Original Reversi: completed natural loss and native Exit

The single native game completed with “You Lost by 25”. The computer had 25 blue discs, the human had zero red discs, and 39 squares were empty. The game ended after the native “You must Pass” banner and an ordinary Game → Pass selection. Game → Exit then closed the native application: actual exit zero after 315.685 seconds, with both output streams at EOF.

Evidence: scratch/runs/20261114-reversi-reviewed-outcome/result.json. This uses the existing registered reversi app; there is no separate candidate corpus row. No alternate package, download, second deal, New Game, guest state write, rule/AI/configuration change, or emulator fix was used.

Original, route and visual evidence

test/binaries/entertainment-pack/reversi.exe was verified before and after: 28,944 bytes, SHA-256 62c4fa62b8bd5c8912b0659e420e370d75c575ca31675fa0afe6546eabcfdcd6. The file is a Win32 PE. Static inspection found USER32, GDI32, KERNEL32 and MSVCRT imports; the existing registered MSVCRT image imports only KERNEL32. Both original executable and CRT are pinned in dependency-pins.json.

The run used node test/run.js --app=reversi, the ordinary native Win32 USER/GDI desktop renderer, an 800×600 canvas, default 1000-block batches, real ticks, cooperative threads, and --no-close. There were no startup input sequences or injected fixture modules in the original. The actual native window was 400×480 at (20,20). Clicks came from the current full-scene pixels and window snapshot: the eight-cell grid spanned approximately x=65…377, y=124…436, at 39 pixels per cell. The registry's differently sized geometry comment was not used to determine clicks.

Selected images, all personally inspected and named by result.json:

Ten legal human placements were chosen individually from native scenes. Each computer response settled before the next review. Four stale precision intents were positively rejected locally before arming/DOWN; each subsequent attempt used a new scene and new intent. No UNKNOWN input occurred, no input was replayed, and the permanent PID/start-token/window/owner binding was not renewed. The maximum consumed placement scene age was 4.697 seconds.

Controller, checks and closure

tools/reversi-controller.py adds a single-click manual policy over the released, unchanged tools/barrage-guard.py transport. It has no board reader, move generator, solver, gameplay loop or score manipulation. Binding requires a submitted, personally reviewed native board. A placement requires a personal legal-bracket explanation, settled opponent, and exact click center from the current full scene. Precision/menu bounds are five/thirty seconds. Pulses are at most 1500ms, with the transport's independent prearmed 1750ms UP margin. Mixed or UNKNOWN receipts close input; only matching all-unsent refusals allow a new scene and intent.

The original had a fixed 1200-second runtime and independent 1220-second watchdog, armed observably before process launch. The mechanical gameplay cutoff reserved 180 seconds for irreversible Quit. Quit began 307.772 seconds after launch, well before the 1020-second cutoff. The native Game menu and Exit selection were separately reviewed. Every nonterminal gesture had a release ACK; their largest DOWN-to-release-completion upper bound was 433.778ms. The final Exit retired the native process before physical UP could be ACKed: exit-up.json accurately records OWNER_EXITED, actual normal wait and both EOFs, with physicalUpACK:false. Port closure alone was not used as proof.

Native, watchdog, supervisor/outer-child and outer waits all completed with exit zero and both EOFs. The watchdog observed actual native termination and did not fire. The heavy lock was released, the native PID was absent and control port 8593 was closed. No host shutdown was needed. No remote box was created: box ID/expiry are null, state NOT_CREATED, stop receipt not applicable.

Before the original, checks passed under the shared heavy lock:

The first policy-test draft incorrectly reused a Quit review for a menu assertion; the test was corrected to use a fresh menu review. The runtime policy was not relaxed. Final production tests started after the final controller change. No causal public runtime fault was observed.

The tested WASM was 2,200,987 bytes, SHA-256 c94ccf1b179f18bd4b47867445afed8cef2f1393b19aabfd259c7f3b94a7be09. launch-pins.json records the exact Git base and 2923 source/media/module hashes; postflight found zero drift. Git plus binary dirty/new-file patches preserves source provenance. Evidence contains no source-tree archive, media copy or WASM copy. Unselected transient PNGs were removed after review; nine semantic captures remain, all named by result.json.

The game/outcome/Exit goal is complete. This run makes no full-27, FPS, audio, browser parity or general compatibility claim. No further original is needed.