MDK2 corrected staging and native scene attempt, 2026-11-04

Corrected small-file staging and fresh prerequisites passed on the same retained box. The single browser reached original New Game and the Medium-highlighted difficulty dialog, then stopped before Medium input at the unchanged menu command cap. No intro comic, native COM_PAUSE consumption, post-cancel request completion or playable scene was observed.

Evidence is in scratch/runs/20261104-mdk2-corrected-staging-native-scene/ (RUN). This follows the staging failure, commit be930bbd6, and uses the tested post-comic continuation, commit 1a207ffb0d02447cea321c66f12d9df93ccd9d79. The long-loading route remains historical evidence, not this run's cancel witness.

Staging and fresh checks

Only 28 direct small inputs were staged into this owned run. The predecessor's corrected runtime_io.py stage_file constructs one absolute repository prefix; its three mocked staging regressions passed after updating the expected owned run name. The predecessor's corrected test-scene-transition.cjs uses memory({ready:true}); predecessor files were not modified. The controller, driver and observer bytes retain their final-input identities. No source, media, module or dependency tree was copied or archived. Puppeteer dependencies were reused through their existing retained path, with one symlink for ESM resolution.

The actual retained repository is /home/user/wine-assembly, established by the predecessor's failure diagnosis and fresh successful file checks. It has no source tools/ prerequisite tree. No such directory was assumed or copied in.

Before Chrome, all 432 historical pins, CommonJS and ESM Puppeteer 24.31.0 imports, registered MDK2 selector/WebGL, isolated HTTP and both full module hashes passed. Historical execution source remains 1329b1ba6eb00144ff77d4cd92a8513d4fc953ad; canonical module cffca869b541e6eff12ae3a6b74c4476977e0a7e23cd43d8ec5ba8faecd4bcee; compat 580f49f3c78d26e6855efa0beb37046d03ea24690ff98cfcb3e94ce5e6cb1ee3; original PE b3e0226e411abff17ddabb723e493e95579490cb49bfe3787b1eade1144e27c1; scripts cfdae6e7d02b1823a59d0b85aa8eaca2f057dbb27e7b86d9a7199fc069f642d8. The complete retained manifest is RUN/runtime-pins.json; today's build was not substituted.

All 177 local helper tests passed. Two repairable remote test prerequisite failures were diagnosed before Chrome: the historical failed-title replay capture is absent, and source decoding requires absent tools/mpq. These tests passed locally; explicit remote name exclusions avoided staging their source/evidence dependencies. A subsequent parallel modeled suite had one late ACK against its existing short test deadline. Running the same applicable suite sequentially, without changing limits or implementation, passed 173/173, including actual controller/driver command code, ACK/barrier, serialized native observer, native readiness, scene, construction, opening and recovery deadline cases. These are modeled tests, not game execution. Initial failures and the passing rerun remain in separate TAP files. No production defect or source fix was demonstrated; no production build was required for this documentation-only commit.

One actual browser and the stopping boundary

Fresh authorization was RUNTIME RELEASE MDK2 TARGET ONWARD, box bx_wm47edh3, expiry 2026-11-04T08:33:42.970Z, closeBy 08:20Z, recoverBy 08:25Z, shutdownOwner coordinator. Chrome15087 launched at about 08:06:34Z under the finite deadline helper and closed normally at 08:12:34.326Z. No second browser or reload occurred.

Input-free startup reached the BioWare splash and original title/legal frame at step3072. One balanced title Escape16/release32 envelope ended at3120. Its actual native opening samples did not show native Escape consumption or a button-timer effect; native-selector-55-escape-boundary.json explicitly reports unresolved consumption and causal attribution. Later original opening animation and menu readiness were observed naturally; they are not attributed to that Escape.

The opening took longer in execution slices than the historical route. Fresh captures observed the original discs opening and the legal overlay disappearing. At3312 native freeze became nil and New Game was visible/highlighted. Review correction: I erroneously reused an opening classification for native-selector-97 despite that PNG already showing the ready menu, then issued one unnecessary input-free16 quantum. The original incorrect approval is preserved. review-corrections.json records the corrected classification; it does not rewrite an approval or retroactively authorize input. The later step3328 New Game review was accurate. This extra observation consumed three menu commands and contributed to the later budget refusal.

At step3376, the single native New Game Enter envelope completed with its native callback effects and balanced release. The personally viewed fresh native-selector-103.png shows Select Difficulty, with Medium highlighted. Its SHA-256 is 22629dec6f5970e2f970cd141bfbf120ff1a5a6820d435fad8e40ffa2512dd2f. Native scheme is menu, current menu diff, Enter raw/previous zero, renderer input empty, and trusted input history balanced.

The menu ledger then held 304 slices /57 commands, from the original shared menu start 08:08:37.804Z. The unchanged Medium helper requires a reservation of nine commands: 57+9 >64. At 08:12:23.027Z, it latched menu command budget exhausted before any Medium state/input command was submitted. The command spool stayed at111 through the rejection; command112 was only orderly quit. No cap/deadline was reset or increased, no guard was bypassed, and no gameplay retry followed. This is a controller allowance boundary, not proof of a game failure.

The requested comic cancel/onward objective remains unfulfilled. There was no comic Escape, COM_PAUSE press sample, cancel witness, Lua Done proof, post-cancel observation, relevant request fate or original playable screenshot. The old skipAttempted ledger was never used as a witness. No guessed controls, forced scene/Lua, guest write/call or other game occurred.

Recovery and handback

validation.json verifies 113 contiguous successful driver ACKs, 65 step commands /3376 slices, maximum64, 22 captures, exactly balanced Escape and Enter pairs, and 200 served files with matching pins. The controller refusal is separate from successful transport ACKs. analyze.py rechecks every recovered browser artifact and those claims.

All 377 closed browser artifacts /12,474,235 bytes were recovered by individual unique path and full SHA-256 before08:13Z. closed-inventory.json and recovery-result.json seal recovery. Chrome exited0 without signal. A subsequent process inventory was empty and debug port41817 was closed. Fresh post-close verification again passed all432 pins and all28 staged identities. Module/media/dependency identities are retained by existing path and hash, without another copy. Git preparation SHA, relevant binary diff, owned small-file patches and artifact seal preserve provenance; Git is the source archive.

Handback: bx_wm47edh3 was left RUNNING / API idle, healthy and hydrated, expiry 2026-11-04T08:33:42.970Z. Coordinator owns shutdown. No allocation, resume, stop, deletion, source-box access, subworker, successor, local browser/heavy game or public deployment occurred. Runtime custody was returned on the messageboard immediately after closure/recovery. A future separately authorized attempt must address the demonstrated finite menu allowance/route accounting before browser launch; this worker makes no successor attempt. Single task stopped.