SuperTux original consequence, 2026-11-14

Evidence: scratch/runs/20261114-supertux-original-consequence/.

Fresh separately authorized local standard CLI task. Preparation verifies the 787 original files, authentic standing CRT and registered C:\ working directory with --sdl --datadir data. No remote runtime, box, download or deployment. The original EXE and 119,228-byte manifest match the owner-supplied SHA-256 pins.

Result: first-level timeout before reviewed gameplay input

One original launched on main 435ce2fc1129da8ed8051f70f41a34cea192e000, module SHA-256 dcf6e8f0ca9d361c10fa556fb30835d34cf5bfe4089f84d298cb57fd170dbd42. source-pins.json, source.patch and new-runtime.patch identify the actual loaded source including foreign WIP, without archiving source trees or media. Postflight found no source, module or media drift. All six bundled SDL/image/ mixer/jpeg/png/zlib DLLs plus the authentic standing MSVCRT actually mapped; the log confirms all seven DllMain calls succeeded and C:\ was the guest cwd.

Personally reviewed Options showed sound and music enabled. Keyboard Setup showed Jump = Space and Power/Run = Left Control. Neither binding nor options were changed. Ordinary reviewed arrows/Enter opened Start Game, free Slot 1, the world map and its first red node, Welcome to Antarctica. The intro passed during loading; no readable story frame or story-skip input is claimed. loaded-story.png is the world map, despite its capture name.

The first reviewed level capture, fresh-firstlevel.png, already showed TIME'S UP, score 0, coins 0, one remaining life. No movement or jump was sent. The old 75/3 and 125/5 screens are not used. Neither meaningful original consequence nor level exit was achieved.

The exact tested limit is the unchanged standard CLI's default service clock: test/run.js selects createServiceClock(5000) when neither real ticks nor an explicit batch-clock flag/registry setting is supplied. SuperTux has no registered batch-clock override. This clock aggregates retired blocks across the guest partitions; it is an abstract CPU-service policy, not wall time. The entry capture published guest time 344,562,171 ms; the first reviewed level capture published 358,437,444 ms, only 7.792846 wall seconds later. That is 13,875,273 ms of virtual progress, about 1,780.5 times wall time. This supports a route/timing blocker under the unchanged-clock constraint, not a new production semantic defect or a historical review-expiry repair. No clock, physics, score, profile or guest-state edit, alternate route or retry occurred.

The controller retained one permanent guarded input binding, unique nonces, the 12-second reviewed-frame limit, independent prearmed release processes, immutable per-pulse deadlines and postrelease batch progress. Fifteen finite reviewed pulses had consumed DOWN receipts, successful independent release receipts and later batch progress. The final Escape intent followed the actual timed-out level frame. Its handler receipt alone does not prove guest dispatch or a menu transition. The CLI's subsequent capture failed because the guest had exited; this was not an UNKNOWN input replay.

Guest [Exit] code=0 occurred at 204.844 seconds, before the fixed watchdog deadline. This proves guest exit, not reviewed native menu Quit or causal attribution to Escape. No host quit/timeout/kill was used for this original. The independent watchdog was armed before launch with an immutable 600-second bound, exited 0 without firing, and was actually waited. Native wait 0, supervisor wait 0 and both output EOFs were recorded; all 18 owned runtime/ release identities were absent. Heavy lock released, free disk above 13 GB. Box ID: none; state NOT_CREATED; expiry null; stop receipt not applicable.

Validation and retained evidence

Current full build passed under both heavy-runtime and build locks. Actual shared-WASM input-owner tests and the 22-check guarded CLI test passed, including independent release, sticky UNKNOWN, ownership replacement, deadline expiry and normal test-harness shutdown. Three run-local controller checks passed, including a real HTTP release child and postrelease progress. Run-local offline clock controls execute the actual CLI selector and clock implementation, confirm 5,000 blocks/second and sibling aggregation, and verify observation does not mint virtual time. Synthetic controls are not original gameplay proof.

An additional existing test-cli-tick-ms-input.js --source-only diagnostic failed before testing: its source-slicing marker for DispatchMessageA no longer matches current test/run.js. That is a stale test marker, not this gameplay failure; its log is retained and no foreign test/source was edited. An optional additional native service-clock suite remained queued behind another heavy owner and was cancelled before execution; no pass is claimed.

Only seven selected reviewed PNGs remain: title, Options, Keyboard Setup, free slot menu, map, selected first node, and first-level TIME'S UP. Compact control/capture metadata, logs, source/media hashes and Git patches are in the run directory. Early action:png captures were stale composited black frames; the run-local controller switched to the standard cmd:png:PATH capture, which repaints. Those black frames are excluded. No production code changed.

Next: owner decision on an explicitly permitted wall-time launch policy before any separately authorized fresh attempt. There is no active runtime to extend.