Execute the deleted-lane countable on the required floor

[gunbai-bot]

Finding

Run 33889629622's uploaded required_floor_disposition.tsv shows that all 12 declarations in witness_floor_workflow_consolidation_witness_test were discovered and all 12 were declined_outside_gate_closure. The apparent 8/4 split in the log is the floor-cost-debt split: eight identities are already declared debt, while four—including w_RED_the_deleted_lanes_do_not_return—were neither debt-held nor recurring executions.

The cause is the module's absence from required_gate_prefixes, not candidate-list construction.

Change

• admit the exact workflow-consolidation module to the required floor
• pin the three rung drops' cited identity as Planned in the discovery-census witness
• retain the existing eight cost-debt dispositions; the four non-debt witnesses now execute
• narrow two newly executing rows from whole-workflow YAML generation to the typed job/roster producers they judge, keeping them inside the per-claim CPU ceiling

This is a partial climb under gunbc.rung_drop.required_gate_bankruptcy, following its 2026-08-31 generated-artifact precedent. It restores four identities in one exact module and does not retire that drop: the restoration trigger remains a required run whose population is derived from the change over the whole corpus.

Price

The first admission receipt (run 33899564106) resolved 2,339 modules versus 2,209 in adjacent predecessor run 33889629622 (+130). Strict preparation was 635,645 ms versus 630,511 ms (+5,134 ms); this is an adjacent-run observation, not a causal estimate. Two whole-workflow assertions were right-censored above the 500 ms ceiling, prompting the typed-producer narrowing.

Green run 33907907019 then planned and passed all four newly recurring identities at 1 ms, 59 ms, 457 ms, and 451 ms CPU. The exact cited row w_RED_the_deleted_lanes_do_not_return passed at 451 ms.

Severity boundary

No other witness found in the corpus covers all three deleted jobs or the full FABRIC_EVIDENCE/calibration spelling set. fabric_evidence_gating and emit_copy_qualification_without_a_consumer therefore cited an uncovered capability. required_lane_claim_agreement_witness_test independently covers only the rust-unit-tests roster absence, so rust_unit_tests_off_the_merge_path has partial independent coverage but still cites the wrong artifact for the broader claim.

Verification

• git diff --check
• full 4,835-file DAG parse sweep passed remotely
• run 33907907019: build, floor, generated artifacts, and aggregate witnesses all passed; floor verdict FloorClean, adjudication PASSED, blockers 0
• disposition artifact: all four newly recurring identities planned and passed
• per-claim cost artifact: 1/59/457/451 ms CPU

Executed RED control

The countable is authorable, not merely planned. In falsification run 33924210916 (temporary PR #10476), removing the negation around the injected ["fabric-evidence"] control made test.claim.witness_floor_workflow_consolidation_witness_test.w_RED_the_deleted_lanes_do_not_return return Bool(false) in 480,283 ns; adjudication recorded cause=claim_failed, and the floor reported exactly one failed claim. In production run 33921261763 on head 706a721, the same identity was planned and passed. Together these runs distinguish a live absence wall from a vacuous always-empty projection.

Scope and permanence

This change retires none of the three citing drops and does not alter their capability-grain restoration triggers. It makes their shared countable real: a deleted lane returning while those rows still stand now makes the required floor fail. Under DESIGN §4b(4), the now-observed RED is a permanent regression control and must remain enrolled for as long as any of the three drops stands.