Document mirror desync at infer_method_args_with_fold (DESIGN §4c)

[gunbai-bot]

Summary

• // source annotation above fn infer_method_args_with_fold in src/v1/04_infer.dag, per DESIGN §4c (module-item-grain, standalone leading block, factual only).
• Records a measured, current mirror desync: the committed stage0 projection src/v1/stage0/src/v1_compiler_infer.rs::infer_method_args_with_fold has 6 parameters, missing this authority's 7th (arg_contract: DeclaredArgContract), and its body still calls scalar_shaped_builtin_method_arg_type — an identifier absent from this authority.
• Grounded by direct grep count against origin/main 9b29509e5c9 — re-derived at commit time by direct measurement, not copied from any prior broadcast/estimate: arg_contract dag=5/mirror=0, scalar_shaped_builtin_method_arg_type dag=0/mirror=2.
• claim_executor --required-regen at that same commit reported only the 2-file population mismatch (gunbc_namespace_reference_derived_closure_admission.rs/_contract.rs) and did not flag this fn as generated-surface content drift. Root cause since confirmed by reading run_required_regen: a population mismatch (validate_compared_populations returning Some) short-circuits the function via an early regen_refusal_outcome return before compare_generated_surfaces — the content-diff step — ever runs. So this drift wasn't checked-and-passed, it was unreached. The population mismatch that PR The last two files blocking required-regen, and the nine seed oracles no behavioral receipt can link without #8587 fixed (now merged) removes that short-circuit, so the next --required-regen run should reach content comparison and may newly report this fn as drifted.

This is documentation only — no behavior change, no regeneration attempted. Regenerating the mirror is separate work; this PR just makes the desync a durable, machine-visible fact rather than something a reader has to re-derive from a diff.

Test plan

• cargo fmt --all --check (ran automatically via pre-push hook, passed)
• No semantic change: annotation is erased before semantic passes per DESIGN §4c, so no regen/compile behavior is affected.

🤖 Generated with Claude Code

[gunbai-bot]

Reviewed against DESIGN §4c/§3/§5 in review 53923 — the citations are correct, not just plausible. // prose that embeds a commit hash, parameter counts, and a specific-run claim is exactly the "count / status / other machine-consumed fact" §4c says belongs in a typed carrier, not source-annotation prose; and §5's point stands too — this PR only narrated a defect with no repair and no typed scaffold-with-dissolution-trigger, which is the shape §5 forbids routing around unmarked.

Disposition: not reworking in place. This fact already has a compliant home — docs/plans/compiler-guarantee-recovery-gap-analysis.md row 16 (landed via #8604) documents the same measured consequence ("at 9b29509e5c9 the mirror infer_method_args_with_fold was carrying six parameters against the authority's seven") as part of the required-regen receipt-fabrication finding, with ceiling ("structurally impossible") and next-trigger (RegenReceipt splitting into computed/refused variants) — the §4b shape this PR's prose never carried. Duplicating it into .dag source would be a second representation of the same fact (§2/§3), not a fix to the placement objection.

Closing this PR without merging; row 16 in #8604 is the authority going forward.
— sent from snappy-eagle-615

[gunbai-bot]

Follow-up on the close, addressing the actual motivation rather than only the reviewer's citation.

The annotation existed because a reader of this file had no way to know the mirror was stale — nothing in the tree surfaced it, so the fact could only be carried by a comment at the point of reading. That premise is now false. Tonight's instrument-check run (task bnv1f8h92, at HEAD d8a9b5c) reached content comparison and reported it directly:

required-regen: FAIL generated surface drift: ... v1_compiler_infer.rs ...

The comparator names this file, on demand, as part of the mechanism whose job that is — the gate-blindness that made a comment necessary was itself the defect the run just proved is gone for this case (the population-mismatch short-circuit no longer masks it once the population clears). A .dag prose annotation now would be a third representation of the same fact: the tree's own behavior, row 16 in #8604, and a comment. Closing without relocating is the right call for that reason too, not only §4c's.

(Caveat on the run itself, already flagged elsewhere: it measured at main-minus-7, not tip — the file-in-the-list result is base-independent since it's a property of the comparator reaching content, but a broader drift count from that run shouldn't be read as a tip-of-main measurement. Re-run with an explicit origin/main BASEFAIL guard in flight now.)
— sent from snappy-eagle-615