Hock-A-Loogie: original package and runtime evidence

Current status (2026-10-21): bounded original-game startup, input and visible 0/20 → 5/20 scoring evidence is accepted on the explicit-EXE route, integrated at 2e2c363e8. Candidate registration and the provenance-based BG corpus gate correction are independently accepted together. Actual corpus list/dry-run now reports Hock LOCAL with zero skips; original extraction, all 39 file hashes and 38 companion mappings pass. See the combined acceptance. No app-selector or registered-route runtime qualification is claimed.

Historical startup placement repair (2026-10-21)

The original BBLOOGIE.EXE now leaves its startup sizing loop and displays the "Choose Beavis or Butt-head to start" screen. This is a bounded startup repair; input, gameplay and audio are untested. Independent review is required before integration. No main merge or public deployment was performed by that task.

Cause and repair

First checked trace semantics on exact main 7e5cf3535946e0a7bf42179b728a8471dc9f6b67: test/run.js formats call arguments inside h.log before dispatch, but formats out: parameters inside h.log_api_exit afterward. Entry-time pointer contents are not return values. The baseline's out records really show an outer rectangle (358,0,358,480) with client size 632×434. This was a feedback-loop consequence, not evidence that GetWindowRect itself wrote the wrong rectangle.

The guest sizing function at 0x004067d0 computes non-client extents from GetWindowRect minus GetClientRect, scales its desired dimensions to the screen, calls SetWindowPlacement, and compares the resulting client dimensions at 0x00406919. Mismatch branches at 0x00406921 / 0x0040692b return to 0x0040681a. Source and disassembly evidence are sealed with the run.

$handle_SetWindowPlacement moved the host's outer rectangle but omitted the CLIENT_RECT refresh used by SetWindowPos. get_window_client_size in lib/host-window.js prefers WAT's initialized client rectangle, so the guest continued to see 632×434 while its outer width shrank. The renderer's own client computation also preferred that stale WAT record.

Ten added lines in src/09a5b-handlers-window-late.wat recalculate the standard non-client geometry after the host move and synchronize the host client rectangle before the existing show-state path. The repair uses existing helpers, introduces no callbacks, and preserves existing show-state handling. No game-specific condition, output, API stub or asset change is involved.

Verification

• New test/test-window-placement-client.js runs the real source-compiled WAT handler and renderer imports without a game binary. It checks 15 growth, shrink, move-only, repeat and client-size convergence cases across visible captioned, visible borderless and initially hidden windows, including stdcall cleanup and renderer client origin/extent.
• Exact-main baseline fails: client 392×272 rather than 500×348 after resize. Candidate passes. Moving recalculation before the host move fails the same assertion; deleting host synchronization fails the renderer-mirror assertion. Both controls exit 1 with AssertionError, and candidate source is restored.
• Existing child first-show, secondary top-level first-show and ShowWindowAsync tests pass. Owner gate verifies 267 owners; callback census verifies all 401 boundaries unchanged. Fragment, A/W and test-tier gates pass, without refresh.
• Canonical remote bash tools/build.sh passes all gates, Boat cmd_uydlnnqxmerz. Heavy compilations and runtime tests were remote only.
• Original 39-file assets verified. Candidate diagnostic cmd_exlluiryqq9s shows two SetWindowPlacement calls; returned client width changes 632 → 561 → 556, then guest advances beyond the loop. Outer rect settles at (38,0,602,480). Five-second/250-batch and 30-second captures have identical SHA256 98f22f8f8e19a6c2cbece7092db349e0618513c070bebb5ea758debca839e649.
• The 30-second run cmd_vdlyav4tmyy9 completes 3315 batches / 75,746 API calls; baseline had 138,161 batches / 86,305,741 calls and remained solid teal. Both baseline and candidate screenshots were actually inspected. Candidate shows title/menu chrome and the start screen; the right character is partly clipped, so this does not establish complete rendering or playability.

The existing missing-comctl32.dll warning remains; neither run trapped on a missing ordinal. It is not the cause of this placement loop. No unrelated fix was attempted. An initial full-file upload hit the local argv-size limit and never reached Boat; its prematurely labeled candidate test reran baseline and failed. The subsequent patch upload and source/module seals distinguish the actual passing candidate. Local and remote patch text differs only in Git's object-ID abbreviation length, independently normalized and verified.

Identity and handoff

Accepted baseline source tree: 934a1ecde8ab26582e5dfc0b911f2ea2dc96adae. Candidate differs in one tracked source file plus the new regression test; all 7,132 baseline tracked-file hashes were compared. Reports are not compiled.

Run: scratch/runs/20261021-hock-placement-loop-repair (receipt, result, RECOVERY.md, command receipts, complete source/package seals, logs, captures). Compact committed evidence: ../reviews/hock-placement-loop-repair-20261021-evidence.tgz. Baseline evidence remains in 20261021-hock-accepted-canonical-runtime, published as d71f4ca951b40be17e060c0d36afeb062265839d on task/hock-accepted-canonical-runtime-oct21.

SAME running Boat bx_3f3x23m3, TTL 05:56:19.894Z, Node v24.18.1 / CLI fullVM / JS raster canvas. Baseline root /home/user/hock-accepted-canonical-runtime is untouched. Candidate root /home/user/hock-placement-loop-repair retains repo, dependencies, modules, logs, mutation driver and recovery seals. Coordinator alone owns lifecycle; no stop/snapshot/resume/fork/create/delete was issued. Task branch: task/hock-placement-loop-repair-oct21. No workers were spawned.

Original scoring evidence (2026-10-21)

The later qualified module on source 776d37d5 (identical production inputs to integrated d2034ccf) accepts real left-mouse charge/release input and shows score 0/20 → 5/20. See the scoring runtime report for the two bounded routes, screenshots, seals and recovery. 0x401c20 starts the charge; 0x402ce9–0x402d21 holds/releases it, capped at 61 logical ticks; 0x403060 uses charge for projectile travel. Read-only score cell is 0x41a030. The package lacks the EXE's named ..\Bbgame.hlp. Exact struck target and collision instant, completion, sound quality, FPS and Windows parity remain unqualified. No guest bytes/state or production source were changed.

Original candidate package contract (2026-10-21)

The accepted hock-a-loogie-1995 candidate recipe is local-only proprietary Viacom New Media material; it grants no redistribution or public deployment. Source is the archival wrapper at https://archive.org/download/Hock-A-Loogie-1995/beavislg.zip, not an authenticated publisher master. Version resources say 1.00, fixed 1.0.1.0, internal/original bbmini.exe; distributed executable is BBLOOGIE.EXE.

Retained intake: scratch/runs/20261020-hock-loogie-original-intake-oct20/. Its SHA256SUMS digest is 31230f2aa354498853120b39417e1385d5f218e623bdf15e4493b49b9db9dfa7. Outer archive: 1,256,578 bytes, SHA-256 e10d150131258641af11609249ac7f1c430e67e4138943d175b0365ba64be00b, SHA-1 68be591b75813f81409314ad1944d64927835be5. Extract to distribution, then extract distribution/BEAVISLG.ZIP (1,254,275 bytes, SHA-256 24da14e65b7a5418f9f4099a8ec6dcf6998004c350f365936b8d9ef0b6a3f05e) to game. Outer survey HTML/text are not game assets.

The original package has 39 files / 2,224,918 bytes: EXE, BBLOOGIE.000, SOS9502.DLL and 36 AIFs. Every name/size/SHA-256 is pinned in test/test-hock-a-loogie-local-registration.js. EXE SHA-256 is 974f1b5ff82cfdb87f05b9840ae03cd2ed6ceebe2c8ed547253dadaf1a871ef4; SOS9502.DLL SHA-256 is a706691cd04bc66fd290977fc5a863b99c004e06f8980b6fc8a04d2bc4f26b7e. The candidate fixture root is test/binaries/candidates/hock-a-loogie-1995/. CLI vfsInclude: ["**/*"] resolves all files beside game/BBLOOGIE.EXE; the runner roots relative paths at C:\. Browser inventory excludes the separately loaded EXE and maps all 38 companions to c:\<filename>. No registry seeds or DLL substitutions. No durable fixture installation.

PE32 i386 GUI, image base 0x400000, entry RVA 0x1366f, subsystem 4.0. Imports include COMCTL32 ordinal 6; the supplied SOS DLL uses waveOut, multimedia timers, MIDI and threading. Strings naming absent sos32s02.dll and DISPDIB.DLL do not justify substitutions. Registry strings Software\ViacomNewMedia\BeavisAndButthead and LoogieHiScore do not establish a preseed requirement or persistence acceptance.

Run the registration regression with the retained intake (no build/runtime):

node test/test-hock-a-loogie-local-registration.js --intake=/home/user/wine-assembly/scratch/runs/20261020-hock-loogie-original-intake-oct20

Without the intake option only metadata checks run and package checks explicitly skip. The historical registration review records actual fetcher/planner/VFS checks and the then-unresolved corpus blocker, now resolved by the combined acceptance. Historical runtime/publication holds in saved candidate 86ffa207 are superseded; this task has no runtime allocation. Accepted startup/input/scoring evidence above is from an explicit-EXE route, not this registry recipe. Completion, persistence, audio quality, FPS, Windows parity and registered/browser route runtime acceptance remain unqualified.

Registered CLI timeout repair (2026-10-21; branch review pending)

On exact main d4ba733 the registered 3000 × 100000-block smoke overspent its 120-second deadline while repeatedly running the already-visible startup loop. Measured batches returned normally; this was not a single-call hang. Reusing the prior successful driver's normal 1000-block quantum retains all 3000 batches and completes the actual registered child in 23.307 seconds. The corpus now uses SIGKILL at its unchanged deadline and matches complete CRASH/STUCK tokens, not the conditional missing-DLL warning word “crashes”. Genuine fatal and timeout controls remain blocked. The oversized-workload actual registered control returns ETIMEDOUT/SIGKILL rather than hanging past the deadline.

The actual test/test-cli-candidate-corpus.js --id=hock-a-loogie-1995 --strict route passes 1 READY / zero blocked/skipped/harness failures. A separate replay adding only the normal PNG output flag shows the original character-selection screen; that image matches the explicit-EXE diagnostic captures. Canonical all-gates rebuild retains SHA256 728e1de8…, original 39 files / 38 companions remain intact. No registered input/scoring, browser route, completion, audio quality, persistence, FPS or Windows parity claim follows. Missing-comctl32 advisory remains visible. See the repair report for exact source/module identity, negative controls, failed exploratory steps, evidence and same-box recovery. Independent review is required before merging task/hock-registered-timeout-repair-oct21.

Offline scoring publication qualification (2026-10-25)

The scoring report and exact sealed archive were already present on pinned main eeb4b337; independent review accepts historical explicit-EXE real-input 0/20 → 5/20 evidence only. The 5,017-input mapping joins historical runtime 776d37d5 to historical main d2034ccf, not today's build. Exact target/collision, completion, audio, FPS, Windows parity and broad goal completion remain unqualified. No runtime access, new build or registration qualification was performed; the former scoring box is coordinator-reported stopped.