DISPATCH, test-side, yours per the dispatch split. Finding 1 on releases#294 is CLASSIFIED, and the
classification is what makes this a test-side job rather than a product one.

RULING: DELAYED PUBLICATION, measured. The wait budget sits INSIDE the measured tail of Windows
sync-stream publication latency. YOUR OWN instrumented reps are what closed it:

  rep1 first wait   iterations=176/400   elapsed 4.448s   ipc 23.6ms (0.53%)   PASS
  rep2 first wait   iterations=392/400   elapsed 9.940s   ipc 48.2ms (0.48%)   PASS (cell 76.558s, SLOW)
  instrumented CI   iterations=400/400   elapsed 10.156-10.200s                FAIL 2/2
  record red + r2 gate                                                         FAIL at sync.rs:109
  same cell, Linux, same r2 run                                                PASS 1.523s

rep2 passed with EIGHT iterations of headroom: 200 ms on a 10,000 ms budget, 2%. IPC is under 1% of
elapsed in every sample, so the wait is not burning budget on RPC overhead -- it is genuinely
waiting, which makes the budget a MEASURING INSTRUMENT for the product's latency. Observed Windows
distribution 4.4s / 9.9s / >10s / >10s against a 10s threshold. Pass/fail is jitter, which is why it
went 2-of-3 at ONE sha on ONE box with the GREEN sample sitting chronologically BETWEEN the two reds.

THE WORK: re-budget wait_for_stream_except (crates/spt-daemon/tests/sync.rs:98-109; callers :201 and
:566) so the budget is OUTSIDE the measured tail, and LAND the instrumentation rather than patching
it in per-diagnosis -- the wait must report iterations/elapsed/ipc/sleep on BOTH outcomes, not only
on failure.

A BARE WIDENING IS REFUSED, and I am amending my own standing ruling rather than quietly dropping it.
The original ground for "no budget widened" was that widening may mask an UNMEASURED delay. The delay
is now measured, so that ground is spent. The amended ground: widening ALONE converts a loud red into
a silent slow pass and discards the latency signal -- exactly what rep2's 2% margin would have
hidden. Wanted shape: generous iteration budget, instrumented on both outcomes, plus a DOCUMENTED
latency ceiling well above the measured tail that still fails loudly, so a real regression in
publication latency reds instead of hiding. The comment at sync.rs:99 needs correcting too: it calls
this a "10s budget" when the real bound is 400 x (net_streams RPC + 25 ms) -- the sleep sum is only
the budget if the RPC is free. Say what it actually bounds.

SCOPE GUARD, censused at a9e786b2 so this does not become a sweep: the fixed-iteration wait shape is
house-wide in crates/spt-daemon/tests -- access.rs:80, attach.rs:114, broker.rs:618 at 400, plus
about twenty more between 100 and 2000 iterations. DO NOT rewrite them all. Fix the site with a
MEASURED tail problem, leave the rest, and say so in the commit so the next reader does not mistake
restraint for an oversight.

Thin lane, test-only, no product files. The box is free: the gate run went terminal 14:13:57Z, 8 of
9 green, test(Windows) the only red. Full ruling with provenance for every figure:
.spt/preserved/gate-r2-34481993681-20260910/CLASSIFICATION-doyle.md

Release stays HELD, v0.69.0 uncleared, core#218 unmerged. 34445961595 remains the verdict of record;
r2 34481993681 is a second red at the same cell.