Penguin Command independent gameplay review, 2026-10-24

Accepted: bounded original Normal Game progression on the retained historical module. Producer d177cdd2a58fef05ae84bbf9c6cd1e845fb404fc, published on task/penguin-command-scheduled-input-gameplay-oct24, supplies 54 unchanged report/evidence files. This independent review adds only this note; no guest, browser, build or native module was executed again. The producer's strict-SSH remote branch SHA matched the reviewed commit.

What the evidence establishes

I independently inspected menu.png, after-selection.png, aim-fire-left.png, aim-fire-centre.png, aim-fire-right.png and final-scheduled.png in the adjacent penguin-command-scheduled-input-gameplay-20261024/run/ directory:

• Batch 32,000 shows the original menu. The scheduled mouse move/down/up at 33,000/34,000/34,500 selects Normal Game.
• Batch 45,000 visibly shows Score: 160 / Wave Scoring and city graphics.
• Batch 60,000 shows Score: 240 / Starting Wave #2 / Defend Cities!, with the red aiming cross at desktop (260,220).
• Batch 85,000 shows Hiscores. Batch 120,000 shows the menu with the cross at (550,250); the final reviewed menu has the cross at (330,160).

All 22 scheduled events appear at their requested batches in the actual 8,039-byte output: 15 ordinary mouse/key actions and seven PNG captures. The exact run/command.json input argument matches scheduled-inputs.json. Mouse messages target hwnd 0x10001; all six A/S/D down/up messages report keyboard focus on that window. The original release README documents mouse aiming and left/middle/right cannons on mouse buttons or A/S/D. Its SHA-256 is df522480c6c682c937ad16c404832625d2246feed60b73d3785c1b664ba5db8b. The empty live-control input-events.json does not contradict the scheduled delivery recorded by argv and output.

Unknown: individual cannon/projectile causality, a player-caused kill or score increment, sustained real-time play, full-game completion, audio, full-client coverage, current-main compatibility, and guest exit. The score change crosses wave scoring; it does not isolate the A press. No frame isolates the S press, and D and the last click follow the Hiscores frame. The green line is not firing proof. --nosound disables sound. The 640x480 canvas clips the client of the original window at (20,20). Guest-clock 6,400,000 ms at the menu and 12,000,000 ms at Wave #2 are accelerated batch time, not elapsed gameplay duration.

Identity, recovery and controls checked

All 50 final seals and eight image identities passed independent hashing. The 54 producer files match their original Git blobs byte for byte, including raw output, recipes and CRLF game options. All 287 recovery manifest inputs were independently read and rehashed: 15,719,953 bytes, unique destination paths, maximum file 1,879,746 bytes. This includes the seven original DLLs, 83 graphics and codec/font/host dependencies. Recovery uses the original absolute immutable donors and exact historical Git revision, not a new build; those standing inputs remain necessary. The producer report retains the fresh-path recovery order and exact recipes. This is investigation evidence; there is no Penguin dashboard candidate card.

The qualified source is b41a7c5f4b752d1f0bdab1141a11d4fed4dc6f9a, module SHA-256 8fac39936b6df2418ad86cab118cef19a65a727f649fbc14c412b37818904686, runner 4d7b9587676db38fcf3314807acb2b2a42b5f09d2c98fcf3dccb8a6cde9e1809, original EXE 14d6e430c26c44c94ef5c176eca2f66edd82dd66fd155d29a5596e3d6e170fc1. Successful destination rehash, production codec/font/renderer probe and separate 291-import ABI instantiation all precede the sole guest launch. The module uses the complete 9a3bc9149d58209c layout contract. All seven actual DLL initializers returned success. ABI probe placeholders were not the gameplay host. Transfer receipts show 431 requests, maximum 35,976 bytes, with 18,000-byte data segments.

The corrected admission-before-guest explicitly retains 5,327,752 bytes of previous history: 65,087,897 / 67,108,864 bytes. Final projection is 62,374,522 bytes with those charges retained. Both include the 8 MiB reserve and meet the host free-space floor. The supervisor's disk assertion still uses an earlier 56,814,417-byte projection; the separate corrected admission precedes launch and passes using the larger complete allocation. Raw recipes are preserved rather than retroactively corrected.

The actual runner reaches the 200,000-batch limit in 7.136 seconds; the supervisor exits naturally with host status 0 in 7.330 seconds, no TERM, KILL or remaining group. The final remote process census is empty. This establishes bounded host cleanup, not the game's own exit path. The observed coordinator watchdog PID/source/deadline is retained; its shutdown had not executed at producer handback. The later coordinator receipt scratch/usage-watch/bx_ts27jq5e-early-closeout-oct24.json records the box stopped at 02:57:37 UTC, snapshot verified at 02:57:37.284 UTC. Original expiry was 03:18:12.037 UTC. This reviewer made no box operation; lifecycle ownership remains with the coordinator.

Producer-wide whitespace checking reports the raw CRLF game options and one space-only line in supervise.py; these exact evidence bytes are retained. Authored report/review prose passes its scoped whitespace check. Independent receipts, manifest verification and strict-SSH publication/resource records are retained under /tmp/penguin-gameplay-independent-integrate-oct24/. Integration reconciles actual remote main and checks the final 55-file scope under scratch/usage-watch/shared-integration.lock; root tracked files, index, refs and all donor repositories remain read-only.