SuperTux: terminal menu failure must notify its waiting sender

The one authorized original reached the Slot 1 menu. First-pillar landing, onward gameplay, and the first-level exit remain unverified. The run reproduced a command-lifecycle defect: the composed driver recorded terminal failure and closed normally, while the numbered sender continued waiting for its absent success reply until its entire 30-second ACK deadline. The production composition now publishes an exact-ID ok:false, noReplay:true reply after the consumer records terminal failure, then rethrows the original error to the existing close handler. There is no success or refresh authorization. Existing replies are preserved; notification failure preserves the original cause.

Evidence: scratch/runs/20261109-supertux-early-landing-progress-0212/. The actual original used canonical Git 3b8958347d7ca5cd09fb9817b0e0932bdfb797bb, including d1893e28 sticky menu failure and 82d028903 explicit airborne settle. Local and remote barriers matched 1,441 providers / 61,415,602 bytes, manifest SHA-256 4a6327cb204138a8fabac9f67d16f414759cb26fa7c2f53abd8bfc45d0c86072. CPU SHA-256 was a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295; native observation stayed OFF. Unchanged 0116/0046 accepted controls and the 0124 locked build/menu tests were verified and reused by PATH+HASH. The current singular menu payload additionally passed actual composed offline execution. An initial provider-source mistake and an offline fixture missing root failed before allocation, were corrected, and remain recorded.

I personally inspected the native menu and resulting Slot 1 image. The menu capture was at 02:18:46.482Z; Start Game's Enter was admitted at age 9,441 ms, inside 12 seconds but missing the 8-second aim. The Slot 1 image was captured at 02:18:56.397Z, SHA-256 a96d9137e93ce5baa87466b11741e0c5df7761e3f6589caee1689c2d45b40294. The next action reached terminal menu admission failure at 02:19:08.498Z, 12,101 ms after that original capture. This is consistent with stale admission; the serialized wrapper does not retain the underlying cause, so its exact internal guard timestamp/cause remain UNKNOWN. Only the successful Start Game Enter DOWN/UP appears in recovered native input. No Slot 1 Enter, later game input, or fresh capture followed failure. No grounded/airborne gameplay review or neutral settle was reached. The earlier Slot 1 image was displayed again while polling the pending command; that was a local read, not a new capture.

The failure record is command 2, at 02:19:08.498Z. Its real sender PID 6764 / start 17900 had started at 02:19:08.254Z, but did not exit until 02:19:38.309Z, status 1, after UNKNOWN command ACK deadline. Its actual wait and both stream EOFs are retained, as are those of the two successful remote menu helpers. The repair was made after runtime shutdown and was not loaded into this original. No second original was attempted.

menu-failure-focused-final passed offline with a real Node sender and the exact consumer call extracted from the actual composed production loop. The stale action caused zero effects; the sender received failure and exited 1 in 24 ms, with stdout/stderr EOF. A later request was refused before producing another numbered command. Normal ArrowDown retained its DOWN/UP plus capture and exact success reply. Missing terminal records and existing replies did not produce fabricated notifications. The generic retained transport, 12-second freshness limit, runtime/ACK caps, controller, choice set and independent UP remain unchanged. flock /tmp/wine-build.lock bash tools/build.sh passed; CPU SHA-256 stayed a320. No browser or game ran in the repair tests.

Renderer0146's exact bx_9g93pmg4 ARCHIVED PHYSICAL_DRAINED board release was posted at 02:15:35Z. Its release watcher actually waited 0 and awaited its tail. No peer box was accessed. An independent immutable UUID watchdog was prearmed at 02:16:54Z, UUID 4585f131-191f-4f91-8ee4-713ae86fba67, deadline 02:41:54.659Z. The single box bx_g7xyqr9c was created at 02:17:08Z, TTL 1,486 seconds, initial archiveAfter=02:41:54.248Z; expiresAt was not returned. The original began at 02:17:54.346Z.

Browser.close ACKed and completed normally in 11,788 ms, browser exit 0, signal null, with no escalation. Client exit 0 and both stream EOFs are retained. Kernel custody verified 189 identities / 189 actual reaps / 1,746 events, including before-wait witnesses. Kernel-parent wait was 1, outer wait 0, and local Boat original wait 0, with EOF. Ports 41895, 35443, 8787, writable run FDs, active profile, input/client locks and runtime lock were drained. Physical pending input was zero. The 20 locally recorded command PID/start identities were absent; the watchdog waited 0, its tail -15, and supervision waited 0 with EOF. Historical unrecorded/nonleader-exec, untraced-clone and external-signal limits remain UNKNOWN; these receipts do not repair earlier missing history.

All 96 remote regular files were accounted for: 79 evidence files and 17 pinned references, 72 unique transports / 1,035,171 stored bytes. Stored/original SHA-256, sizes, independent reopen, GNU gzip CRC and exact set passed before normal archive. No inactive profile/cache exclusions were actually needed; the normal close had removed the profile. Free /home stayed above 1.5 GB at recorded checks. Git SHA plus small owned patches preserve the source; no source/media/module archive, peer deletion, agents, parallel runtime, local heavy execution, RAM gate or public deployment was used.

The box was archived at 02:20:28.756Z. Stop receipt stop_5af07274747100b7 completed at 02:20:28.779Z, forced=false, error=null, final archiveAfter=null, expiresAt not returned. Physical release was posted after recovery and watchdog drain. Next gameplay needs a separately authorized original, fresh canonical composition including this repair, and personal frame decisions submitted early enough for the unchanged 12-second bound.