Redline interrupted native-action recovery — 2026-11-08
Recovery ended at the ordinary Start Up menu with New Game selected. No firing, ammo decrement, vehicle entry, driving, or first-person gameplay consequence was demonstrated by this run. The retained startup-evidence failure marker made further gameplay inadmissible under the explicit recovery instruction. This is a recovery blocker, not a demonstrated emulator input defect.
Evidence: scratch/runs/20261108-redline-repaired-native-action/ (ignored). The interrupted launcher handle77169 exited143 without a final/result; its Codex PID3047760 was subsequently confirmed absent. Cause remains unknown. The coordinator reported no input or replay. This fresh worker explicitly took ownership on the messageboard at 02:12:01Z and did not rewrite the attempt as a clean completion.
The live audit recovered all original numbered commands 0–14 and exact successful ID-matched ACKs. Command14 was new-selected, a bounded ArrowUp; it was not replayed. Controller PID6409 and Chrome PID6421 were live with the assigned runtime/profile. No original first-failure.json, transport-closed.json, or first-browser-failure.json was present. However, local first-error.json retained a failed fetch-original-startup-dadbcd read: at 02:08:03Z, startup.png did not yet exist. That marker was preserved. It is not evidence of an unacknowledged input effect. The predecessor subsequently continued through command14; this recovery did not erase that history or infer authorization to disregard the marker.
Only two new numbered commands were issued: passive capture15 recovery-passive and lifecycle quit16 recovery-quit, both exactly ACKed. Personally reviewed review/recovery-passive.png confirms New Game selected. Its process/document/canvas/guest Worker/module/render Worker identities match command14 across all 13 compared identity fields; sampled physical keys were all released. CPU and renderer both used module SHA-256 f9650607181301b5d2a9990420f437bd9f1ca10d40ac73067339f07499d53121, process1000, guest/render endpoint1, slot0. The recorded renderer is WebGL2 ANGLE Intel Mesa. No new selection, Control press, movement, config edit, or gameplay input followed takeover.
Personally reviewed review/controls.png preserves the predecessor's native On Foot Controls screen: Forward PAD8/UP, Backward PAD2/DOWN, Left PAD4/LEFT, Right PAD6/RIGHT, Fire LCTRL/RCTRL and mouse LEFT, Jump SP/PAD5, Next Weapon TAB. No remapping was performed. Existing fixture qualifications remain retained (exact-controls-qualification.json, positive-request-qualification.json, observer-qualification.json, release-control.json); they do not substitute for actual in-game fire consumption or a rendered firing consequence. The earlier texture-repair run's 250ms Control attempt with no distinct shot/ammo decrement remains unresolved; it was not repeated here.
Tested source identity is 7849be78c2c4cac23bbe3ab6f9ec6e0d26926a73, containing repaired boot commit37ce9588e. Recovery recipes retain source-sha.txt, runtime-pins.json, relevant-source.patch, and new-wininit-boot.js.patch; Git is the source archive. All 65 original media pins were rehashed successfully locally (recovery-original-pins.json), with paths and SHA-256 in canonical-original-verification.json. Media/source trees were not archived. The tested original and compatibility modules are recovered under retrieved/.
retrieval-final-verification.json records 257 artifact references, 233 unique SHA-256 values, and 78,210,705 reference bytes. Recovery retained every generated trace rather than using the earlier inventory's trace exclusion. recovery-verify.py independently rechecked every retrieved size/hash, all 17 original ACKs, unchanged owners, released physical keys, and all recovered PNG chunk CRCs plus IDAT decompression. retrieval-verified.json passed before normal-stop request. The original boot regression passed again in recovery-boot-regression.log. No emulator code changed, so no new build/fix claim is made.
Quit16 closed Chrome normally at 02:12:43.909Z (code0), removed its profile, and closed both listening ports. The initial drain assertion failed because the Node controller PID remained live after browser shutdown; its exact failure is retained in recovery-drain-remote-result.json. No successful controller-drain assertion is claimed. The driver emitted cleanup.json with generic pending: UNKNOWN; never replay; the complete 0–16 ledger independently shows no outstanding command. Evidence retrieval and hash verification preceded box shutdown, which terminated the remaining controller. No driver fix was attempted.
Assigned existing box bx_8e659hv2, runtime /home/user/redline-repaired-native-action-runtime, initial expiry 2026-11-08T02:32:14.885Z. Existing supervisor3049281, watchdog3049283, and normal delegate3049284 were verified live before recovery; their immutable deadline was unchanged. The normal delegate completed stop stop_789a68d2b93cdd2b, forced=false, at 02:13:41.846Z. Final state archived, expiry null (normal-stop-confirmed.json). Both delegates exited0 and supervisor terminal has no remaining children. No new box, resume, deletion, browser restart, deadline extension, nested worker, guest/config/clock change, or public deployment occurred. Sampled local free disk remained above5GB; remote free disk exceeded58GB.
Next: a separately authorized clean action run should reuse the repaired boot and these native bindings, review the first-person HUD, then establish actual guest-consumed input and one visible firing or vehicle-door consequence. This interrupted run supplies neither. FPS, audio, software/backend parity, and full playability remain separate.