gnomonlearn
Docs/Admission candidate and preserved inconsistency observations

Admission candidate and preserved inconsistency observations

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.

This evidence accompanies FORMAL_MODEL_ADMISSION_PROPOSAL.md and the independently written FormalModelAdmissionCandidate.lean (repository access required). Neither is adopted. The strict known-pin blocker, external formal sources, public RDF/theorem identities, runtime policies, and immutable dependency pin remain unchanged. No site generation, preview build, or deployment is part of this slice.

Preserved actual AX-016 observation

statement-observation/ contains byte-identical copies of the other agent’s owned .work/statement-population-observation-20261013 source, native output, compiler-version log, original text/binary ValidationEvidence, and explanation. The original typed evidence’s absolute paths and argv retain their original execution context. The original compiled .olean and imported compiled framework objects remain at those persistent .work paths; external source bytes were not copied into this repository.

The native Lean subprocess exited 0, deriving False through AX-016 from nonempty-kind Resource values and two structural Statements with equal keys but raw IDs 0 and 1. Its axiom dependencies include Spec.k1 and predicate symbols, with neither Spec.kindNonempty nor sorryAx.

The outer [python3, -] inline wrapper subsequently exited 1. Its offending post-compilation check was:

assert b'[Spec.k1]' in run.stdout and b'kindNonempty' not in run.stdout

The reported original terminal suffix was:

TERMINAL_EXIT_CODE 0
Traceback (most recent call last):
  File "<stdin>", line 31, in <module>
AssertionError

That suffix survives in the original tool transcript, not a persisted raw wrapper log. wrapper-outcome.textproto records this reported historical outcome separately and deliberately has no fabricated combined-log binding or saved-driver claim. The native source/log/evidence had already been written before the wrapper assertion. No recompilation was used to erase this outcome.

Independent candidate compilation

The candidate imports only Lean Init, declares no axioms, and contains no sorry. Its six inspected proof declarations report no axiom dependencies: local nonempty-kind projection and empty-kind rejection, guarded statement uniqueness, parameterized non-overlap, inhabited admitted population, and conflicting-pair joint-admission rejection. It permits raw construction and separate singleton admission for each conflicting row; local validity alone does not supply a population-wide uniqueness certificate.

The first source edit had seven elaboration errors because decide could not infer a Decidable instance through the validity definition. Lean’s error recovery printed a sorryAx dependency for the incomplete witness. That source and log are preserved as failed work, not proof evidence. Its shell wrapper ended with cat and exited 0; the native compiler exit was not captured separately. The corrected source uses an explicit concrete witness and constructive singleton-membership proof. The second direct native invocation exited 0 and all six printed proof dependencies are empty. Final compilation and its typed evidence are recorded by compile_candidate.py with a separately captured native exit code and an output artifact in an owned persistent .work path.

These are small conditional proofs about the candidate certificates and an inhabited example, not consistency certification for the complete external framework or Gnomon model. Caller-supplied temporal-validity and overlap predicates select no interval endpoint, create/upsert, allocation, or diagnostic policy. Runtime preservation and the three actual consumer migrations remain acceptance work described by the unadopted proposal.