gunbc-private CI: every module-scope annotation refuses; establish the rule before editing

[briansrls]

Summary

fail_closed_gate_refuses_its_own_repair describes a gate that predicates on base-revision health and so refuses the very change that restores it. In namespace_wave_admission, that fired on annotation grain: a base blob whose only diagnostics were SourceAnnotationRefused was declared unreadable, so a transition that merely moves // onto the declaration the grain admits could never be adjudicated.

The prior repair worked around it by substitution: is_annotation_grain_repair compared the base and head sources with standalone // lines stripped, and if the remainders matched, copied the head records in as the base side. That is a reconstructed baseline defended by an argument, and the arm sat below the class's ceiling.

This change replaces it with construction. base_records now asks the real question — did the parser produce a module? Annotation bind runs after parse, so a file whose diagnostics are all SourceAnnotationRefused and which produced a module has a genuinely observable baseline, and it is parsed rather than inferred. Everything else keeps refusing: any other diagnostic, or a file that produced no module, stays unobservable and returns Err.

is_annotation_grain_repair and its helper annotation_erased_lines are deleted, and the substitution arm goes with them — Err(reason) => return Ok(WaveAdmissionOutcome::NotEvaluated { reason }) is now the whole failure arm. §4b(4): the climb deletes the lower-rung production handling, and the refusal it stood in front of remains. The narrowing is real and is the point — the old arm admitted any file whose annotation-erased remainders happened to match, including one the parser never read; the new one admits only files the parser actually read.

Also on this head: the roster's single TransitionAdmission row — gunbc#10945 mutation_status_is_commit_ambiguous stranded-caller repair — is deleted. #10945 merged, required floor run 34517633122 reported it CONSUMED ... already satisfied at the base, and the rule charges the deletion to the next change touching the roster. This change edits that file, so it is the toucher; the twentieth dissolution is recorded in place. The resting state is empty again, and empty is not permissive.

The v1 PURPOSE-admission question

The review's second finding is advisory and is not a defect, and it is answered here rather than acted on: hand-written Rust expanded in the v1 seed is a PURPOSE-admission question under gunbc.v1_maintenance_standing v1_seed_standing — whether the change serves the v2 self-host program — not a correctness one.

This lands on the admissible side, and the reason is what the Rust does rather than how much of it there is. The v1 standing is semantics frozen, maintenance active, closed to growth for its own sake. This change is net deletion in the seed (two functions and a fallback arm out, one predicate in), it adds no v1 capability and no new v1 growth surface, and its subject is the required-CI wall itself — the instrument the v2 program is adjudicated by. A wall that refuses its own repair blocks migration changes specifically, so repairing it is reachability for the v2 program, not investment in v1.

Scope is deliberately not expanded to satisfy the finding. Migrating evaluate_wave_admission to .dag is the real climb and is a different change against a different authority; doing it inside this one would make the exception look bought by the fix.

Test plan

• cargo fmt --all --check — clean (pre-commit and pre-push hooks).
• Required floor and build lanes on CI. The prior red on dd8b394 was the consumed-admission refusal described above — the build lane was green on that same head — and it is fixed by this push.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UwLGmTkGxdNTHZo3qpZnAg

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-10T18:09:27.818864Z	bb931e1	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.