NFS3 native driving attempt, 2026-11-14

Fresh original setup is blocked before its renderer chooser or game UI by COMCTL32's native thunk connection. No fresh cockpit, Controls/transmission, Pause, Continue, held driving, speed or scene consequence was observed. Historical cockpit 0 MPH is not driving evidence.

Evidence: scratch/runs/20261114-nfs3-native-driving/. Runtime source HEAD was 9afbc3e6df17030388d49fad017fdea047f8870e, with relevant foreign WIP captured in relevant-source.patch and new/helper files in owned-and-relevant-new.patch. Loaded module path was build/wine-assembly.wasm, SHA-256 c7f5bada4da4b4ed95b235add0aa1c9843176805d29d5b1ee79359752a19d348. All 135 original media pins matched before launch and again at finalization. No source archives, repeated media or binary module dumps were made.

Actual original result

Reconstructed preparation helpers from retained patches, then used registered nfs3_device_data followed by nfs3_device_setup, carrying only each fresh native registry export into the next process. The catalogue exited 0 and exported 118 entries. Setup loaded original COMCTL32 at 0x551000, DllMain 0x56b1cd, then trapped at 0x57d5d2, yield 18, halt 3: Missing NE image COMMCTRL. Its captured contract is LS01/LB01, checksum 0x1e9a, commctrl.dll!Cctl1632_ThunkData16, QT offset 36, FT offset 68. setup/stdout.log and setup/stderr.log retain the actual failure and stack. The control endpoint never opened; the read-only capture request was refused and produced no PNG. No input attempt, binding, DOWN, pointer action or replay occurred. No fresh chooser selection/export or game launch was attempted.

This is the same original connection prerequisite described in the COMCTL32 investigation. Current foreign lib/ne-guest-bridge.js first resolves the missing companion, then explicitly refuses LS relay publication until a native connection owner and FT prolog exist. Recovering a DLL alone cannot complete this route. A quarantined candidate companion belongs to another worker and was untouched. No production repair was attempted: finishing the shared native connection implementation exceeds a bounded NFS input/render repair and would overwrite foreign WIP. No guard, native chooser, guest state, clock or configuration was weakened or injected. The old chooser export was hashed only, never imported.

Verification and lifecycle

The actual guarded CLI consumption regression and real shared-WASM UI-owner regression pass. They were rerun under the heavy lock with logs and actual exit/EOF receipts in focused-tests.json; an initial invocation lacked that lock and is not the serialized verification receipt. The reconstructed pure controller validations also pass (eight control requirements, thirteen invalid pointer reviews). These fixtures establish no gameplay result. No production code changed and no new full build was run; this is a docs-only commit. Runtime helper patches preserve the exact tested source.

Both original phases held /tmp/wine-heavy-runtime.lock, checked the current source/module/media pins and actual child census, and armed an observable independent finite deadline/disk watchdog before launch. The setup diagnostic logs were buffered by the helper until process exit. After bounded endpoint failure, SIGTERM requested orderly shutdown but did not exit within eight seconds; owned SIGKILL yielded actual return -9 with both EOFs. Catalogue returned 0 with both EOFs. Both watchdogs returned 0 with both EOFs, all four PIDs are absent, and the heavy lock is released. No deadline was extended. Box ID NONE, state NOT_CREATED, expiry null, stop receipt N/A.

Next: complete and qualify the native COMCTL32/COMMCTRL connection prerequisite, then repeat fresh native catalogue/setup/Direct3D selection and unchanged export before Controls/transmission review, two settled Pause frames, ordinary Continue and independently released held driving. This task stops at the concrete original setup blocker; no screenshot or gameplay acceptance is claimed.