Redline native vehicle route — 2026-11-08

Vehicle entry and driving remain unverified. The original native menu provides Training Levels → Vehicle Training. It starts on foot with a car directly ahead. Reviewed forward/right movement reaches the rear bumper and clears its right corner, but subsequent bounded left inputs do not produce a reviewed heading change or side-door view. The immediate blocker is unresolved guest consumption of left steering during this approach. Its underlying cause is not established; this run does not claim an emulator diagnosis or repair.

Evidence is retained in ignored scratch/runs/20261108-redline-native-vehicle/. Personally reviewed views include review/training-open.png, vehicle-select.png, vehicle-start.png, flank-advance.png, around-rear.png, clear-corner.png, face-car-flank.png, toward-known-flank.png and final-route.png under that run's review/ directory. The last views retain the on-foot weapon and health100. There is no changed vehicle HUD/camera, contacted visible door, acceleration/braking/steering consequence on a road, or successful exit to report.

Native route and control boundary

The shipped original/MAINDIR/Readme.rtf records two Training missions, entry by walking into a driver/passenger side door, arrows for acceleration/braking/turning, and Delete/Page Down for left/right exit. No menu label was invented. One initial mouse selection intended for the visibly present Training Levels option instead opened the ordinary mission grid. That mistake and its capture are retained as training-levels; Escape returned to the menu. Individually reviewed Down/Enter navigation then selected the actual Vehicle Training option. Stadium City was not started.

The training starts directly in the on-foot scene; no textual mission briefing was observed. No audio-based instructions are claimed. The car was kept visible during the approach to its rear/right corner. A short forward step cleared the corner; the subsequent left turns targeted that immediately adjacent known car, rather than surveying unknown routes. A side door never became visible, so the documented door-contact entry condition was never tested.

All28 numbered commands0–27 have exact successful ACKs, including final passive capture26 and quit27. There is no refused, failed, outstanding or replayed command. Three separate2000ms ArrowLeft attempts and one2000ms Up+Left attempt are explicitly preserved; none demonstrated a heading change in the reviewed after/settled views. Their wholly-held passive samples observe physical VK37 down; the combo also observes VK38 down. Independent release receipts record2000ms intervals. This verifies host input delivery/release, not a guest keyboard poll or simulation-step consequence. Period-strafe attempts are also preserved; the inherited physical sampler does not include VK190, so no in-hold physical-period observation is claimed.

passive-input.json and passive-input-final.json show physical/async movement keys released and an empty input queue. CPU slices and render completion counters advance. Recorded cumulative renderer synchronization time is large, and several held samples retain one completed slice sequence throughout their interval. These observations make timing a candidate explanation, but do not prove missed guest polling, identify an incorrect key binding, or establish that rendering caused the steering failure. Further input was stopped at this unresolved dependency. No firing, controls-menu loop, mission/configuration edit, remapping, guest-state or clock forcing, browser restart or backend comparison followed.

Reproducibility and checks

Tested source SHA is efe297deaf4a3daae1d8fbd6beb05258babc5d58. source-sha.txt, relevant-source.patch, runtime-pins.json and small driver-input patches preserve the source/driver recipe; Git remains the source archive. All65 original media references were rehashed unchanged. No source/media archive was created. The actual guest and renderer module SHA-256 is f9650607181301b5d2a9990420f437bd9f1ca10d40ac73067339f07499d53121; recovered modules are retained. All18 compared owner identity fields match from training spawn through final capture, with process1000, guest endpoint1/slot0 and render endpoint1.

The exact-ready startup waiter regression passes, including delayed publication, missing-artifact rejection and hash mismatch rejection. Existing arrow/combo paths were reused. Only the newly admitted Delete/PageDown controls received focused blank-page native trusted-event checks and local payload/independent-release tests; neither exit key was used in the original game. The existing browser DirectX-vtable boot regression passes. No emulator source changed, so no new locked-build or emulator-fix claim is made.

verify.py independently passes420 artifact references,391 unique hashes,195,522,943 reference bytes, all28 PNG chunk CRCs/IDAT streams, all28 ACKs, owner matching, release/drain evidence and65 original pins. This verification passed before the normal-stop request. Actual owning renderer inspection reports WebGL2 ANGLE Vulkan SwiftShader, legacy/WebGL,640×480, context intact and fallback count0. The adapter's software:false is not evidence of hardware GPU execution.

Shutdown and next work

One UUID-idempotent fresh box/browser was used. Observable independent watchdog/normal-stop delegates were armed before allocation, with the original immutable25-minute deadline. Chrome exited0 at02:53:40.411Z. The first drain found the controller briefly present; the retained second drain confirms controller5886 and Chrome5898 absent, both HTTP/CDP ports closed, and profile/writer lock absent.

After retrieval and hash verification, delegated normal stop archived bx_9h52bbtt at 02:54:42.884Z, receipt stop_fc7e06f699e7d313, forced:false. Final expiry is null; initial expiry was 03:05:25.132Z. Both lifecycle delegates exited0 and supervisor terminal has no remaining children. Sampled local free disk remained above5GB. No resume, deletion, deadline extension, nested worker or public deployment occurred.

Next work should resolve why the qualified physical left-input windows produce no reviewed native turn before resuming this door approach. Vehicle entry/driving, objectives, FPS, audio, paired backends and full playability remain separate and unproven.