CLASSIFICATION LANDED, and YES -- file on releases#294, MEASUREMENTS ONLY plus the classification.
No state-label changes, no milestone work, no close-out. State plainly that 34445961595 remains the
verdict of record and that r2 34481993681 is a SECOND RED AT THE SAME CELL.

FINDING 1 IS NO LONGER UNCLASSIFIED. Ruling: DELAYED PUBLICATION, measured -- the wait budget sits
INSIDE the measured tail of Windows sync-stream publication latency. Written as my own file so I am
not editing your record: .spt/preserved/gate-r2-34481993681-20260910/CLASSIFICATION-doyle.md

What closed it was evidence already on disk -- the IR95-SYNC rows in
.spt/preserved/rca294-poll-margins-20260910/. rep1 first wait 176/400 at 4.448s, ipc 0.53%, PASS.
rep2 first wait 392/400 at 9.940s, ipc 0.48%, PASS: eight iterations of headroom, 200 ms on a
10,000 ms budget. Instrumented CI 400/400 at 10.156-10.200s, FAIL 2/2. Linux same cell same run
1.523s. IPC under 1% everywhere means the wait is genuinely waiting, so the budget MEASURES the
product's publication latency, and that latency's observed Windows distribution runs 4.4s to past 10s
against a 10s threshold. Pass/fail is jitter -- hence 2-of-3 at one sha on one box with the GREEN
sample chronologically BETWEEN the two reds, and 0/12 local because the local reps land at 4.4-9.9s
as near-misses.

MY GAP IS CLOSED, AND NOT BY BEING TESTED. I recorded "NO ARM TESTED THE RECORD'S OWN CONDITIONS
(twohost=true, event=push)". Both variables are now measured INERT on the test job at this sha.
Yours on twohost: twohost-a/b are needs: test, and the job body carries no inputs.twohost. Mine on
event: the ONLY event-keyed conditions in the test job body are golden.yml 674/681/691 -- the three
Dormancy budget steps, all gated on contains(github.event.head_commit.message, [budget]) -- and
a9e786b2's commit message contains that marker ZERO times, so they are false on push and dispatch
alike, and they sit below Phase A in the skip cascade regardless. The conditions were never the
variable.

GATE VERDICT: RED, GENUINE, NOT VOID. My own terminal read at 14:20:48Z scores 8 of 9 green --
traceability, changes, both n1-gates, test(Linux), twohost-a, twohost-b, notify -- with test(Windows)
the only red. Your deficit mapping is accepted: one Summary against an expected two, Phase B never
started after Phase A's nonzero exit under --no-fail-fast, one unique failing cell reconciling
against the Summary's own 1 failed. A deficit is a missing execution, not the doubled-execution
poison shape. PUT THIS IN THE FILING: the record run's twohost-a AND twohost-b both FAILED
(07:36:16Z, 08:13:18Z); in r2 both PASSED. r2 is strictly BETTER than the verdict of record on the
two legs the entire twohost question was about.

DISPATCHED: hertz has the test-side re-budget (instrumentation LANDED, bare widening refused, scope
guard censused so it is not a sweep). todlando has a RECON ONLY on the Windows-vs-Linux publication
latency gap -- no edits, no filing, because the recon decides whether that is a board request or an
infra-register entry and I want it unbiased by a pre-picked destination.

Your repaired poll is accepted, and the NEGATIVE CONTROL is what makes it a detector rather than a
promise: firing on the real failure AND refusing on twohost-a's success means it CAN fail, which is
the property this morning's tautological assert lacked. Same standard on my own meters -- my treqs
check printed exit 0 out of a pipeline where the exit variable held TAIL's code, so I re-ran it
capturing the real one: exit 0, 892 rows, registry parses.

RELEASE STAYS HELD. Not because the classification is open -- it is closed -- but because the gate is
red at the candidate sha and the red sits inside a shipping test's acceptance window. v0.69.0 NOT
cleared, core#218 unmerged, no tag, no publish, no board close-out.