NFS3 ordinary browser driving and Pause/Continue, 2026-11-18

Ordinary production-browser input raised the original Corvette from 2 to 15 MPH, advanced the road view, opened the original pause menu, and resumed the cockpit with its clock advancing. Steering changed the wheel, but the separately reviewed captures allowed the car to coast to zero: they do not isolate a causal road/yaw effect. Ordinary Quit reached the original System confirmation; native process exit was not observed within the predeclared 15-second terminal window. This is partial control evidence, not completed driving/steering/Quit acceptance.

Evidence: scratch/runs/20261118-nfs3-current-browser-driving-pause/result.json. All screenshots are named whole private-display XGetImage captures, without crops or replacement pixels. No full-27, parity, FPS, or audio claim follows.

Actual runtime and route

The first original used immutable published Git ac2e2b69ae1428dac7610c736a9607730ee4db1b; the affected second original used published helper commit 95df0faa906cba19b7987d1417a38f061e63bade. Their src/lib/host/index bytes are equal. Both used the same retained compiled module, 2,413,749 bytes, SHA256 a6e2bb71152d588f71e9059f9df3105349e1fd64811a81de0bb5aeeaa18d2f42. Debugger-extracted guest/render Worker bytecode, live endpoint, primary window, canvas, and bridge generation were checked. Prepared manifests retain exact source hashes. Foreign working-tree overlays were excluded from the served immutable runtime and preserved.

Launch was the supported registered nfs3_demo production Worker route with unchanged 135 media, all rehashed. Its unchanged registered startup registry chose Glide/WebGL. This is not fresh Direct3D DIM acceptance of the earlier cf7044e37/1fe5e1baf repairs. The original game automatically reached its race; no game, INI, registry, or state was authored. The first original showed rainy Hot Pursuit, the second dry Single Race, reflecting their actual unmodified starts.

Concrete helper repair and validation

The first original's throttle worked, then the old harness armed its independent UP child before checking the full live owner twice. Those checks exhausted the 1,750ms physical release window and the left DOWN was refused before delivery. The helper now validates the owner before arming that short clock, checks remaining hold capacity, and always waits for independent release after DOWN. It introduces no guest poll, wake, focus, clock, handler ACK, or runtime write.

Focused old-order-fail/new-order-pass tests, the existing browser controller loopback and NFS control-observer tests passed. The required locked full build passed before the explicit helper/test commit 95df0faa9, followed immediately by strict pull/push (both zero). Calibration4 and calibration5 exercised the changed helper through trusted ordinary browser Enter, visibly changing the native fixture blue to red. Calibration4's later Escape was refused on the separate 1-second release-child startup bound; a 5-second startup bound was declared before fresh calibration5 and original2. Calibration5 did not establish native exit with its immediate terminal assertion. Calibration3 independently established the earlier capture/release route and fixture ExitProcess(0).

Original2 predeclared capture 3s (measured whole capture/encode 282–507ms), review 90s, transport 3s, child startup 5s, fallback UP 1,750ms, holds at most 1,200ms, runtime 900s, Quit reserve 180s, terminal observation 15s and enclosing cleanup 20s. No in-flight deadline extension or closed-scope replay occurred.

Original2 observed consequences

Trusted DOM edges and independent UP child waits are retained per gesture. Available host check_input logs show Up/Left/Escape messages actually dequeued; the passive native API observer collected zero handler records, so no custom per-action native consumption ACK or complete DirectInput proof is claimed.

Remaining blocker and cleanup limits

Next bounded work is the original System-Quit transition and a steering hold while throttle is maintained, with a preplanned capture short enough to avoid coasting between controls. Do not infer a runtime crash or an exact Quit failure cause from a 15-second non-exit alone. No runtime input/render code was changed.

All seven browser/display scopes have actual browser, Xvfb, Openbox and native supervisor waits; all native supervisor receipts report no leaks. Original2's outer tool session was actually reaped with status 143, but its enclosing launch script did not retain its pipe-EOF wait receipt. The board watcher was also actually reaped 143. Consequently enclosing-launcher EOF and complete cleanup attestation are unavailable for that scope; process absence is not substituted for a wait. Earlier six enclosing-launcher receipts exist. No test box was created: ID/state/expiry/stop receipt = null/NOT_CREATED/null/null.

Git is the source archive. The run retains the unique tested module, exact Git and relevant binary/source diffs, owned new-file patches and recovery recipe; no source snapshot, media copy or continuous frame series was made.