required_floor_gate_selector: assert the current gate, not the withdrawn v2.test. family

[gunbai-bot]

Repairs the latent red in v2.test.required_floor_gate_selector (hit on #12461 at b1cfc75, run 36366239387). Two claims assumed the v2.test. family prefix, which v2.workflow.required_floor required_gate_prefixes no longer carries.

• the_match_names_the_kind_that_admitted: the family arm now matches test.claim.namespace_resolution_witness and requires GateModuleFamily { prefix: "test.claim.namespace_" }. The subject is unchanged: an authored-module admission and a family admission are reported as different selector kinds.
• a_family_row_still_admits_its_members_textually (sibling): v2.test.claim.ordinary_witness is replaced with test.claim.infer_ordinary_witness, which the test.claim.infer_ row admits. It still goes through required_floor_site_disposition.
• The module comment no longer says this module is in the v2.test. family.

The other two siblings made no withdrawn-prefix assumption and are unchanged.

Evidence

Local claim_batch --source-root dag --source-root src/v2 --entry src/v2/test/required_floor_gate_selector_test.dag --functions <all 4>. The binary was built from this branch.

claim	before	after	namespace_ row removed	infer_ row removed	runner_capacity_plan_witness row removed	eval_steps (after)
the_match_names_the_kind_that_admitted	FAIL	PASS	FAIL	PASS	FAIL	328
a_family_row_still_admits_its_members_textually	FAIL	PASS	FAIL	FAIL	PASS	9,873
an_authored_module_row_admits_the_module_it_names	PASS	PASS	PASS	PASS	FAIL	10,158
an_authored_module_row_does_not_admit_its_suffix_neighbour	PASS	PASS	PASS	PASS	FAIL	17,156

Each claim goes red when any row it names is removed. All four are well under the 72.3k new-witness budget.

🤖 Generated with Claude Code

[briansrls]

APPROVE-MERGE — content at 6693eb8

Inspected the complete one-file patch, all four claims and the current required_gate_prefixes authority at this SHA. No blocking source finding.

The two replacement subjects are admitted by live family rows (test.claim.namespace_ and test.claim.infer_). the_match_names_the_kind_that_admitted still rejects an unmatched result or the wrong selector kind and checks the actual admitting prefix; its authored-module arm is unchanged. a_family_row_still_admits_its_members_textually still goes through required_floor_site_disposition, whose helper accepts only Planned, not changed-selection or any decline. The suffix-neighbour negatives and contained-child positive are retained. No claim is removed, converted to a tautology, or moved to a weaker helper.

The change does not restore the withdrawn v2.test. family or claim that this module is ordinarily selected by it. The reported row-removal controls are suitable discriminators for the fixtures being asserted; the reported eval-step observations are well below the stated budget. I did not independently execute those experiments.

Exact-head PR workflow 36413686714 completed/success. This review is source inspection plus checked CI status, not a locally executed claim_batch receipt.

Landing caveat: GitHub currently reports mergeable=false. Content is approved, but reconcile main and report/rebind the resulting SHA before treating it as queue-ready; retain normal exact-head/merge-queue gates. Do not broaden the production gate to repair these fixtures.

No merge, enqueue, workflow dispatch or runner mutation performed.