ReVolt manual driving and ordinary Quit, 2026-11-18

Manual throttle and steering now have visible consequences in the original PC Gamer demo. In manual-tmpfs-deferred, ordinary Controller Settings → Configure Controller established and displayed Left=LEFT, Right=RIGHT and Accelerate=UP. The operator manually selected Single Race, PLAYER, RC Bandit and Toys in the Hood 1. A fresh whole-display review authorized Up for 1200 ms at age 3904 ms; the car moved from the road toward the curb. A later, separately reviewed Up+Left for 1200 ms reached DOWN at 4438 ms, changed the heading toward the wooden fence and produced a personally inspected 8 mph HUD. Both operations have independent release ACKs and physical UP0. Lap remains 1/3, position 8th; no checkpoint, completed lap, course completion or broad playability claim follows.

Ordinary race Quit returned to the front menu. Application Quit → Yes reached the demo's promotion screen, rather than immediately exiting. Declaring that gesture final prematurely prevented further normal input; the fixed 60-second terminal check failed and permanently closed that scope UNKNOWN. No endpoint followed closure. A new, explicitly declared manual-tmpfs-quit-nag scope changed the exit plan to review the promotion before final dismissal. Its ordinary menu Quit → Yes, reviewed promotion, and personally chosen Enter for 200 ms produced processExited=true, running=false, and both _exitCode and process exit code 4097 within the unchanged 60-second terminal bound. The final whole display shows the application gone. Driving and verified application exit are separate scopes, not one uninterrupted successful driving-to-exit run. Code 4097 is recorded as observed; it is not described as exit zero or Windows reference behavior.

The handoff retained the published loading qualifications 8c4052149/0f00d8c01, atomic operator 5e4e89fc9, UTC/tmpfs helpers, original module SHA-256 a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8 (2,418,710 bytes; compiler source f6962ba5b1cf2829a66f5cc4d80b293c30719b37), and 463 verified input/media pins. No unchanged runtime-module compilation or media copy was performed for launch. Both successful scopes served source graph 6d5558bacfa78946c235d4ac9fb79dd774445db0; their actual HTTP source/module hashes match the retained manifest. Executing module/worker-role attestation remains UNKNOWN; served bytes alone do not prove it.

Two focused published corrections address measured overhead and capture readiness. 93ee0d25c uses ReVolt's own D3DIM identity reader to wait passively for already queued publication/paint before issuing a native capture ticket. Unexplained crop, identity, visibility and pixel failures still close; failed real tickets are never retried. A mistaken attempt to reuse the GL reader first proved a render-endpoint mismatch and permanently closed its scope. cce80c7c4 moves independent release-child imports before capture. Measured post-command preparation previously cost 1547/1278 ms. The standby child has no configuration, connection, arm, DOWN or release clock until the chosen command arrives, and expires after 120 seconds if unused. Release fallback, identity checks and the five-second review limit are unchanged. Six focused tests, exact-adapter syntax checks, a real Chrome loading-overlay integration, the authored native preparation fixture (1437 ms age, independent UP0), and both required locked builds passed before publication. Each commit was immediately strictly pulled/pushed.

Actual final gameplay phases are retained in audit.json: throttle capture 75 ms, capture completion to tool image return 102 ms, image return to arm 3704 ms, arm to DOWN 23 ms; steering 122/142/4143/31 ms respectively. Imports and source/readiness preparation happened before capture. One intervening steering intent expired before arm at about 5291 ms, was NOT_SENT, and its unused child was joined; the successful steering used a new capture and intent. The operator awaited each genuine complete capture, immediately forwarded the whole image, personally inspected it, then used a short tool invocation for one finite choice. No unrelated reads or logging occurred between those successful image reviews and commands. No input was pre-sent, timestamp recycled, review fabricated, limit waived or play automated. Scratch-only startup adaptation deferred automatic native tickets until naturally loading UI settled; it granted no input admission and every subsequent command still required the normal qualified whole capture.

Evidence is scratch/runs/20261118-revolt-capture-review-handoff-driving/; result.json is published last. Seven original scopes and one authored fixture retain permanent closure and nested waits. Earlier scopes failed on disk, capture crop, incompatible reader or startup proof; none was reopened. The final audit finds zero ordinary requests after any UNKNOWN, 95 actual nested wait/EOF receipts, no supervision leaks, and all 299 recorded runtime PIDs absent. Both final enclosing tool calls returned 143, despite their nested launcher/supervisor/watchdog receipts and successful Quit-scope driver/browser waits being zero; the separate invoker-child receipts and outer EOF are missing/UNKNOWN. These gaps are not reconstructed. Full-task continuous disk compliance is false: the shared filesystem fell below 1.5 GB during earlier closures and again between scopes. New launches required actual space above the floor; peer-owned verified recovery restored capacity. No foreign files, holds or assigned boxes were cleaned by this worker.

Unique tmpfs evidence was copied, gzip round-tripped, hashed and fsynced before removing owned tmpfs directories. Only selected named whole screenshots remain; omitted redundant frames retain size/hash records and are not recoverable images. Git commits and binary diffs archive source; owned new inputs have patches, and the module/media recovery recipe names existing pinned inputs without duplicating them. No agents, workers, boxes, Wine, Python or deployment were used. Box ID null, state NOT_CREATED, expiry null, stop receipt null. Ownership is released after final sealing.