Original registered Mad Bomber: catch and score qualified

On exact main 72e7a30d6e36df896eb25007b6139d750e715720, the ordinary registered --app=madbomber route reaches One Player, drops bombs, moves buckets with mouse input, catches a bomb with a visible blue splash, and advances the score from 0 to 1. No runtime blocker occurred in two bounded 60-second runs. This qualifies the requested bounded gameplay/catch/scoring behavior, not full-game completion, sustained play, browser, FPS, or audio quality/synchronization.

Observed behavior

Run 1: 2026-10-21 16:31:20.773–16:32:21.091 UTC (Boat command cmd_l0hcin8zdr4w). The 3-second capture shows the menu, 5 seconds shows the score-0 playfield after Enter, and 8 seconds shows multiple falling bombs and buckets moved from the left edge to the center. The predetermined mouse path misses the bombs, loses a bucket, and stays at score 0. No scoring acceptance is claimed from this run.

Run 2: 16:32:34.163–16:33:34.523 UTC (cmd_ld480wv6ceuc). The second ordinary input plan places buckets at x=532 before Space starts the bombs. At 7.8 and 8.2 seconds, the score is 0 and bombs descend toward the buckets. At 8.5 seconds, blue splash pixels appear above the top bucket beneath the bomb previously visible near x=546. At 9 seconds, the score is visibly 1 and the buckets have moved to x=408; at 10 seconds they are at x=346 with score 1. Other bombs were missed, and Game Over appears by 30 seconds. The controller's 8.5-second move and capture are adjacent asynchronous commands; the image still shows the previous bucket position. The image and acknowledged input log are the evidence, not an assumed instantaneous move.

Both processes exit 0 through the normal CLI bound, without controller guard termination. Postflight confirms both process groups absent. Inputs are only ordinary balanced Enter/Space key pairs and mouse moves. No guest memory or state edits, patches, replacement EXE, bypass, or stubs were used.

Source, build, and assets

All 7,451 exact-main Git blobs verified before and after execution, with a clean remote Git status. Module SHA256: 728e1de80ec20a284fc41b339895ee8ac29f16e4ded83325b27576cfacd4466c (1,878,508 bytes). The preserved canonical module was reused, not rebuilt. Its build source 30a2c3fb1b5c2bfa8a6b001d7394d5cf50df295d, source manifest, successful canonical build log/exit, and module match prior seals. Complete build-source/current-source manifest comparison proves identical production and build inputs. Differences are docs/evidence/status and the unrelated Army Men observer/test files; neither is loaded by this route.

Fresh cheap canonical gates pass: CLAUDE size, WAT manifest, fragment balance, and WATX compiler provenance (23 required files). The three Mad Bomber corpus registration tests pass. These are source checks, distinct from the observed gameplay. One initial invocation mistyped the provenance script name and failed MODULE_NOT_FOUND; the correct canonical script then passed. Both logs are retained. No failing source gate was repaired or bypassed.

All 168 staged pins verify before and after: 140 original files, the original support MSVCRT, and 27 boot assets. No closure item is missing or substituted. Original archive SHA256: b5353b73674a1af6109cb9bf3af5f7d0383308b7ef0b47d52139813ba3ee98ef (3,069,702 bytes), retained at the supplied local path without redownload.

Environment: assigned Boat bx_jh469zy9, Linux x64, Node 24.18.1, CLI headless renderer, cooperative --no-threads, 640×480, real ticks and calendar. Remote lockfile installation and 2,395 actual dependency-file hashes are retained. No browser, FPS, or audio quality claim is made.

Evidence and handoff

Run folder: scratch/runs/20261021-madbomber-current-main-gameplay/. evidence.tar.gz holds 63 sealed artifacts, including all screenshots, timestamped launch/input receipts, logs, terminal/postflight records, source/build/dependency manifests and tested module. Archive SHA256: 22121970add638c66203d278b55ee734f668abf36a389cdec2c9c901298d673c. verify.py checks archive/seals, source equivalence, module, closure counts, normal exits and ordinary input acknowledgements. RECOVERY.md gives exact recovery paths and commands. Inspect run2/sample-8.2s.png, run2/sample-8.5s.png, and run2/sample-9s.png for the catch/score sequence.

Remote source /home/user/madbomber-source and evidence /home/user/madbomber-evidence remain intact. The SAME assigned box is left RUNNING (API idle, health ok), expiry 2026-10-21T18:26:26.518Z. Coordinator owns shutdown and any subsequent task. No additional worker, runtime owner, deployment, shared source/status/index/config edit, or main merge occurred.

Local work used a sparse evidence-only task worktree, with full source and dependencies remote. Shared free-space dips temporarily blocked local writes; work resumed after actual recovery, without deleting unrelated data. A tiny transport-helper write followed one failed admission check due to tool sequencing; this deviation is disclosed, and subsequent writes used explicit reserve guards. API transfer once returned 502; exact partial-byte readback allowed continuation, then the complete archive hash verified. Neither event was a guest runtime failure. Total task allocation remains below 64 MiB.