RFM accumulator_copy_lens_binds_no_carrier: fixed by #13337, bisected to #12506; 4 residual reds re-measured

[gunbai-bot]

Updates the stale RFM row. It said 5/24 and named #13028 as an unbisected candidate.

• Bisect: the carrier half was introduced by v2 resolve/infer: the dotted-path decision keeps its binding kind, and projection inference stops calling an unretrievable declaration a fieldless receiver #12506 (861c829ac6, the fresh_fold_carrier mint), not N7-1: lambda-binder projection base as lexical reference; dependent-child infer driver types the fold member #13028. The port half came from MQ-1 PR-B: named-argument labels conserved and bound by name (infer + eval through one binding plan) #12382. Both were fixed by accumulator-copy lens: bind the authored carrier and named actuals on lowered folds (5/24 → 20/24) #13337.
• Measured on main d7ac842 (claim_batch, BuildBuddy): 24/28. The module grew from 24 to 28 claims.
• The 4 reds are expecting-red probes in gunbc.explicit_witness_admission, each with a capability dissolution. I re-probed each with uncommitted claims (not in this diff):
  • let_alias_is_proven_suspect: ingest returns Absent on the specimen (grammar, let-then-call).
  • named_step_fold_refuses_accumulator_unread, report_counters_state_the_domain, declared_named_step_fold_copy_is_caught: the named-step specimen ingests, and count_fold_sites = 0 (fold_shape_visible_to_no_lens / call_expression_erased_at_v2_body_lowering).

No code changes. Data row only.

🤖 Generated with Claude Code

[briansrls]

REQUEST_CHANGES at exact requested head ceb7b0b93c0135559019299c11abf9e93241abef — final RFM receipt accuracy only, against DESIGN.md §§4b and 4d. No lens repair, new bisect, changed witness or CI-lane restoration is requested.

The main correction is accepted: distinguish the generated-carrier/role mismatch attributed to #12506 from the named-actual port change attributed to #12382, and record #13337's role-based carrier read and application_binding_plan repair rather than retaining #13028 as the leading candidate. The reported 24/28 measurement is scoped to its named main snapshot. The four residual identities do have existing known-red admissions and capability dissolutions: the let-alias case cannot reach the lens because ingest refuses; the other three concern the erased named-step call/site. Those are not evidence that the repaired inline-carrier regression remains unfixed. This PR changes only the one data row.

[P2] The new final RUNG receipt still conflates two different boundaries

First, it now says the class sat at mitigatable, a clean answer standing where a Poly2Suspect was owed. That is not the previous typed-refusal case. At the classifier boundary described by the new opening receipt, returning clean for the supported planted copy is a false negative. DESIGN §4b places silent wrongness outside the ladder; the receipt names no containment mechanism that would justify calling that answer Mitigatable. Separate the historical classification defect from today's execution/coverage standing. Record the false-negative boundary honestly, or identify an actual independently established mitigation and its boundary rather than carrying the old rung label across the changed diagnosis.

Second, gunbc.rung_drop.accumulator_copy_positive_controls_off_every_lane does not cover the population this receipt attributes to it. At this exact head its declaration.population contains six specific fold-shape controls: three in accumulator_copy_corpus_path_test, plus nested_fold_inside_a_sole_body_fold_is_visible, a_lambda_in_a_non_step_argument_mints_no_carrier and an_inline_step_beside_a_decoy_lambda_is_still_caught. It does not name quadratic_fold_has_suspect, real_branch_body_quadratic_is_caught, or the complete 28-claim module. Its retirement requires those six identities to execute on a live cadence, not every repaired inline-copy discriminator. Consequently the sentence tying the broader population to that drop, followed by NEXT TRIGGER: that drop's restoration, is too broad: that trigger can be satisfied without establishing execution coverage for the other named controls.

Bounded correction: preserve the bisect/repair receipt and the four residual classifications; rewrite only the final standing/trigger to distinguish (a) the historical false-negative defect, (b) the repaired behavior measured under claim_batch, and (c) the precisely scoped executing-lane obligation. Cite the six-identity drop only for its own population. For the remaining discriminator population, name its actual coverage authority/obligation or state the missing coverage and sufficient capability directly; do not silently enlarge an existing drop by citation. This does not require implementing that lane in a data-only correction.

Verified the exact-head witnesses workflow 37345514271 completed successfully. I inspected the one-file diff, exact-head DESIGN ladder, the cited drop's complete population and trigger, the witness source and all four explicit admissions. The bisect and 24/28 remeasurement remain author-run evidence; I did not execute them. The requested head remained unchanged and mergeable before submission. These accuracy corrections do not reopen #13337's accepted implementation.

[gunbai-bot]

Addressed review 5419747804 at fb75139. Only the final receipt changed; the bisect and residual receipts are unchanged.

• (a) The historical defect is now recorded as silent wrongness, a false negative with no containment, which is outside the §4b ladder. It no longer says mitigatable.
• (b) The repaired behaviour is described as author-run claim_batch evidence at d7ac842, not as gated.
• (c) Coverage is scoped by population. The six-control drop is cited only for its six identities. quadratic_fold_has_suspect is enrolled as evidence on the census row accumulator_copy_root_lens. The other inline-copy discriminators are covered only by an identity-only frozen_path_deferrals row, stated as missing coverage. The next trigger names the capability directly: an executing lane that reaches a verdict on every claim in the module.

— sent from sleek-koi-528

[briansrls]

APPROVE at exact requested head cd41ae75dc93e6a902ade08ab1ebfebc2bea40e6, against DESIGN.md §§4b and 4d. Both receipt-accuracy points in review 5419747804 are resolved. No remaining blocker in this data-only re-review.

Historical defect and measured repair are distinct

The final receipt now describes the supported-copy false negative as silent wrongness with no containment mechanism, outside the guarantee ladder rather than Mitigatable. That is the correct classification for a clean answer where Poly2Suspect was owed. It separately identifies the repaired behavior as author-run claim_batch evidence at d7ac84233a, not an executing-lane guarantee.

The previously accepted #12506 carrier / #12382 named-actual diagnosis, #13337 repair and four residual classifications are preserved. The reported 24/28 is still a measurement of its named snapshot, not a new measurement of this head. The four separate expecting-red probes are not presented as a failure of the repaired inline-carrier mechanism or as repaired by this receipt.

Coverage references no longer enlarge their populations

I checked the referenced authorities at this head. gunbc.rung_drop.accumulator_copy_positive_controls_off_every_lane contains exactly its six fold-shape controls; the receipt now expressly denies that retiring it establishes coverage for other identities.

The accumulator_copy_root_lens identity in gunbc.v2_compile_obligation_census::accumulator_copy_obligation names quadratic_fold_has_suspect as a witness and includes retirement of the relevant witness-exclusion frontier in its capability trigger. Naming that evidence is not represented as proof that it executes.

The gunbc.witness_deferral_freeze::frozen_path_deferrals population contains the other cited inline-copy controls, including real_branch_body_quadratic_is_caught. Its identity-grain debt marker is now explicitly described as missing execution coverage, not a cadence or a safety mechanism.

The next trigger therefore correctly names the missing capability directly: a required or scheduled lane, with a declared ceiling, that plans the whole v2.test.long.accumulator_copy_fold_analysis module and reaches a verdict on every claim. Retirement of either neighboring row alone cannot satisfy it. This does not require the four separately owned expected-red probes to be falsely reported as passing.

Evidence and scope

The PR diff against main remains one RFM row. The row's blob at this head is identical to the receipt rewrite at fb75139c144c7780835cb6a356c676cf5b8f9d7e, so the subsequent merges preserved the correction.

Verified requested-SHA workflow 37408994522: floor, generated, emit-build, rust-unit-tests and aggregate witnesses all completed successfully. Those green jobs are not credited as execution of the entire deferred accumulator module. The bisects, 24/28 result and residual re-probes remain author-run evidence; I did not re-execute them.

The requested head remained unchanged and mergeable before submission. No lens modification, new bisect, lane implementation or additional ledger is required for this correction. The broader execution obligation remains outstanding; this approval does not discharge it.