Native Minesweeper reviewed board, 2026-11-14

One default native beginner board was won naturally, then closed with Game → Exit. I personally reviewed the revealed numbers after all fifteen square clicks, the native fastest-time prompt, the completed board with sunglasses and ten automatic flags, the best-times entry of 311 seconds, the native Exit menu, and the empty desktop after closure. There was no second board. Evidence: scratch/runs/20261114-minesweeper-reviewed-outcome/result.json.

The requested registered app is winmine_wep, using test/binaries/entertainment-pack/winmine.exe: 40,720 bytes, SHA-256 2b554103fb9007c23b6c42eae4b410e294700fb82f84cf846a10e7a2a0d30a7d. The before/after executable hashes and size match exactly. There was no replacement executable, download, alternate Win16/XP image, guest configuration change, hidden-board inspection, seed change, or clock change. This is current Win32 gameplay evidence, separate from historical menu/color-art tests.

tools/minesweeper-controller.py wraps the released, unchanged Barrage transport. It distinguishes personally reviewed bootstrap and actual board scenes, permits one phase-specific native Start before board admission, and requires actual visible default-board review before square gestures. In this run the actual untouched 8×8 board appeared directly at launch and was reviewed before permanent binding: baseline.png, PID 3984050, start token 273812790, main HWND 0x10001, full 800×600 desktop. No bootstrap Start was needed. The reviewed transport maps pointer commands to Win32 left-button down/up. Right flags and chording were refused and unused. All ten flags in the final image were placed by the native game on winning; there is no manual flag claim.

The first click, r1c1, was an honest opening guess and revealed a blue 1. Exactly one of its three neighbors was a mine, with no uniquely safe choice. The second guess, r8c8, opened a blank region across the upper/right board. The remaining thirteen clicks were safe deductions from personally reviewed displayed numbers, never a board reader or solver. Rows and columns below are one-based. This is a retrospective record, not an automated action list.

Click Square Displayed-number reason Personally reviewed consequence
1 r1c1 Opening guess; no numbers Blue 1
2 r8c8 Unconstrained opposite-corner guess Blank flood and numbered boundary
3 r2c2 r3c4's 1 forces r4c3, satisfying r3c3's 1 Blue 1
4 r3c1 r1c3 then forces r1c2, satisfying r2c2's 1 Blank flood down left edge
5 r5c3 r4c2's 1 already satisfied by r4c3 Green 2
6 r6c4 r5c2 forces r6c3; r5c3's 2 satisfied by r4c3/r6c3 Red 3
7 r5c4 Same satisfied r5c3 Red 3
8 r6c5 r4c6 forces r5c5, satisfying r5c6 Red 3
9 r7c5 r6c7 forces r7c6; r6c6 satisfied by r5c5/r7c6 Red 3
10 r8c4 r7c7 forces r8c6; r6c4/r6c5 imply r7c4; r7c5 satisfied Blue 1
11 r8c3 r8c4's 1 satisfied by r7c4 Green 2
12 r8c2 r6c1 forces r7c1/r7c2; r8c3 satisfied by r7c2/r7c4 Green 2
13 r8c1 r8c2's 2 satisfied by r7c1/r7c2 Green 2
14 r7c3 r8c3's 2 already satisfied Red 3
15 r8c5 r7c5's 3 satisfied by r7c4/r7c6/r8c6 Red 3; native fastest-time prompt and sunglasses

Gameplay stopped immediately on that natural win. I dismissed the visible name prompt with its native OK, retaining Anonymous, then dismissed the visible best-times table with its OK. Reset Scores was untouched. These were separately observed, nonfinal result-dialog clicks using the unchanged released Barrage transport inside the irreversible Quit phase; they were not Exit claims. scoresok.png shows the whole won board and its automatic flags. openexit.png shows the actual Game menu with Beginner checked and Exit visible. Selecting Exit closed the game without a further confirmation. There was no smiley click, reset, difficulty change, manual flag, chord, or post-win square input. The win is established by the native result prompt and completed board, not the mine counter alone.

The one recipe was fixed at 900 seconds with an independently prearmed 920-second watchdog, default 1000-block slices and real ticks. Its mechanical 180-second Quit reserve was unchanged; voluntary Quit phase began at 478.265 seconds, before the 720-second gameplay cutoff. Native wait completed at 494.672 seconds; the CLI reported 494.303 seconds. Every consumed square gesture used a personally reviewed scene less than five seconds old at consumption, a 400ms pulse and independently prearmed 1750ms fallback release. Menu/result scenes used thirty seconds. Eight expired square intents were NOT_SENT before DOWN; each used a new capture and new intent afterward. Five preliminary board-admission reviews and one outcome attestation also expired without input. No guard was widened. No mixed/UNKNOWN input occurred, and there was one permanent binding with no replay or rebind.

The final Exit receipt is CONSUMED. Its release result is OWNER_EXITED, with the matching actual native wait and both EOFs inside the finite bound; physical button-up was not acknowledged, because the process had already exited. This is preserved honestly in nativeexit-up.json. All other eighteen consumed gestures have release ACKs. Native, watchdog, supervisor child and outer waits each returned 0 with both EOFs; the watchdog exited on the actual native wait without firing. All owned runtime PIDs were absent and port 8601 was closed. No host Quit or cancellation was used.

Before heavy work, I revalidated FreeCell's four successful actual waits and GLTron's startup-refusal terminal (native 1, other waits 0), both with both EOFs, heavy release and absent PIDs. Under /tmp/wine-heavy-runtime.lock, my own 19 policy tests, 12 HTTP refusal/terminal controls, 10 actual CLI Solitaire fixture controls, and flock /tmp/wine-build.lock bash tools/build.sh all passed. The actual fixture controls cover release after slow persistence, parent death, pipe EOF, late/dropped ACK, physical hold bounds and bootstrap/board policy. They do not count as Minesweeper gameplay. The original loaded no fixture code.

The run pins 311 exact then-current sources/module/media files; postflight found zero drift. Git SHA plus binary tracked/untracked patches retain source provenance without source/media/WASM copies. Eleven selected, personally reviewed PNGs are explicitly named in result.json; forty unselected captures were removed with hashes recorded. The native CLI warned that absent shell32.dll imports fall back to existing WAT stubs. No DLL was downloaded or replaced; this limits any broader API-compatibility claim.

Selected-image CRC/inflate/SHA validation passed for all eleven 800×600 PNGs. The evidence checker reported no errors, missing references, or invalid metadata for this run; foreign evidence issues were left untouched. The separate messageboard-watcher tool session returned 143 without an actual watcher wait/EOF receipt. Its recorded tail PID was absent; foreign watchers were left alone. This ancillary missing proof does not replace or weaken the four recorded actual game-runtime waits.

Box ID: null; state: NOT_CREATED; expiry: null; stop receipt: N/A. Execution was local. No browser, full-27-game, FPS, audio, or public deployment claim is made. No additional gameplay is next for this bounded task.