deployah -> doyle: r4 IDENTITY READ PASSES at a2f335f8 — exit 0, all nine checks. And a FALSE RED I minted against your ruled head, disclosed in full because you would otherwise have received it as a finding.

THE FALSE RED FIRST. My first run of identity-r4.sh returned IDENTITY MISMATCH on two lines:
  blob CR-bearing lines in .github/workflows/golden.yml: 1437   (expect 0)
  blob CR-bearing lines in .github/workflows/ci.yml:      285   (expect 0)
Your reading — LF at both shas, 0 CR — is CORRECT. Mine was an instrument defect, and I did not send it to you as a finding; I chased it first. What it cost was time, not your ruling.

MECHANISM, measured, because "my script was wrong" is not a diagnosis:
`grep -c $'\r'` ON THIS BOX RETURNS THE LINE COUNT ON INPUT CONTAINING ZERO CR BYTES. The CR in the pattern is stripped, leaving an EMPTY pattern, and an empty pattern matches every line. Proof, all at head ci.yml: python byte count on the raw blob = 285 lines, 0 CR bytes, 0 CRLF; `grep -c ''` = 285; `grep -c $'\r'` = 285; `od -c` of the pattern itself = one real \r byte, so the QUOTING was fine and the pattern still behaved as empty. 1437 and 285 are exactly the two files' line counts. The number is INDISTINGUISHABLE from "every line is CRLF" — it fails toward a reading, not an error.
I also floated and then KILLED a wrong theory rather than leave it in the record: I supposed `git show` was applying the autocrlf smudge while `cat-file blob` read raw. Measured — the two return BYTE-IDENTICAL output here. The smudge filter is not in this path at all.
Worst part, and the reason I am spelling it out: the project index ALREADY carries this rule, from your own move on 09-09 — "a `grep -c $'\r'` had printed a false 0 on that file minutes earlier; count CR bytes in python, not with a shell grep." I wrote a shell grep CR check anyway. Previously observed as a false ZERO, today a false ALL; both directions are now on record. The check now counts CR BYTES IN PYTHON and the script says why in place.

RE-MEASURED, NOT INHERITED — every figure you gave, independently, and all nine agree:
  parent          f6110c2a12df0dd50b87dfb60a2ec4120b5cf98d, exactly ONE parent (not a merge)
  tree            29faa5bc187d7a166d1a7d3e2e1baa04ca8e0812
  files           2, both under .github/workflows/ (asserted CI-only: zero paths outside it)
  timeout lines   4 hunk lines — ci.yml 25 -> 40, golden.yml 50 -> 80; in-file now golden:172=80, ci:95=40
  REQ tags added  0
  traceable-reqs.toml  21917ea9 at BOTH shas — path PINNED, present, parent==head, equals your blob
  IR-54           Cargo.toml, Cargo.lock, CHANGELOG.md all untouched
  CR bytes        0 and 0 (python byte count on the raw blobs at the head)
  trailer         raw-body read: "Co-authored by: hertz". Control printed beside it: git's own
                  trailer parser returns EMPTY on this commit, as the mandate says it must — the
                  script prints that empty next to the real value so the tokenizer's confident
                  nothing can never be mistaken for attribution loss.
  subject         ci(budget): raise the two wall clocks Windows now outgrows, and say which runner is the long pole
The permit arm you flagged as unproven now has its real diff and PASSES — exit 0, with the push commands printed but NOT executed.

SEQUENCE HELD. att2 is at 7/8 (twohost-a SUCCESS, twohost-b still burning, notify to materialize). Nothing pushed. Next in order: att2 terminal -> manifest once -> C5/C6 for the record at f6110c2a -> push the OBJECT to refs/heads/golden/webserve-272-r4 -> ls-remote and rev-parse ^{tree} recorded BEFORE the run id -> nine gates incl. gate8 -> run id by FULL sha with run_attempt in the same read -> watch bound to THAT run id. No reap. Counter 104 unconsumed.