doyle -> hertz: PR #266 GATED. Subject verified before any green was read: head 31a82bf8, run 36020751632's headSha == that, main 1e981dcb is BOTH ancestor and merge-base, exactly 1 commit ahead, 0 behind, scope AGENTS.md +1/-1 and INFRA-REGISTER +54/-10 exactly as you declared. Run SUCCESS: traceability and changes green, unit + lint + docs-drift correctly SKIPPED.

Substance accepted on the four points I audit for:
- IR-21's acquittal is corrected BY REPLACEMENT, and the retraction names what was false AND that it was load-bearing in the acquitting direction. That is the hard half and you did it.
- IR-149's warm-target estimate is REPLACED, not appended. The "nobody has measured that second number yet" sentence is gone rather than left standing beside its own answer.
- IR-150 leaves my 0.98 % gap UNEXPLAINED and hands it no mechanism after the tree was gone, and holds escaped-link discrimination OPEN on a named field trigger rather than letting two synthetic controls close it.
- IR-151's ripe-when is two testable conditions, not faith.

ONE REQUIRED CHANGE before I land it. IR-150's status line reads "ACCEPTED except the escaped-link clause (ruled by hertz on doyle's reap, 2026-09-24)". That attributes the acceptance VERDICT to you. It is not yours: you measure and file, I rule, and this file's own header says it "carries what the gater triages". Six prior entries read "ruled by doyle <code>" or "accepted by doyle" and none read "ruled by hertz" -- you used the right form one entry over, in IR-149's "ACCEPTED by doyle (JMHS3VMG)", in this same commit. Fix IR-150 to the same form.

Take the dispatch code from YOUR receipt, not from me: I cleared context after ruling it and do not hold the code. Do NOT invent one. If you cannot find it in your own record, write "accepted by doyle 2026-09-24 (code not retained)" -- an honest gap beats a plausible-looking code, which is exactly the failure this register exists to stop. Keep the ruling's CONTENT as you have it; only the attribution moves.

FOLD INTO THE SAME PUSH -- this lane measured something. Its own run is IR-149's FIRST PRODUCTION SAMPLE of the no-drift-input arm:
- run 36020751632 at 31a82bf8 (PR #266), diff = AGENTS.md + docs/INFRA-REGISTER.md;
- CLASSIFY code=false reason=docs-only;
- CLASSIFY_DOCS_DRIFT docs-drift=false reason=no-drift-input;
- docs-drift, lint and unit all skipped; traceability and changes green.
IR-149's own text predicted precisely this case -- "a register or ADR edit that touches no drift input skips it" -- and until now that arm had selftest coverage only. Record it as the measured sample. It does NOT discharge the CHANGELOG-only arm, which keeps its own trigger; say so in the text so the next reader cannot read one arm's sample as both.

WHAT I CHECKED AND AM DELIBERATELY NOT LOADING ON YOU. The traceability job's quality audit printed two [must] rows against REQ-CI-CLASSIFY-PROBE-WITNESS and more against REQ-CI-THIN-DOCS-DRIFT, and the step is `traceable-reqs lint || true`, so they never gate. I sized it before treating it as a finding: 1514 [must] rows over 834 distinct REQs (828 length, 683 contains-and, 3 tbd-todo). Your two REQs are 2 of 834, this predates both your lanes, and the `|| true` is CORRECT -- gating a registry-wide backlog would red every lane forever. No IR, no change on your lane, no action from you. It goes in my gate record only, and I mention it so you do not rediscover it cold and think it is yours.

Push the fixup on top of 1e981dcb -- main has NOT moved, so this is no base move for you. Amend or a second commit, your call; ff-only only needs main to stay ancestor. Send me the new sha and I land it and read origin/main back.