Q2 observed autoattach and native fixture, 1205

Retained both actual startup autoattach requests, scopes, sessions, raw replies before pending lookup, and target/context events. The 1135 timeout did not reproduce. Whole actual native fixture PASS preceded one original; original stopped at its first passive health/motion UNKNOWN, before guard/door progress. No later guest input, capture, retry or speculative production change.

Evidence: scratch/runs/20261109-quake2-observed-autoattach-native-1205/. Recovery uses git-base.txt, exact historical Git provider identities, run-input-recovery.patch.gz and zero relevant uncommitted production diff. No source archive, media/module evidence copy, local browser, worker or agent.

Fresh 727-provider inputs authenticate health, authored fixture and outer 1150 observer. Actual fixture and original compiled-consumer receipts verify both flags, health integration, raw-reply observation and timeout-before-delete hook. Eager HEALTH_PROVIDER precedes actual main's ROOT rebind; health wrapper is byte-identical to 1135. Explicit waitMs: 0, first HP-loss/death terminal, independent UP and existing caps remain unchanged. Nine generated startup old-missing/new-evidenced pairs, observer provider/paired startup controls, 21 actual generated command controls, staged-root/tamper fixture and original preflights, composition assertions and locked build PASS. The Chrome-launching quake2-browser-startup-failure.test.cjs was never invoked.

Actual fixture request3 is browser-scope autoattach; request6 uses explicitly attached page session AFA238AF0626C4DAC2C53C9AA19FBEEE. Raw replies match their sessions, remain pending and are not late; Node send-to-reply intervals are 2.296ms and 9991.588ms. Original request3/request6 likewise succeed, at 1.986ms and 4928.408ms; page session 983204549ABCC607BDFDAE3B9DC8A412. Both use autoAttach/flatten true and waitForDebuggerOnStart false. Neither run has a request-timeout event: timeout pending-before-delete and timer overrun are not applicable here. There are zero diagnostic drops/errors/sink failures. These Node intervals do not identify browser service/application latency or explain 1135; its cause remains UNKNOWN.

Fixture comma/period 500ms holds retain wholly-held physical-high samples, release ACKs, post-release zero, capture ACKs and ordinary joined quit ACK2. Browser close is normal exit0, ACK, no signals, 1749ms. Independent retrieval and the whole-fixture gate verify 212 actual descendant reaps, command EOF, separate parent/caller EOF and complete drain before original launch.

Personally reviewed original main menu, Easy menu, loading, first gameplay and short-turn PNGs. Four numbered ACKs complete; gameplay remains pictured HP100. The next 1000ms right turn terminates at 11:43:25.817Z during its action sample: bindings/identity still SNAPSHOT, received motion and damage history UNKNOWN. All 411 valid health-history rows report HP100 without recorded decreases; the last is 11:43:25.783Z. No health-after-UNKNOWN or damage-cause claim. The failed hold retains its independent UP ACK; no fifth numbered ACK or post-terminal capture exists. Independent no-input cancellation immediately joins cleanup: 275 owned/275 actual reaps, command EOF, no remaining/unresolved entries, ports closed, locks free, no profile writers. Original parent wait1, launch transport -9 and separate EOF are retained; browser normal-completion receipt is absent, so normal original close remains UNKNOWN despite its raw Browser.close reply. No ordinary original quit was sent after failure.

Runtime followed Q31200's exact UUID release and API-confirmed completed normal stop. Its historical OWN label was matched by exact UUID/box and API name; initial offline queue-parser failures remain excluded. One observable immutable 25-minute UUID watchdog was armed before create. Box bx_cq8zqp74, UUID 179a8a60-1e58-4425-9a22-15f4fedc2187, is ARCHIVED / PHYSICAL_DRAINED. Initial expiry 12:02:52.623Z; normal stop stop_fad5e3232932c556 completed 11:44:31.856Z, forced false/error null. Final archiveAfter null; expiresAt not returned. Watchdog/lifecycle and board watcher actually wait0 with separate EOF.

Before stop, fixture 79 rows/62 newly stored contents and final root 182 rows/84 additional unique contents pass SHA/size/gzip CRC verification. An initial root retrieval detected a compressed-file hash mismatch: repeated content shared a mutable compression filename. Exact failed bytes/inventory are preserved; final retrieval uses immutable per-path/content names and verifies every row. No guest retry occurred. Logs were compressed before transfer. Free home stays above 1.5GB at recorded checks.

Final SHA/size/CRC reread passes. Own dashboard metadata has no errors, missing references or invalid fields; the global checker retains 22 unrelated historical errors. Initial metadata/seal development failures are retained, not passing tests.

Next: offline diagnosis of the failed received-frame read; no native retry authorized here. Native REVOKED; all-writers/edict ABA UNKNOWN. SingleTaskSTOP.