Redline readiness gate passes; request success-exit blocker corrected offline — 2026-11-09
The actual Redline filesystem evidence consumer now waits for production
ready.json, checking failure markers, owning-process birth identity, child
exit and a finite startup deadline before any capture-file fetch. Its new
60-second controller bound stays inside the existing immutable runtime deadline;
startup admission, CDP, input holds and fixture caps are unchanged. First
terminal observation immediately closes further work and invokes independent
no-input SIGTERM through an identity-checked pidfd. Cleanup is sticky and
idempotent. The actual generated fetch branch has oldFAIL/newPASS delayed-ready
coverage plus never-ready, child-exit, production-failure and missing-file
controls, with real child waits and separate EOF.
The 775-provider closure, actual generated fixture/original invocations, corrected-caller preflights, composed cursorOnly oldFAIL/newPASS, focused Redline suites and locked build pass. Prior 1020 fixture proof and accepted unchanged-source references remain SHA/size verified references. The whole current native fixture passes before original: 600ms comma/period holds, wholly-held native highs, release-zero, capture ACK0/1, ordinary quit ACK2, 186 actual reaps, caller/Chrome/watchdog waits and separate EOF.
Fresh original proves the readiness correction: startup capture completes at
10:40:11.228Z, production READY publishes at11.232Z, and controller readiness
admits fetching at11.247Z. The reviewed image is the ordinary Redline launcher.
An ordinary 150ms Play Redline click produces the game title image and actual
ACK0. However, the new guard incorrectly treats the generated request wrapper's
successful SystemExit(0) as a terminal exception. It publishes terminal at
10:41:03.912Z and sends no-input SIGTERM at03.967Z. This is the first blocker,
a controller error, not a game/input defect. No later guest input, capture or
retry occurs; the later local fetch is refused before issuing a fetch. The
retained title capture was made before terminal. There is no reviewed Start Up
game cursor, Training list, Vehicle Training or gameplay consequence.
The success-exit error is corrected offline after that terminal, with no
runtime retry. The actual generated Python request wrapper, using real inert
subprocess exit statuses, reproduces oldFAIL/newPASS for zero exit; nonzero7
still triggers immediate cleanup. No ACK or native pass is synthesized by
these tests. All fetch readiness controls also pass with the final helper.
runtime-pins.json and executed-run-inputs.patch.gz preserve the executed
version; offline-fixed-runtime-pins.json and owned-run-inputs.patch.gz retain
the corrected version. Only readiness.py differs among staged providers.
The final success-exit correction is not native-qualified. Git base
9c9e5bbc2d24f5ddaac4c4cba2d080155bea9e39, closure diff and small run-input
patches reverse-check; no source archive, repeated media or module copy.
Original cleanup has 215 owned tasks and 215 actual reaps, caller1, Chrome0
without signal, separate untruncated EOF, closed ports and absent profile
processes/writer locks. Large logs are compressed before transfer. Cumulative
127 files /110 distinct SHA-size contents /32gzip are independently reread and
CRC-verified before delegated normal stop. Free home stays above1.5GB.
Independent immutable25-minute watchdog is actually joined with separate EOF.
Evidence: scratch/runs/20261109-redline-ready-controller-gameplay-1035/.
Box bx_m7a8vegs, UUID c0b5e7f9-1853-4ed5-83bb-8ae423840d73, is
ARCHIVED, PHYSICAL_DRAINED. Initial expiry 2026-11-09T11:02:03.120Z;
unforced normal stop stop_21cd0c37665566a6 completed
2026-11-09T10:43:42.495Z. Final archiveAfter is null; expiresAt was not
returned. Exact predecessor Q3 UUID33932372 archived/drained stop10dc was
verified by API and board before allocation. Shared clients, custody, Q2/Q3,
host/WAT, HeroesII and Claude are untouched; no agents/workers or deployment.
Next: separately authorized fresh closure with the final success-exit guard, whole actual native fixture before original, then reviewed ordinary Start Up cursor steps <=24 counts per axis, reviewed Training hover and150ms zero-displacement cursorOnly click; review the Training list before Vehicle Training. No Down fallback, absolute locked jump or private-cursor inference from Win32 coordinates. Native lifetime REVOKED; all-writers/edict ABA UNKNOWN. SingleTaskSTOP.