Penguin Command matched ordinary-firing experiment, 2026-10-24

Historical source last updated, 2026-10-24. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.

Immediate firing/projectile causality remains unresolved. One original Normal Game control and one matched A-key variation reached visible gameplay. A down/up reached the focused SDL window. Their first differing reviewed capture shows different green lines with Score 0 in both, but no isolated friendly projectile, blast or ammunition change. This experiment rejects a claim of proved immediate firing under this schedule; it does not show that firing is broken. No third run was made.

Evidence and executed recipes are adjacent in penguin-command-controlled-firing-gameplay-20261024/. Worker session 01a16be7-16d7-7430-8d0e-2bb4c0be527d owns this bounded task on task/penguin-command-controlled-firing-gameplay-oct24. The explicit sparse checkout starts at accepted main 3f75d7afdfc93819d4874b2cf126ca83e66d5bd2, confirmed against actual remote main using strict SSH. No main merge occurs. Root tracked files/index/refs and all input donors were read-only.

Matched actions and actual delivery

The authentic bundled README (original-controls.txt) specifies mouse aim and A or left mouse for the left cannon. Both runs use identical original assets, runner/module, CLI settings, normal-game entry, aim, batch-clock and screenshot schedule. The only additional ordinary events in firing/command.json are A down at batch 34,800 and A up at 34,900. No config, media, guest state or memory was patched. All events were queued in argv before each launch.

Batch Both runs, except the stated variation
32,000 Original menu capture
33,000 / 34,000 / 34,500 Move (420,360), left down, left up to select Normal Game
34,510 Capture, still menu
34,600 Same fixed aim (260,220)
34,790 Capture before scheduled A down, still menu
34,800 / 34,900 Variation only: A down/up, balanced
34,810 / 34,850 / 34,910 Adjacent captures, still menu
35,000 / 35,200 / 35,500 / 36,000 Captures, unchanged menu
37,000 Identical paired frame, before logged release/aim/key delivery
40,000 Both Score 0, red aiming cross at (260,220); green-line geometry differs
45,000 / exit Wave Scoring; control Score 160, variation Score 80

The guest did not consume ordinary events immediately at their scheduled batches. Both full logs place WM_LBUTTONUP and WM_MOUSEMOVE after the 37,000 capture. The variation then records WM_KEYDOWN 0x100 / VK 0x41 and WM_KEYUP 0x101, each routed to keyboard focus 0x10001, before the 40,000 capture. Exact delivery batch/time inside that interval is unknown; 6,960,000/6,980,000 ms are scheduled key clock values, not delivery timestamps. The 37,000 and 40,000 captures log guest times 7,400,000 and 8,000,000 ms. This delayed consumption invalidates the intended adjacent firing window.

matched-analysis.json decodes and compares all 28 original PNGs. Eleven paired captures (including the menu and every capture through 37,000) have zero differing pixels. At 40,000 there are 2,378 different pixels; at 45,000 and exit there are 10,153. Visual inspection of the control 35,200, both 40,000 and both 45,000 frames found no isolated friendly shot/blast. A change after an input is insufficient to distinguish firing from input-induced guest progress/timing differences at these batch boundaries. Scores are observed during wave scoring and cannot establish a particular shot or kill.

The original window starts at desktop (20,20), with the client below the title bar. The 640x480 host canvas clips the right and bottom of the game; cannon and ammunition areas are below the reviewed viewport. The captures preserve the visible upper play area and crosshair but cannot show a complete launch path or ammunition counter. This is the concrete limit on the causal result. A separately assigned experiment would need captures bracketing actual input consumption during play and adequate client-area coverage. There is no current-main qualification, audio, FPS, complete-game or sustained-play claim.

Identity, gates and bounded execution

All local source hashes passed before the one staging transfer. Destination rehash passed for every file before execution. stage.py preserves the accepted serialized streaming method (431 requests, command bytes <=35,975, segment data <=18,000 bytes); no archive or local media duplicate was created. The fresh UID1000 PNG encode/decode, rasterCanvas/renderer, font readability and module layout gate passed. The exact 291-import/512MiB historical ABI acceptance is reused by module hash; no additional native ABI instantiation was performed. runner-preflight-initial.json retains an insufficient literal flag search; the corrected receipt records actual parser spellings and default tick 200. Accepted quiet flags and output guard were checked before guest execution.

Control: API command cmd_3dtbnvbrgyzk, runner 2.593s / 30,619 API calls, supervisor 2.913s. Variation: cmd_ngudujslgeqx, runner 3.039s / 30,328 API calls, supervisor 3.370s. Both host exits were 0 at the batch cap; neither is claimed as a guest exit. Both initialized seven real DLLs without an observed trap/unimplemented API; TERM=false, KILL=false, remainingGroup=false.