StarCraft: personally reviewed enemy consequence (2026-11-14)

This is a distinct combat objective on the existing registered starcraft_shareware route (ophelia terran1). The completed Supply Depot construction run is closed and is only a read-only controller reference.

Controller and evidence contract

tools/starcraft-combat-controller.py permits one personally chosen left click per invocation. The released Barrage transport supplies permanent native PID/start-token/window ownership, unique nonces, sticky UNKNOWN and a separately prearmed 1750 ms UP fallback. Right clicks, drags and action lists are refused. Every gameplay/navigation scene expires after five seconds; ordinary native menus expire after thirty. Each input also checks freshness immediately before DOWN. The final 180 seconds of the immutable runtime belong to irreversible Quit; reentering gameplay is refused.

Binding requires a personally reviewed actual native title/menu. A splash or movie cannot be relabelled as a menu, and a later mission scene is not a binding shortcut. Mission admission is separate and requires a current unobstructed mission review.

The combat extension requires the selected armed unit's HUD/weapon, the visible ordinary Attack command, and a separate current visible enemy target after the consumed Attack command. Move/Stop/Hold support finite native navigation; construction is refused. Each camera/minimap click needs its own reviewed intent. There is no recognition, tracking, waypoint list or model action loop; ordinary native AI may continue after an order.

A result record requires a current settled scene after the attack target order. Damage/destruction needs a described persistent consequence. Animation alone cannot qualify. Other units firing forces ambiguous attribution; such a record does not assert this objective was proven. Every result explicitly declines a mission-win claim. These fields are human visual attestations, not automated verification of game pixels or private world state.

The CLI supports capture NAME, bind SCENE, admit-mission SCENE, MODE NAME mousedown:X:Y MILLISECONDS SCENE [QUIT_ROUTE], and result NAME SCENE ATTACK_INTENT. Reviews are append-only SCENE-review.json records tied to the complete raw PNG hash and native identity. See the focused tests for the required review fields. FORSAKEN_CONTROL_OUT selects the unique run directory and FORSAKEN_CONTROL_URL the direct loopback endpoint.

Bounded original and validation

Run-specific evidence belongs in scratch/runs/20261114-starcraft-enemy-consequence/. The original uses one fixed 1200-second registered route, a prearmed independent 1220-second watchdog, real ticks, default 1000 batches and cooperative mode. It follows the actual Pegged and GLvisible terminal waits and heavy-lock release. No download, alternate media, cheat, map seed, speed, AI, guest memory or configuration change is allowed. Source recovery uses exact Git SHA plus binary diffs; media is referenced by existing path and SHA-256.

Observed result

Enemy destruction proven; native Quit proven. Outer-wrapper wait evidence is incomplete. This does not establish a mission win.

After separately reviewed native briefing, Start and Tips dismissal, the actual mission was admitted. The first selected Marine followed finite Move orders while scouting and died before an Attack order; this was a friendly loss, not credit toward the objective. Another supply reduction was observed without attribution. No second mission, cheat or construction substitute was used.

A replacement Marine was personally selected on the eastern platform. Its HUD showed 40/40 HP, Kills: 0, with the weapon panel visible. One reviewed minimap click revealed a red enemy Marine firing beside a burning friendly barracks. At 417.105 seconds I clicked the visible Attack command. A separate current raw scene showed Select Target and that red enemy; at 424.520 seconds I clicked the enemy at the personally reviewed screenshot position.

The following observation showed red death remains where the enemy had stood, with the selected friendly Marine below the barracks, 22/40 HP, Kills: 1. Later settled scenes retained the same selected Marine's kill counter and HP; the enemy remained absent. No other friendly shooter was visible in these scenes. The selected unit's counter increase supplies attribution beyond the mere disappearance of an enemy. I did not sample the selected Marine's muzzle flash; the death remains and persistent native HUD consequence are the proof. A later disappearance without that counter change would not receive the same attribution. The friendly burning barracks is ambient damage and is not claimed as an effect of my order.

Eight selected complete 800×600 PNGs remain, with hashes and personal notes in personal-scene-reviews.json:

Scene Observation
original/bootstrap-current.png Actual native briefing menu, before permanent binding
original/select-infantry.png Admitted unobstructed mission
original/replacement-hud.png Selected replacement Marine, 40/40 HP, Kills 0
original/base-enemy-view.png Visible red enemy, selected armed-unit HUD and Attack command
original/enemy-target-review.png Separate active Attack target review
original/combat-observation.png Red death remains, selected Marine 22/40 HP, Kills 1
original/result-final-persistent.png Settled enemy absence and persistent selected-unit kill
original/quit-confirmation-review.png Native exit confirmation with the result still visible

Transient PNGs were submitted for personal review during operation, validated, and then removed; their metadata/hashes and input receipts remain. There is no frame series, source-tree archive or copied media/WASM in this run. Several reviews expired before admission/arming and sent nothing; a new raw scene and new choice were used each time. No input UNKNOWN, replay or rebind occurred. The original contains 21 consumed finite left clicks, all 400 ms. Gameplay intent ages were 3.442–4.609 seconds and were checked again before DOWN. All 21 independent release children exited 0 with actual waits and both EOFs, and physical UP was acknowledged, including the final confirmation.

Native Quit and missing outer proof

Quit began irreversibly at 488.589 seconds, before the 1020-second gameplay cutoff. Each native step was separately reviewed: MENU → End Mission → Exit Program → “Are you sure you want to exit Starcraft?” → Exit Program. The final confirmation was consumed at 518.388 seconds. The log records guest [Exit] code=0, CLI duration 519.380 seconds, and the actual native wait completed at 519.694 seconds.

Native, watchdog and supervisor child each have actual exit-0 waits with both EOFs. The watchdog exited unfired, the supervisor recorded heavy-lock release, all four relevant process IDs were absent, and port 8607 was closed. No host quit, cancellation request or deadline signal ended the native process.

The launching tool session returned 143 without its Python wrapper writing outer-wait.json. The surviving outer process did record the actual successful supervisor-child wait, but its own wait and separate EOFs cannot be recovered from process absence. original/outer-wait.json and launch-wrapper-missing-proof.json explicitly record actualWait: false, missingProof: true; they do not invent exit 0 or EOFs. Thus combat and native Quit have their own positive evidence, while the requested complete outer closure proof remains missing. The coordinator explicitly released the next independent task on the established terminal/heavy-release evidence while retaining this gap. The closed original was not reopened or rerun.

The native log also retains two secondary-thread EIP=0 diagnostics and 26 abandoned parked critical-section enters; one thread remained active in the final summary. Native Exit Program and process exit 0 do not establish clean shutdown of every emulated thread. No runtime patch was attempted. The ancillary read-only messageboard watcher likewise returned tool status 143 without a separate child/EOF receipt; this is disclosed independently.

Reproducibility and checks

Tested Git base: ee9c89feb84901561d1999d9d5a9ba3fcdbaa8be. Relevant uncommitted runtime hunks and owned new controller/test files are retained as binary patches in the run directory. The final report also has a new-file patch. The tested WASM is referenced by path: 2,200,987 bytes, SHA-256 c94ccf1b179f18bd4b47867445afed8cef2f1393b19aabfd259c7f3b94a7be09 (the authoritative full hash is recorded in launch-pins.json).

Validation passed: 34 own policy tests, 12 actual HTTP tests, 10 actual CLI and child tests, plus flock /tmp/wine-build.lock bash tools/build.sh under the heavy lock. The new combat sequence runs through actual controller CLI children. Fault injection and synthetic review fields belong only to isolated Solitaire fixtures; none was loaded into StarCraft.

Postflight checks found 2,948 pins, zero drift. The unchanged registered ophelia terran1 app and all twelve existing media/system-data dependency files matched the prior pins and their own before/after hashes. The existing missing non-seed shell32.dll/builtin fallback was unchanged. All 64 transient/raw PNGs passed CRC, inflate and 800×600 validation before pruning; the eight selected PNGs were retained. No FPS, audio, browser parity, full-suite compatibility or mission-completion claim follows from this run.

Box ID and expiry are null; state is NOT_CREATED and stop receipt is not applicable. No box or worker was created. Work ends with publication of this controller, its tests and this report; no further native action is required.