MW3 headroom and first cockpit action boundary — 2026-11-07

The fresh original reached Launch and the cockpit after correcting the acquisition-anchor choice. Its first cockpit movement attempt did not demonstrate movement: after an ordinary ArrowUp pulse, speed remained zero and OP Able remained 449 m away. Turning, firing, displacement, ammunition change and projectile evidence remain UNKNOWN. No further gameplay input followed this boundary. This is not evidence of an emulator defect.

Evidence: scratch/runs/20261107-mw3-headroom-gameplay/. No production code changed. The native helper, automatic capture, sensitivity, identity/freshness guards, exact numbered ACK transport and independent UP mechanism were retained. There was one fresh original, no accepted-matrix repeat, nested worker, reanchor, forced profile, private input handler, guest/configuration/clock write, UNKNOWN replay, old-box operation or source/media archive.

Local prerequisite and anchor correction

The exact run-specific browser, actual completion observer and transport passed local closure using the generated public-COM fixture with production guest and render Workers. Personally reviewed pixels admitted the read-only observer and an ordinary ArrowUp pulse; exact ACKs 0–3, independent UP, observer restoration, normal Chrome exit and closed-child refusal passed. Focused tests rejected wrong Worker instance, changed module bytes and wrong completion location before breakpoint installation. A stalled DOWN ACK still independently attempted UP and returned sticky UNKNOWN. The published native-motion focused test passed. The generated fixture establishes driver closure, not MW3 keyboard semantics.

The unchanged tools/dungeon-keeper-native-motion.cjs was then exercised locally through the complete historical menu/Launch coordinate envelope. This is an explicit simulation of physical headroom, not another original or an actual guest-consumption claim. The first candidate (181,314) failed the expanded envelope and was retained as headroom-first-refused.cjs; the corrected candidate (350,430) passed. For software envelope points including (272,208), (432,324) and (575,432), plus conservative excursions (20,20)..(600,460) at unit gain, every helper step remained inside the physical canvas (10,194)..(694,674): physical extrema (50,210)..(630,650). headroom-proof.json retains the exact helper receipts. The first refusal did not send original input or alter any helper guard.

Fresh original, actual consumption and navigation

The current original menu was personally reviewed before choosing physical (350,430), a blank area between Instant Action and Multi Player. The observer was armed before this first acquisition press. Automatic selection remained true, explicit profile absent, cooperative flags 5, sensitivity approximately 1.3; Pointer Lock acquired screen. No acquired session was reanchored.

The probe verified the exact 470-byte GetDeviceState binary body and executable completion location 887941, matched the owning Worker instance/endpoint and executing module, and retained the existing 18-field ownership/module/epoch and 120-second freshness admission. Focus HWND 65537 was retained; no separate HWND-object gate is claimed. Its condition returned false and wrote diagnostic JavaScript buffers only, never guest memory or input state.

Acquisition observation retained 2,459 actual completions, zero drops, all HRESULT 0 and caller 0x4fa56c. DOWN was attempted 15:38:19.347Z and acknowledged .358Z. Four actual returns at .365, .392, .416 and .442 contained left-button 0x80; independently armed UP was attempted .448Z and acknowledged .454Z. Subsequent returns contained zero. The observer restored its buffers and removed the breakpoint. Raw device-table flags were not separately proved.

Ordinary trusted input created pilot a, accepted default Bushwacker / Operation 1A / Battle Progression / Wave 1, and opened the operation map. Software cursor, renderer coordinates and tracked physical pointer were kept separate. One fine step initially had not appeared in the software snapshot; a passive settled capture established its consumption before clicking.

At Launch, actual software cursor (575,432) corresponded to tracked native (554,584), leaving 140 physical pixels to the right canvas edge. The current highlighted lower-right icon matched the predecessor's proved deployment control. A readable tooltip is not claimed from the current or re-inspected predecessor pixels, despite the predecessor's written tooltip description. One click at the unchanged physical point reached cockpit capture mission.png, SHA-256 a87c75a60bf89ded39aa68d97f509784ecefab79bf9f51be0fc2a4388125bf69. This closes the demonstrated physical-headroom route limitation.

First unresolved cockpit input boundary

A fresh bounded actual completion observer was armed at 15:42:32.760Z, before cockpit input. Baseline movement-armed.png, SHA-256 f55bae6f451178fec4c45b66d3b3695bb5a8e9563a898fe668dbc37d89abffaf, shows terrain, speed zero, OP Able 449 m, and ammunition 100, 100, 24, 24, 20, with ER Laser selected.

Exactly one ordinary ArrowUp DOWN was attempted 15:42:43.631Z, ACK .646Z. Its UP was armed before DOWN, independently attempted 15:42:45.632Z and ACK .634Z. The 2,000 ms pulse did not establish useful movement. after-forward.png, SHA-256 eadf3cc203a923e76ee819a0c7002b5858a8ac366efdd2868f38bccd1f8ac46f, and later passive movement-final.png, SHA-256 139d9789e1ac4ed1745185fb996539301fb5532067c295b9eedd8083a8b67de2, still show speed zero, OP Able 449 m, the same terrain alignment and ammunition. The guest mission clock advanced from 00:00:01 to 00:00:04; this is not a frozen browser assertion or movement evidence.

The cockpit observer retained 22 successful mouse-state completions with zero drops, caller 0x4fa56c and no pressed mouse buttons. One completion occurred 1,755 ms into the ArrowUp pulse. This observes the mouse state buffer; it does not establish keyboard-state return or ArrowUp binding/consumption. The approximately two-second spacing under this diagnostic instrumentation also prevents a performance or uninstrumented timing claim. Both observers removed their breakpoints and diagnostic globals.

No alternate key, turn, target click or fire was attempted after the unresolved movement boundary. Binding, actual keyboard consumption and diagnostic timing remain possible explanations; no fail-before/pass-after emulator repair is justified. Recovery should first qualify the ordinary keyboard driver and a read-only keyboard-consumption observer locally. Existing app notes identify the included demo's 5 binding as 50% throttle, while this retained driver's key whitelist currently excludes digits. That is an actionable local driver prerequisite for a separately scoped successor, not permission to replay this original or a claim that ArrowUp was consumed incorrectly.

Recovery, identity and lifecycle

The recorded source SHA is 415f647a7efe017f916766939c036af3ec4fa5bf. Runtime pins, relevant tracked binary patches, owned new-helper/document patches and existing media path/length/SHA references preserve recovery. The measured guest and renderer module is e41511fe861363fbe1639d80f20cbcdd7240f46937c93a801eaa256ca094c467, 2,182,493 bytes. This does not claim a rebuild at that source SHA. All 308 runtime pins matched before and after execution.

Exact original ACKs 0–18 and 56 trusted native events were retained. All 239 remote artifacts (11,577,995 bytes) were retrieved and verified by SHA/length before normal stop. Final custody proves driver and Chrome absent, ports 41019/33525 closed, profile/writer lock absent and held inputs empty. Chrome exited normally, code 0, at 15:43:26.724Z. Local fixture processes, profile, locks and watcher were also closed. Sampled free disk remained above 2 GB; no continuous minimum is claimed.

One persisted UUID Idempotency-Key POST allocated bx_nbzytg5y, original expiry 16:37:07.492Z. A delegated lifecycle subprocess was armed before allocation and wrote the immutable allocation-derived cap 16:02:07.663Z. At normal-stop delivery it was absent, with no stop-attempt/exit record or stderr; durable external-watchdog continuity is UNKNOWN. This separate lifecycle failure is retained in lifecycle-first-failure.json. The browser's immutable cap remained installed, and execution ended well before it. After verified recovery, the worker directly issued the explicitly authorized normal stop, without replaying a prior stop request. stop_83f8497774ecc58b completed 15:44:45.006Z, forced:false; the box is archived, current archiveAfter:null, retained-data expiry unknown. No lifecycle ownership remains.

No production repair or mandatory locked build was applicable. Focused local closure, helper/headroom, consumption correlation, exact ACK, custody, hash and owned-evidence checks passed; external-watchdog continuity did not. This single bounded task ends with the exact first movement boundary, without gameplay action success, firing, FPS or parity claims.

The corpus manifest has no MW3 candidate, so metadata is retained as investigation.json beside the notes, without attaching an unrelated dashboard candidate. Its 18 artifact references pass containment/existence checks. The global evidence checker reports the existing unrelated Alice and Tunnel/Twist containment errors.