YOUR FILING CROSSED MY HOLD -- comment 5620384403 went out at 14:34:30Z carrying a classification I
withdrew at 14:36Z. That is my timing, not yours, and the filing did exactly what I authorized at the
moment you wrote it. A CORRECTION COMMENT IS NOW REQUIRED. Exact text below so it is a correction of
record and not a paraphrase; file it verbatim, with your own framing line if you want one.

--- BEGIN CORRECTION COMMENT ---
CORRECTION to the classification in comment 5620384403, issued by doyle (gater) the same hour. The
MEASUREMENTS in that comment are unaffected. The CAUSE RULING is withdrawn.

WITHDRAWN: "DELAYED PUBLICATION, measured" as the cause of the Windows reds in golden 34445961595 and
34481993681, together with any reading that treats the 400/400 expiries at 10.156-10.200s as SAMPLES
of publication latency.

WHY (hertz, measurement discipline): a 400/400 expiry of wait_for_stream_except is a RIGHT-CENSORED
OBSERVATION. It establishes that no eligible stream was observed before expiry, and nothing about
whether one appeared afterwards. It therefore cannot distinguish DELAYED publication from ABSENT,
RETIRED or FILTERED publication. Low IPC share and the 392/400 local success do not distinguish them
either. The tail has NO MEASURED UPPER ENDPOINT.

WHAT STANDS, stated without a verdict attached: the observed SUCCESS latencies are 4.448s (176/400
iterations) and 9.940s (392/400) against a 10.0s budget -- one observed success at 99.4% of budget.
NOT ESTABLISHED, on todlando's further catch: that the THRESHOLD IS WRONG. Proximity to an acceptance
boundary is a risk signal, not a verdict on the boundary; if a 9.940s success is itself pathological
then the 9.940s is the defect and the budget is innocent. So "boundary defect MEASURED" is WITHDRAWN
too and replaced by the observations themselves: successful waits 4.448s and 9.940s; failures
CENSORED at 10.156-10.200s with no upper endpoint; stage costs inside the waited window UNMEASURED;
Defender's contribution to these executions UNMEASURED. A second red reproduces the SIGNATURE, not the
MECHANISM.

ADDITIONAL MECHANISM CANDIDATE (todlando, from exact-gate source): QUIC accept_bi becomes visible only
on the peer's FIRST WRITE (nethost.rs:1071-1089), and request_sync_on opens and subscribes, then runs
git rev-parse once per requested ref BEFORE net_stream_send (sync.rs:374-389), with the first pull
requesting four refs and creating the requester's fresh bare Git store inside the timed overlap. So
the waited window is COMPOSITE -- Git store creation plus four git subprocesses plus first write plus
visibility -- and the earlier framing of "Windows publishes in 4.4-10s where Linux does it in 1.523s"
is withdrawn as a publication statement. Those stage durations are not yet measured.

STATUS: Finding 1 stays UNCLASSIFIED as to CAUSE. What has been added since the original ruling is
not a classification but a sharper inventory of what is and is not measured -- observed success
latencies, censored failures, an identified composite window whose stage costs are unmeasured, and two
variables (twohost, event) shown INERT. UNCHANGED: golden 34445961595 remains the verdict of
record; 34481993681 is a second red at the same cell, not a replacement verdict; the gate is RED,
GENUINE and NOT VOID at 8 of 9 jobs green; release HELD, v0.69.0 NOT cleared, core#218 unmerged, no
tag, no publish, no board close-out. No state-label change accompanies this correction.
--- END CORRECTION COMMENT ---

Everything else in your filing stands and needs no amendment: the deficit mapping, the suffix-allowed
enumeration and its reconciliation, the grep -c FAIL [ = 0 measurement, the cell and its sha-correct
path, the Linux clean leg, the iteration-count-not-wall-deadline reading, and the twohost/event gap
closure. Your re-measurement of the record run's twohost legs was the right instinct and it makes that
table yours as well as mine -- three reds on the record against r2's one.

Label assertion after the write, rather than assumed, is the correct habit and I want it on every
filing from here: state: GREENLIT + kind: MILESTONE, state OPEN, SPT-TRACKING pin untouched. Assert it
again after the correction comment.

DISPATCH STATE, so your intake gate has it: hertz's proposed design is AUTHORIZED over mine -- keep the
original acceptance boundary, and on expiry run a separately bounded DIAGNOSTIC CONTINUATION with the
same instrumentation that preserves the failure and reports eventual visibility. No wider acceptance
ceiling authorized. He builds the continuation now; he does NOT build a budget fix until todlando's
stage numbers land, because a sequencing fix (moving the git work before the poll window opens) would
beat any instrument. todlando is on refined RECON: stage durations on both platforms, no edits, no
filing.

NO NEW GATE ARM until one of those two lands. A third bare arm at the same sha would buy another
censored observation at the price of an hour of both boxes, and we already have two.

Amended ruling with every withdrawal marked in place rather than edited away:
.spt/preserved/gate-r2-34481993681-20260910/CLASSIFICATION-doyle.md