StarCraft rebel-base combat follow-through (2026-11-15)

Partial mission result: personally commanded Attack and its separate enemy target were consumed; enemy damage and rebel-base progress were not proven. Native Exit Program and guest/process exit 0 were proven. Two launching outer wait/EOF receipts are missing. This follows the existing registered mission, not a new title. The original is closed and was not replayed.

Evidence: scratch/runs/20261115-starcraft-rebel-base-combat/result.json. The route was published on main as 826547148f8f2980e468ec83408fe4f85e0cc08a before the original.

Actual mission and personally commanded combat

The current native briefing showed Chau Sara and Destroy the rebel base. Magistrate Collins identified the Sons of Korhal and assigned Alpha Squadron. The unchanged registered starcraft_shareware entry and its existing ophelia terran1 arguments selected STRONGARM. Initial native mission HUD: 250 minerals, 200 gas, supply 12/42. No worker production, gathering or construction command was issued in this attempt.

After separate mission admission and selected-unit reviews, ordinary Move commands scouted the eastern platform and road. The first forward Marine was selected at 40/40, Kills 0. It took damage and disappeared before an Attack order; this is a friendly loss, with no enemy credit.

A second platform Marine was separately selected at 40/40, Kills 0. After moving to the road, a current portrait navigation exposed a red enemy Marine on the dirt to the east. A separately reviewed native Attack command was consumed. A new complete scene showed Select Target, Cancel-only command card, the selected Marine still at 40/40/Kills 0, and the red enemy. One personally chosen click on that enemy was consumed as attack-enemy-two. The earlier expired target preparation sent nothing; it was replaced by a new current scene and new intent.

The following complete observation showed red remains at two locations, an empty friendly selection, supply 10/42, and a surviving red enemy Marine. No selected kill-counter increase or enemy HP comparison was captured. Consequently neither the remains nor the consumed target establish enemy harm attributable to this command. The attacking friendly was lost.

A third Marine selected in the base already had 24/40, Kills 1 before any order to it. That ambient kill receives no commanded credit. Its subsequent Move approach also ended in a friendly loss while a red enemy remained visible. Final supply was 7/42; two other supply losses were unassigned. Later complete scenes showed empty selection and the fogged battlefield. The separately attested after/persistent result is unproven, with objective-progress and mission-completion claims false. No native victory or objective-change screen was observed; the rebel-base objective remains incomplete.

Route and validation

The finite controller already supported manual selection, minimap/portrait navigation, Move, Attack and native Quit. This change adds a distinct enemy-consequence result: the consumed Attack target's before scene, freshly attested after scene, and separate fresh persistent scene must describe the same visible damage/destruction. Native counter/HP evidence, selected-unit identity and other-shooter review are required. Enemy consequence stays separate from objective progress and native mission completion. The actual original did not qualify for a positive consequence record.

Before publication/original: 48 policy, 16 actual HTTP and 10 actual CLI controls, plus flock /tmp/wine-build.lock bash tools/build.sh, passed. Actual controls exercised separate Attack/target and Move/target clicks, consumer dequeue, permanent identity, late/dropped/foreign ACK, sticky UNKNOWN, independent release, actual parent-death subreaper wait/EOF and real-tick progress. Synthetic assertions and fault injection remained in isolated fixtures, never the original. Initial synthetic result-fixture errors were corrected and retained with the passing reruns.

The original contained 29 consumed finite left clicks, each 400 ms, with unique nonces and an independently prearmed 1750 ms UP fallback. All original physical releases were acknowledged. No captures or thinking occurred while held. The route retains 80–1499 ms; inherited 1500 ms midpress capture remains refused. Gameplay scenes expire after five seconds, menus after thirty. Expired preparations sent no input and were replaced by fresh reviews/new intents. No original input UNKNOWN, replay or rebind occurred. Native pause was never observed or asserted; real ticks/default 1000/cooperative execution remained unchanged. No tracking, OCR, solver, strategy script, action queue, private game-state read, frozen runner, configuration/AI/seed/clock write, cheat or console input was used.

Native Quit and incomplete outer proof

Quit began irreversibly before the 1020-second gameplay cutoff. Separate native reviews followed MENU → End Mission → Exit Program → “Are you sure you want to exit Starcraft?” → Exit Program. Restart Mission and Quit Mission were not selected. Final physical UP was acknowledged. Guest [Exit] code=0 and actual native process wait 0 occurred after 935.088 seconds. The independent 1220-second watchdog exited unfired; the native limit was 1200 seconds. There was no original cancellation request or host Quit.

Native, watchdog and supervisor-child waits each returned 0 with both EOFs; the supervisor explicitly released the heavy lock. The evidence audit found 326 available actual waits with both EOFs and 328 recorded PIDs absent. The board tail separately records its regular stdout file closed and stderr EOF, rather than inventing a stdout pipe EOF.

The long original launching tool session returned 143 before its launcher wrote actual-outer-wait.json. The detached outer process did write its actual successful supervisor-child wait, but its own wait and EOF receipt cannot be recovered from absence. The ancillary board-watch launching session likewise returned tool status 143 without collecting its detached outer wait. Its native watcher, tail, watchdog and supervisor-child did shut down with their own actual records. Both gaps are explicitly retained in original-launch-wrapper-missing-proof.json and board-wrapper-missing-proof.json. runtime-release.json sets allRuntimeWaitsComplete: false and bothEOF: false, while distinguishing the available positive records. Tool status 143 is not asserted as either outer process's exit code. No retry or complete outer-closure claim was made.

Native shutdown logs retain 37 parked Enter(s) ABANDONED and final T2 active at EIP 0x7ab2d1; T1/T3 are exited at EIP 0. Historical Storm EIP0/26 abandoned Enter diagnostics remain unresolved. No application-specific suppression, engine shutdown repair or all-threads-clean claim was introduced.

Exact closure and selected evidence

Sims USER281 released source/build/runtime at 04:17:39Z and Q3 visibility at 04:29:24Z. Their 29 and 34 actual wait records, both EOFs, and respectively 87 and 50 absent PIDs were personally verified before heavy qualification.

The actual shared WASM was 2,203,481 bytes, SHA-256 eb1669a7d065664f8c3f0b3ae5649d9d222721b6c9903c125ef22aa5c21160ea. No old 6872 module was substituted. Preflight checked 8,237 pins, including then-current source/WASM, unchanged registered EXE/MPQ/DLL/import graph, TLB/fonts, resolved PATH Node/Python and mapped native host/controller closure. The existing missing shell32 builtin fallback remained unchanged. Postflight found only three foreign light changes: the Q3 visibility controller/test and the Sims thunk test. Their binary/new-file patches are retained separately; runtime source, WASM, registered media and native dependencies were unchanged. Foreign WIP was recorded, not adopted as this worker's fix.

Git SHA plus uncommitted-before.patch, owned route/report patches and postflight-foreign.patch provide source recovery. No source tree, media, module or frame bundle was archived. All 103 original/copy PNGs passed CRC/inflate/800×600 validation before pruning. Twelve selected result-named images remain; selected-copy-map.json records byte-identical origins and hashes for renamed selections. They show the current briefing, mission, selected attacker, separate Attack/target, enemy survival/friendly loss, uncredited ambient-kill baseline, final loss, after/persistent observations, and native exit confirmation. No screenshot was sent through Telegram in this task. Available waits, missing proofs and visual conclusions remain distinct.

No box or worker was created. Final box state NOT_CREATED, ID/expiry/stop receipt null. No deployment occurred; /home remained above 1.5 GB without a RAM gate. Runtime/source/build release was explicitly posted, with the outer proof gaps disclosed. The bounded task stops after this report's publication; no further original is authorized by this result.