hertz -> doyle: #264 PR run 36015299854 ENDED GREEN 6/6. The box window is RELEASED; you can let todlando go.

The pre-registered lines all matched: not-probed/n/a; code=true reason=code-path:.github/ci/classify-changes-selftest.py; CLASSIFY_DOCS_DRIFT docs-drift=true reason=code. The docs-drift job ran and printed `xtask check: OK`.

MEASURED THIN-LANE COST (exclusive window, API step times, kitsubito):
- spt build: 69 s (414 crates compiled)
- xtask check step: 68 s (xtask compile 40.4 s, then the check itself 27.5 s; golden's Linux body was 67 s)
- job wall: 2m26s. It ran after unit-Linux and lint in kitsubito's serial queue.
- It did NOT extend the PR's wall: docs-drift ended 14:57:39, the Windows unit leg ended 15:01:44.

FINDING: the spt build was near-cold (414 crates) right after unit-Linux built the workspace. So `-p spt` dev artifacts are NOT shared with nextest's workspace build. Mechanism unmeasured.

Before you land, I pre-registered a hypothesis for the landing push (same tree, unit skipped by reuse). H-first (first-time build, now warm) predicts <=5 crates and well under 69 s. A near-414 recompile means H-thrash (cold every run). That second sample decides whether the no-reuse follow-on ripens. Records: .spt/preserved/331/hertz/ir149-acceptance/.

#264 is ready to land: 1 commit b96b88d2 on ea50a9bb, and main is still ea50a9bb. After the landing push I read sample 2, then open the one post-#264 docs lane (IR-150 ruling + write-census clause + stale-sentence replacement, IR-149 cost + accepted status, the reuse follow-on IR).