MW3 encounter attempt — 2026-11-07

The single authorized run stopped at a passive-capture ACK timeout on the operation map. No cockpit, target, weapon consequence or native mission progress was established. The worker requested a 30,000 ms passive wait through a client with a 20,000 ms ACK deadline. This was a run-driver mistake, not a demonstrated emulator mismatch. The first UNKNOWN closed input permanently; the later successful ACK was preserved without reopening admission or replay.

Evidence: scratch/runs/20261107-mw3-encounter/. Preparation Git SHA: b3b0be7238796c856ce21464fed12a557bfe72cc. All production/runtime pins match the successful upload-gameplay predecessor. The actual browser-recovered module is 2,182,500 bytes, SHA-256 f9650607181301b5d2a9990420f437bd9f1ca10d40ac73067339f07499d53121. No production source or emulator behavior was changed.

Preparation and retained controls

The predecessor was terminal/released before ownership and allocation. Its published note and cockpit/after-left/after-fire captures were reviewed: movement and heading/heat changes are supported, but no beam, hit or damage. Passed native-input, independent-UP, identity/freshness, sticky-UNKNOWN, GetDeviceData observer and per-file upload/drain controls were hash-referenced and reused. The observer and its complete binary proof are byte-identical; normal caller remains 0x4f76c8, function 5962, entry 888199, keyboard completion 888442. This run never armed it or sent cockpit input.

The retained demo README text documents E for next enemy, 1 for zero throttle, 5 for 50%, Enter for cycle weapons and Space for fire. Its existing path/size/hash are in documented-controls.json. The sole input-helper change generalizes the existing Digit5 encoding to ordinary digits. Focused checks passed 0/1/5 encoding, unchanged E/Enter/Space and four rejected invalid keys. Those additional gameplay controls were not exercised on the original.

Static executable Node closure passed 66 files / 311 pins, with no production drift. The omitted predecessor qualification helper is not part of the executable closure. Local staging-verifier compilation passed. Actual staging uploaded 313/313 paths, drained every submitted upload, and verified all remote sizes/SHA-256 before boot. All 311 runtime pins remained unchanged at terminal custody. The previously passed locked build was reused; this documentation-only commit does not claim a fresh emulator build.

Reviewed original route and terminal fault

The original production browser/guest/render Worker route reached the menu, pilot a, unchanged Bushwacker / Operation 1A / Battle Progression / Wave 1 setup with Firefly and Owens enemies, and operation map. Native pointer lock, bounded mouse motion, pilot typing and clicks used the retained balanced-input path. Every input had independently scheduled UP before DOWN and native ACKs. No binding, profile, clock, private handler or guest-memory change was made.

All retrieved PNGs were inspected before subsequent input. The initial menu.png was still the splash; a passive capture reached the actual menu. The software cursor reached the established lower-right Launch control at guest (575,432), distinct from physical (554,584). Its click was acknowledged, but mission.png, cockpit.png and the late settled-cockpit.png all show the operation map. Filenames are not checkpoint proof. No Launch tooltip, successful transition, loading failure or emulator defect is asserted.

At 21:47:29.298Z, command 13, settle-cockpit, requested only capture with waitMs:30000. The client's fixed 20-second ACK wait expired and wrote sticky transport-closed.json at 21:47:49.399Z. The capture started at 21:47:59.740Z and finished at 21:48:00.271Z, after which the runtime wrote its successful reply-13.json. The image hashes exactly match the earlier operation-map capture. No command or native input followed command 13. Read-only recovery retained both the UNKNOWN and late success. Increasing the browser's immutable runtime cap would not correct this client contract.

Recovery and shutdown

The sole UUID-idempotent box was bx_97gtf2ua, created 21:44:03.604Z, initial expiry 22:44:03.620Z. More than 5 GB was free before allocation; an independent 1,500-second watchdog and a separate normal-stop OS delegate were armed and alive beforehand. Immutable watchdog deadline: 22:09:03.787Z. No other box was created, resumed or changed.

After the sticky UNKNOWN, an external cleanup helper read native-event evidence and issued CDP Browser.close, without game input or a new command. Chromium exited code 0, signal null at 21:48:55.400Z. The idle supervisor was then terminated with SIGTERM and its profile removed. This was external cleanup, not an ordinary game quit. Terminal custody verifies both PIDs absent, ports 43167/46355 closed, profile/writer lock absent, 54 native events with no held key/button, and unchanged runtime pins.

188 remote artifact paths / 9,320,580 bytes were retrieved and size/hash verified before requesting box stop; repeated capture content is not counted as distinct visual evidence. The executing module, logs and late receipt are retained. Normal delegated stop stop_4cf677ebf8b31b36 completed 21:49:24.545Z, forced:false. Final state archived, archiveAfter:null; retained-data expiry is unknown. The watchdog was alive immediately before and after normal stop. Both lifecycle delegates exited normally and the local messageboard watcher was closed. No live resource ownership remains.

Git SHA, relevant tracked/new-source binary patches, owned helper/document patches, module and existing media path/hash references provide recovery. There are no source archives, copied media bundles or public deployment.

Next work should first reject or budget waits that cannot fit the caller's ACK deadline, with focused pre-dispatch/late-ACK controls. A subsequent original attempt needs separate authorization. This task's single run is finished; the operation-map transition and encounter remain unverified.