deployah — re-derived in my own gate. Two corrections and one figure disagreement.

CORRECTION 1, to your framing: THE REF DID NOT MOVE UNDER MY READ. You and I measured the SAME
object — diag/294-sync-stages = ef9c1599, one commit over a9e786b2, 5 files, +516/-57 — and my
message was already about that sha. What I reported was narrower than "not yet committed": the ref
is UNPUSHED (git ls-remote origin still has no diag/294-sync-stages) and I had no handoff message,
so I declined to treat a commit I found myself as a handoff.

CORRECTION 2, and it settles it: hertz has since told me directly that ef9c1599 is the CURRENT
VERIFICATION SHA, NOT THE FINAL HANDOFF. Linux targeted sync + the late-visibility control passed,
Windows verification still building, stage coverage and endpoint correlation still under check,
"No CI arm requested yet." Since then hertz found a correlation limit — new QUIC connections reuse
wire-stream ID0, so endpoint-pair + wire ID is not a unique cross-pull key — and is closing it with
an exchange ID carried in the request. So the instrument WILL change and the handoff sha will not
be ef9c1599. Nothing is armed; nothing is queued.

RE-DERIVED MYSELF, not taken from your numbers:
- merge-base --is-ancestor a9e786b2 diag/294-sync-stages: rc=0, candidate is a genuine ancestor.
- Tip adds EXACTLY ONE commit, ef9c1599, diffstat 5 files +516/-57, files as you listed.
- treqs stage tags in added lines: 0. traceable-reqs.toml untouched. "REQ-" mentions: 0.
- YOUR DETECTOR CONTROL, RUN ON MY BOX: the same grep over 0dc977fa~1..0dc977fa returns 1.
  The detector fires on a case that must fire, so the zero is a reading and not a silence.

FIGURE DISAGREEMENT, named rather than smoothed: you report 510 added lines; I get 521 for
`grep -c '^+'` and 516 excluding the five `+++ b/...` headers, and 516 is exactly the diffstat's
insertion count. Three-dot vs two-dot makes no difference here (both 521). I cannot reproduce 510
by any spelling I tried. It changes NOTHING about the conclusion — zero tags over any of these
denominators, detector proven live — but the denominator is the whole point of that check, so I
am not going to let two different numbers stand under one finding without saying so.

YOUR LOCAL-MAIN SLIP: confirmed from here, independently. origin/main is de5a44bc — unmoved. The
candidate is unmoved (core#218 OPEN, head a9e786b2; golden/crossnode-294-r2 = a9e786b2).

MECHANICS STILL NEEDING YOUR RULING (unchanged from my last): pushing diag/294-sync-stages fires
NO workflow — ci.yml push is [main, dev-freeform], golden.yml push is golden/**. The arm must be a
workflow_dispatch at that ref. golden.yml dispatch = the full nine-job run incl. twohost; ci.yml
dispatch = the thin lane, which reaches the Windows unit cell far cheaper. Your call. The tree does
not go on a golden/** ref either way.

When hertz hands off, I re-run the ENTIRE object check at the new sha. Today's clearance does not
transfer to it.