feat(pm): H35 — flag a gate-label removal with no dual-carrier clear behind it (#12407)

H31 compares the contract-review gate's two carriers as they STAND; its own
header says what it cannot do — 「闸门被剥不是红灯是放行」, and 「被剥」 and
「从未挂过」 are indistinguishable in the state. H35 reads the EVENT that
produced the state, which is the first reader of history in this file.

Transport: `GET /repos/{repo}/issues/events` is a repo-wide, newest-first
stream carrying the full issue payload, so this costs ZERO per-card timeline
fetches — the trade H15 and H16 decline by name. One time-bounded window
(12h = two patrol cycles) with a 30-page quota backstop that reports when it
binds.

Evidence is STRUCTURAL, not prose. The protocol's 「PASS 双载体同笔清标」 makes a
legitimate clear two removals seconds apart, one per carrier; a strip leaves
one. Measured (160 pages, 16,000 events, 3.41 days): the same-stroke cluster
ends at 101s and the next observation is 275s, so the 120s threshold sits in an
empty region. A PASS-comment predicate was measured and REJECTED — the verdicts
are free prose (three spellings live in one 18h window), a strict marker matched
5 of 35 removals and an "any PASS" reading matched 26, which is a check that can
barely fail.

Three outcomes, never two (#4690): `half-write` (hung dual, cleared lone) is the
finding; `unjudgeable` (hung lone, cleared lone — a single-carrier gate) is
reported AS unjudged, because 34 of 36 lone removals in the corpus were gates
the PR side never carried and flagging them would report the board's majority
shape as a violation; `undated` declines. Report-only throughout.


Claude-Session: https://claude.ai/code/session_01JANH3y7qe3MD8aLaLXci8N

Co-authored-by: Claude <noreply@anthropic.com>

Author: claude[bot]