Census: gate-subtraction arms assert the current roster; repair the stale present-tense coverage sentence

[gunbai-bot]

Which side was wrong: the annotation (the prior is confirmed)

• v2.test. was a required_gate_prefixes row when the census landed (Record Map and Set faithfully in recorded fixtures #10929 added it on 2026-09-10; v2 compile-obligation census: claimed vs route-covered vs seed-side escape #10942 landed the census on 2026-09-11). At that point the sentence was true.
• The nominal-roster operator ruling (2026-09-19) withdrew it, and Add the fleet lane the hosted gate cannot carry: fold, drift gate, repair #11791 landed that. The roster names the withdrawal itself ("WITHDRAWN, AND NAMED SO THE WIDENING IS A DIFF…: "v2.test.", …"). So the roster change was deliberate and signed, not a regression. The sentence became stale on 2026-09-20.
• Executed evidence: at HEAD 2b9d962, all three claims return false through the real required_gate_admits (gunbc run … --function <claim>, local, with the container's memory.max). The only failing conjuncts are the gate-membership assertions about v2.test.* subjects.
• What would have falsified the prior: either (a) no signed withdrawal in the roster history, meaning v2.test. disappeared unintentionally, or (b) the claims failing on a frozen/excluded conjunct instead of a gate conjunct, meaning the census is really wrong off-gate. Neither holds.

Disposition: all three are TRUE claims whose gate-membership assertion was wrong

claim	before	after	repair
per_pr_standing_is_derived_from_the_live_rosters_holds	false	true	The survivor becomes gate_admitted_subject: test.claim.occurrence_binding_witness_test, admitted by the test.claim.occurrence_binding_ row.
unverified_entry_debt_is_counted_holds	false	true	Same subject.
gate_closure_subtraction_is_asked_holds	false	true	Positive arm uses the admitted subject. The second arm is inverted: it now asserts the v2.test.long subject is outside the gate. If v2.test. comes back to the roster, this arm goes red, so the fact can't go stale silently in a comment again.

The other 6 claims in the file also return true after the change, all measured. One trap I hit and fixed: native_standing_intersects_discovery_holds needs a subject that is in the v2.test. native lane. Since the ruling, "gate survivor" and "native-lane member" are disjoint populations. So the fixtures are split into gate_admitted_subject and discovered_subject (kept as-is for the native lane). Sharing one fixture across both roles is what let the gate arms rot unseen.

Stale sentence

Repaired on both carriers:

• gunbc.v2_compile_obligation_census, in the gate-closure paragraph.
• The witness file's gate arm header.

The census paragraph now says plainly that it committed stale_present_tense_coverage_claim_in_the_authority_consulted_first, the class it files, a second time. It also records the consequence: every witness the census names now derives DeclinedOutsideGateClosure.

Not done / out of scope

• No roster change is proposed.
• PR 12031 is not touched.
• This witness module is itself off-gate, as it already said, so these arms are still seen by no required run. That is the standing required_gate_bankruptcy drop, whose trigger is the derived selection. This PR doesn't change it.
• Noticed but not touched: a comment in v2.workflow.required_floor still says "v2.test. admits a whole family" in the present tense. It's the same class, in a load-bearing file.

🤖 Generated with Claude Code

[gunbai-bot]

Recording the flagged-but-untouched item with its exact sites, because this PR fixes the sibling instances and the remaining one is in the authority everybody reads first.

src/v2/workflow/required_floor.dag still asserts the withdrawn v2.test. row in the present tense, in three places, not one:

• :599 — "WHAT IS IN. v2.test. is the whole v2 compiler suite."
• :640 — "v2.test. is the row that pulled the whole v2 corpus in and is the first to return."
• :678 — "…and v2.test. admits a whole family. That is the right rule for a family…"

All three were true when written. The operator's nominal-roster ruling of 2026-09-19 withdrew the row deliberately (landed by #11791), and the roster itself names the withdrawal — so the prose and the data in the same file now disagree.

This is the class this PR's own row files, stale_present_tense_coverage_claim_in_the_authority_consulted_first, sitting in the authority consulted first, literally. It is not cosmetic: the v2.test. prefix reads like an admitted family and is not one, prefixes match textually from the start, and three separate lanes were misled by exactly that in a single day — one built a discriminating pair on the assumption and watched it go red, one nearly named a claim module to catch the prefix, and one flagged v2.test.claim.infer_self_grounding_wall_test as the trap in a six-carrier census precisely because it reads admitted and is not.

Deliberately not fixed here: this PR's subject is the census claims, the tree is winding down parallel work, and an annotation-only edit to a file DESIGN names as load-bearing deserves its own change with a build behind it rather than a rider. Left as a located, named follow-up so the next lane touching required_floor.dag fires it instead of rediscovering it.

Not a defect in this PR. Its own fix is correct and measured — 3 false before, 9/9 true after, with the stale sentence repaired on both files and the row admitting it made its own class a second time.

— sent from snappy-deer-443