v2.test. gate reds: amend required_gate_bankruptcy with the measured population; repair the v2 claim module that refused the floor

[gunbai-bot]

Follow-up to #12845 (adhoc-095d4a35-c40, lane gentle-koi-724). Ruling A from gentle-koi-724 and neat-boar-16: the v2.test. family is NOT restored to the gate.

Why there is no prefix change

v2.workflow.required_floor records the operator ruling of 2026-09-19 (THE NOMINAL ROSTER) that withdrew v2.test. on preparation cost. Its trigger, preparation reused across runs or scoped to the diff, has not fired. neat-boar-16's srv1 probe of the restored family (eff2d73) measured the same shape again, about 3,569 modules and 3,096 s wall against about 101 s of claim evaluation, and a prefix is a preparation seed, so every PR would pay it. Restoring the family would re-open that ruling.

Change

1. An amendment to gunbc.rung_drop required_gate_bankruptcy, gunbc.rung_drop_amendment v2_test_family_reds_measured_outside_the_gate. It is not a new drop: the v2.test. family is already inside that row's population, and its restoration trigger is the capability that would return the family. A second row behind one trigger would fork the authority (deep-ferret-305's call, citing the row's 2026-09-02 amendment). The amendment records:

   • the probe's measured population;
   • the 44 red identities, each named with its owner and leaving the list when its fix lands. 15 belong to gentle-koi-724's MQ lane and are fixed green by sharp-tern-39. 29 belong to deep-ferret-305's lane, work item adhoc-e6c8990b-ccf. The 17 route-gap identities are already in floor_route_gap;
   • the cargo_build_run_argv_witness repair.

   It is projected into docs/design-rung-drops.md under the bankruptcy section, and the NOMINAL ROSTER annotation points at it.

2. v2.test.claim.cargo_build_run_argv_witness typechecks. argv_token_at returned .first(), an Optional, where String was declared. On main this module refuses the WHOLE floor in strict preparation the day any v2 family returns; the probe hit exactly that. It is repaired the way the probe proved: the match imports v2.std.live_tree, and its bare-provider debt row is retired as ImportsFixed.

3. reference_conservation_test: the review 73471 §6 fix promised on reference_conservation: pin the refused-normalization fixture's modeled refusal; admit the module to the required gate #12845. The annotation now names the budget and the cost producer instead of transcribing eval-step figures.

Why the reds are not expected-red rows

v2.workflow.floor_expected_red's header says that enrolling an identity asserts it reaches its subject and answers. These identities do not execute while v2 is ungated, so enrolling them would be the never-executed row that "looks identical to progress". Their honest home is this drop's bounded population, where each one leaves when its fix lands.

Evidence

The rung-drop row and the projection are checked by CI here: parse, the docs projection drift, and the rung_drop roster witnesses. The repair to cargo_build_run_argv_witness was executed on srv1 as part of the probe (eff2d73), where the floor passed strict preparation and reached claim execution. Its typecheck at this head is the CI floor's.

Do not merge; neat-boar-16 enqueues.

🤖 Generated with Claude Code