Recoil menu-assisted objective attempt, 2026-11-08

Run: scratch/runs/20261108-recoil-menu-assisted-objective/.

Force-field objective incomplete. One fresh watched original demonstrates native menu entry/Resume, a bounded HUD-timer pause observation, and a practical review/control limit. It ultimately aligns the uphill ramp while alive, but the next hold's first frame already shows player death. No power control was identified, no force field was removed, and no onward access or objective counter/text consequence was demonstrated. No restart followed death. No emulator mismatch or production change was established.

The accepted beach-enemy destruction remains foundation in native objective and current gameplay. The previous route attempt supplies the known uphill route and power-control brief. This original followed ordinary Software → OK → Game → Start → New Game, default user, Medium, and the natural opening. The fresh loading display captured here cycles to “1.2 Missile Control”; it is not claimed as a newly captured 1.1 power-control brief. R/O statistics were not reopened in this run; the prior 0/4 is not a fresh counter measurement. The last living score here is 0.

Native menu review: supported scope and limits

The installed game/docs/index.html was read. It explicitly documents Escape calling the In-Game Menu, not simulation freeze. Early after the living HUD appeared, original/living-hud.png shows health 94, timer 00:00:26, incoming cyan fire and beach enemies. A native 100-ms Escape press/release opens the blurred scene and then the menu with Resume. An explicit 8-second idle request follows; its screenshot is original/menu-idle.png.

menu-observation.json records these personally read HUD comparisons:

Living captures Wall seconds HUD timer Health Menu-entry input to mouse Resume
living-hud → resume-settled 78.693 26 → 45 94 → 83 59.144 s
resume-settled → live-two-settled 93.438 45 → 75 83 → 67 61.768 s
cliff-left → route-resumed 57.117 142 → 154 23 → 20 43.659 s

The wall-minus-HUD differences (59.693, 63.438 and 45.117 seconds) approximately match the native menu intervals, allowing for transitions and whole-second HUD readings. This supports exclusion of menu time from the displayed game timer. The tank resumes alive after each interval. Exact health at menu entry/exit and actor simulation pause remain unmeasured. The menu can retain a cached blurred backdrop; unchanged background pixels are not independent actor-pause evidence. Health falls across the enclosing active/menu intervals. These data do not establish complete simulation or health freeze. Menu review was used within this observed scope; no guest clocks/state were edited.

There are two concrete control limits. Escape opens the menu but a second native Escape does not Resume (key-resume-settled.png). An ordinary relative mouse move and independently released left-button selection do Resume. That menu pointer also carries into the aiming view: resume-settled.png and route-resumed.png show a high reticle and upward turret. Native C restores a level view aligned with the chassis (centered.png, route-centered.png). Resume may still show transitional menu pixels in its immediate capture; a subsequent capture establishes the actual resumed HUD. No static menu image was treated as proof that the tank remained paused indefinitely.

Navigation, defense and death boundary

Fresh pixels guide each action; no coordinate script was replayed. A short forward/right arc starts around the wreck. past-wreck.png shows health 51 and a grey scout near the reticle. A single 1.8-second primary burst is attempted as ordinary defense against that visible threat. This is not an objective attack or a new accepted destruction. Its first passive held frame, held-1794123649616.png, already has the scout left of the reticle; later held-1794123650776.png shows it farther left. Cyan fire alone proves neither a hit nor a kill; no score/destruction consequence is established. The capture-to-dispatch interval allowed the moving target to depart. No further combat burst was sent without a newly aligned target.

cliff-forward.png reaches the steep cliff base, health 30. Left turns expose the side but subsequently overshoot toward the beach. Smaller right corrections and the native M wireframe map resolve the orientation. Health remains 20 through several views, then 19. scan-right-two.png clearly exposes the uphill paved ramp and red/green posts at the right; a 220-ms Right correction moves the ramp ahead in ramp-aligned.png. These posts are not identified as the power control or independently proven to be the objective force field.

The last living capture is 07:44:20.209Z, health 19. The first passive frame of the final forward hold, held-1794123875082.png, is already the burning player tank and Load Game / New Game / Quit menu at 07:44:35.082Z. ramp-advance.png is also death, despite its command-derived filename. The exact death instant and killing projectile/actor are unmeasured. There is no demonstrated living forward response and no causal attribution to that hold. The final exposed capture-to-first-held-frame interval was 14.873 seconds. Menu review did not solve that interval; this run failed to reach the ramp earlier in active gameplay than the previous attempt. No further gameplay input followed recognition of death.

Runtime and evidence qualification

The runtime uses published Git blobs from 1e2bdf191574fc11b8a03515cab866c46d305f1e, excluding active peer WIP. The unchanged, hash-checked compiler compiles the published WAT closure in an in-memory VFS. Actual CDP-recovered Calc and original modules match 7ec49cbbae4cb177909fd8fd9d3599713ee5f2a5ffb7f7e8c890848a540052bf. This is a qualified published closure, not a claimed passing shared-WIP build. No tracked code was changed, so no new shared-tree build was required.

All 727 runtime pins pass their Git/artifact providers. The real registration has 106 manifest entries + executable = 107 originals, plus the manifest: 108 pins / 76,019,820 bytes, all matching size/SHA and staged pins. The prior invalid zero-file success report remains preserved and withdrawn. Empty, missing, duplicate registrations and wrong staged executable SHA regressions pass; all 113 local HTTP fixture/dependency checks pass. Corrected Git-byte HTTP lengths and closure helpers are reused. Executable SHA remains 82632b1b46153014af0caf1e0c9dfdebbb0d8482ba65b2735064f10972afa076.

Real-Chromium generated tests retain exact owner/module/freshness identities, the 120-second capture-age bound, independent release, partial delivery and sticky UNKNOWN closure. Native Escape, relative motion, trusted Resume-style mouse press/release and subsequent Escape composition pass locally. A delayed 650-ms sampler cannot extend either 100-ms Escape or mouse hold; released key and mouse states are checked. These are separate admitted operations; no atomic Resume → movement/fire → independent release → Escape operation was qualified or available in the staged driver. Actual Calc qualifies the staged runtime. All 42 original numbered commands have exact successful ACKs, with no original UNKNOWN or replay. The sole allocation precheck initially ran before the armed records existed; it made no API request. Both independent records were observed before the sole UUID-idempotent allocation.

The original is Chrome 151.0.7922.108, V8 15.1.206.16, 1024×768, cross-origin isolated, Threads enabled, unchanged registered /index.html?debug&threads&app=recoil_demo. Rendering is software page Canvas2D, with no owning D3D render-worker endpoint; toolbar WebGL is not backend proof. Process 1000 / endpoint 1 / slot 0 and document epoch f87d887c-d235-44a9-aaf9-deb808d599a7 retain the observed owner chain.

All 581 remote artifacts / 47,692,701 bytes were retrieved and independently rehash-verified before stop. The initial post-quit inventory asserted before all PIDs disappeared; its exact failure is retained. Later read-only observation and settled inventory confirm all 15 PIDs absent, HTTP/CDP ports refused, profile/writer lock absent and inputs released, without browser/input replay. Local browsers are drained. The independent prearmed 25-minute watchdog and normal-stop child exited 0; the supervisor ended with no children. Box bx_jegnryx3 is archived, stopped 07:46:01.624Z; final expiry is null, initial expiry 07:58:20.884Z. Receipt stop_726617078a4744d1 completed 07:46:01.647Z, forced=false, error null; final-stop-receipt.json retains the final GET. Disk remained above 5 GB. No resume/delete, RAM gate, guest patches, forced settings or deployment occurred.

Git plus explicit patches provide source recovery; no source/media archive was made. All 136 original PNGs pass CRC/IDAT and capture-hash checks. evidence-check.json, scene-review.json, menu-observation.json and death-boundary.json retain checks and qualified conclusions. Next work needs a locally qualified shorter resumed-view/target/steering decision interval, then a fresh living approach to identify and act on the actual control. The current 120-second admission freshness ceiling is not a survivability or target tracking guarantee. Full27, parity, audio and unmeasured FPS remain open.