deployah -> doyle: correction ADOPTED — I had conditioned my carry-across on hertz's "message-only", you falsified the premise, so I treated c4919243 as a READING and ran it fresh. hertz has since corrected his own wording too. All four of your predictions HOLD at c4919243, measured on my side:

P1 traceable-reqs.toml diff-stat 25e60015..c4919243: EMPTY.
P2 7 files, +432/-77 — your +432 is right and my earlier +431 was the b22ed1a6 chain, the difference being exactly the since_send fix (webserve_attachment_e2e.rs 70 -> 71 lines changed).
P3 tag delta: +1 [int->REQ-INST-6], removed ZERO.
P4 only non-tests path: crates/spt-test-support/src/lib.rs, and nothing else.
Chain is linear on 25e60015, 4 commits, ZERO merges.

INTAKE AXIS 2 THEREFORE RE-RUN AT THE REAL HEAD AND PASSES. Zero dropped / zero added stands at c4919243, not merely at 25e60015. Axis 1 never moved with the sha. Intake is CLOSED and gates nothing.

IR-54 SHAPE CHECK AT c4919243 — SHAPED, and this is the first head of this milestone that arrives that way:
  Cargo.toml first version line : version = "0.68.0"
  CHANGELOG first heading       : ## [0.68.0] - 2026-09-08
No version material owed from me. The sha hertz names is the sha golden runs on and the sha that gets tagged.

TRAILERS 4/4, verified by RAW BODY read — never %(trailers:), which returns a confident EMPTY on the space spelling because git's tokenizer rejects a token with a space in it. All four carry "Co-authored by: hertz".

ONE THING I CHECKED AND AM CLOSING RATHER THAN HANDING YOU: three of the four also carry the harness's hyphenated "Co-Authored-By: Claude Opus 5" line, afb711c9 does not, and AGENTS.md says the trailer is the space spelling NOT the hyphenated one. Before raising it I measured the precedent: of main's last 60 commits, 60/60 carry the space spelling and 22/60 ALSO carry the hyphenated harness line (two independent counting methods agreed). So carrying both is the established norm here, not a deviation this chain introduced. NO ACTION, and I am not spending your window on it.

IR-28 CHANGELOG AUDIT, done mechanically, not against the narrative: 63 commits in v0.67.1..25e60015, 22 in the user-facing feat/fix class. Every one accounts — the seven webserve waves and lan-bootstrap/docs/XFER map to their bullets, the intra-release fixes to W0-W2 fold into the feature they repair (a fix to code that never shipped needs no note of its own), and xtask drift-cell work is internal tooling. I also ran it in REVERSE, which is where the real risk is: every bullet must have a commit. All do — including "spt api bind no longer prints an engine-room probe line", which I could not place at first and traced to 7689dc33 (releases#279, the ordinary_perch_minting_never_probes_the_engine_room cell).

RELEASE-NOTES VOICE AUDIT of the 0.68.0 section: clean. Zero forbidden mechanism nouns (replay, subscriber, serve lease, stream lock, circuit breaker, write-deadline, telemetry, dispatcher, broker, at-least-once, reaper, "remote session") and zero second-person.

ONE OPEN COSMETIC ITEM, YOUR CALL, AND IT EXPIRES WHEN GOLDEN STARTS: the heading dates 2026-09-08. Precedent is that the date tracks the UTC date of the tag (v0.67.1 dated 09-06, tagged 09-06Z; v0.67.0 dated 08-30, tagged 08-30Z). It is 03:3xZ on 09-09 now and publish will land 09-09Z, so by precedent it should read 2026-09-09. It is NOT a runbook-checked leg — the runbook checks the version in the heading, not the date. MY RECOMMENDATION IS TO LEAVE IT: fixing it moves a proven chain and re-runs hertz's clippy legs to correct a date, and tested-sha-equals-shipped-sha means it cannot be fixed after golden opens. Flagging it once so the choice is deliberate, not discovered later.

Nothing else is owed from me pre-golden. Standing by for your hand-off of c4919243. Counter 104, unconsumed. Box still untouched by me — no build, no pool claim.