Dungeon Keeper original PNG runtime, 7 November 2026
PNG staging and actual Canvas2D presentation now work on the original demo. I reviewed the intro and authentic Main Menu. The first gameplay blocker is native menu input does not move the game's software cursor to Start New Game: the acknowledged click leaves the menu unchanged. No dungeon, digging, building, persistent consequence, FPS, audio or parity was demonstrated.
Evidence is in scratch/runs/20261107-dungeon-keeper-png-original/.
Published/released prerequisites c790ca544 and 7f4486b53 preceded allocation.
The retained original driver imports the accepted Canvas2D binding and uses
the unchanged registered dungeon_keeper_demo route. The fresh base manifest
was augmented with the retained stage.cjs complete pngjs closure before
upload: 644 files, including the staging verifier. Every SHA/length was
checked before transfer and remotely; isolated stage.cjs --verify passed
before Chrome. No sample adaptations or accepted fixture matrices ran.
I personally reviewed these fresh owner-scoped immutable captures:
review/startup.png: red circular intro vortex on black.review/after-intro.png: Main Menu with Start New Game, Continue Game, Load Game, Multiplayer, Options, High Score Table and Quit; imp cursor near Multiplayer. Ordinary Escape successfully reached this menu.review/after-start-new.pngandreview/after-menu-enter.png: the same menu and imp position after the click and Enter respectively. All three menu captures have SHA-256f1cd5f82862e641363e1af66a55a3d7b444827b3b6878f5d8465b3e13464a88d.
Command 1 clicked the reviewed Start New Game location at normalized
(0.5,0.265). Trusted Chromium movement/DOWN/UP events were recorded; both
input ACKs succeeded, with UP independently armed before DOWN and attempted
101 ms later. Renderer coordinates became (320,127), buttons returned to
zero, and the queue drained. The screenshot at 14:30:25.606Z remained
unchanged. Command 2's ordinary Enter also completed with balanced trusted
events and no menu transition. Commands 0–3 have exact numbered ACKs; command
3 closed the browser. No input outcome was UNKNOWN and none was replayed.
The passive owner-checked snapshot at 14:31:26.340Z in
recovered/passive-cursor.json reads the original game's documented variables:
software cursor (320,240), last mouse position (320,127), relative-mode
flag 1. Both guest threads were active and natural publication advanced.
Thus this is a specific game-cursor versus native-position mismatch, not a
capture failure or an assertion that the guest crashed. The existing
dungeon-keeper-demo.md documents dropped single-position moves and continuous
small-motion steering. That is a plausible explanation here; semaphore timing
was not instrumented, so the exact internal drop cause remains unproven.
No handler, guest memory, clock, installed asset or configuration was edited.
Staging records main source SHA
0c71c5a625fad675e61ddda0942616a9fe6aa7fd, relevant binary patches and
run-helper recovery patches. Relevant retained source/media pins showed zero
drift, and all 168 media/support pins and all 644 deployed pins remained
unchanged. Media stays at its existing paths with SHA/length references.
Chrome 151.0.7922.108 executed page and guest module SHA-256
e41511fe861363fbe1639d80f20cbcdd7240f46937c93a801eaa256ca094c467,
recovered from CDP. This identifies the executing module without claiming a
fresh build. The actual backend was page-owned Win98Renderer/Canvas2D,
process 1000, HWND 98305, with no D3D endpoint. Strict 22-field continuity,
publication, immutable-hash review, freshness and sticky failure gates stayed
intact. Source recovery uses Git and patches; no source/media archive is retained.
One UUID-idempotent POST created bx_6g45252h. The immutable runtime deadline was 14:54:38.631Z, fixed at 14:29:38.631Z. Chrome exited normally at 14:31:35.655Z, code 0. Initial custody validation failed before preserving its individual predicate values; a subsequent passive observation and completed inventory verified both owned PIDs absent, no matching live processes, both ports closed, profile/writer absent and held input empty. No runtime or input was restarted. SHA/length recovery verifies 68 artifact paths / 3,210,537 bytes, including captures, logs, module and passive observation.
Under the explicit shutdown delegation, normal stop followed recovery at
14:32:37.118Z. Final state is ARCHIVED, stop completed at
14:32:37.505Z, forced: false, current expiry none; original running expiry
was 15:09:15.538Z. No old box was touched and no lifecycle ownership remains.
The one-second local monitor observed minimum free disk 2,775,019,520 bytes;
the remote 2 GB guard also remained active. Neither recorded a floor event.
Checks cover driver syntax, augmented deployment verification, unchanged media/source pins, recovered SHA/length, balanced input and exact ACKs, cleanup, owned JSON/evidence containment and documentation whitespace. No production code changed, so no new emulator regression or build was warranted.
Next: separately qualify bounded native continuous mouse motion in the retained driver, preserving review/identity gates and independently scheduled releases, before another authorized original runtime. This single attempt stops at the menu cursor blocker; gameplay success remains unclaimed.