule is aligned in authored §3/§4.5/§10, Lean, the reviewed claim projections, diagnostic definitions, and generated references. Test an AI-created field, AI modification of an existing field, document extraction, and rule inference through type checking and storage serialization. Check no-null handling at the serialization boundary and compare the RDF query outputs before release.