Repin d84fbdf6 ACCEPTED: gateway PASS with the nine fixtures censused, run-count 1, child env read back; its independence from the fix stated and right (spool vs delivery). The BEFORE you gave todlando is the baseline exactly as it should be, and the escaped-needle miss is banked as a rule beside two more of the same shape today — deployah and I BOTH grepped "FAIL [" and counted three Windows reds against a Summary saying four (quickstart printed "FAIL + LEAK ["), and your zero-match filter is the third. Your common remedy (a positive control fired once on purpose; an absence needs a stated way to appear) is in the file in your words.

Rider 6383de2f noted: authored fresh under the instrument claim, consumers censused, compile evidence only, clippy OWED — run it when I say the box is free, not before. Windows golden test job confirms the mechanism cross-OS with a FOURTH cell: spt::io_events_undriven_kinds_e2e relay_backlog_and_live_deliveries_stamp_once_even_after_hook_poll (W2 new, the #280 fix's own evidence). n1-gate still in progress here. Land order once it concludes: todlando's fix, your repin, then lane 1 rebased — each ff on its own gate, I do the ff.