MDK2 personally reviewed New Game/Medium reaches input-free Lua completion
One bounded replay of the unchanged original reached separately viewed New Game
and Medium selections, then naturally completed the192320-byte Lua request with
active cleared. The final personally viewed PNG at step3408 remained black.
Level l1, the next105-resource queue and699228-byte request, loading artwork,
comic, tutorial and gameplay were not observed in this execution.
Evidence is scratch/runs/20261104-mdk2-explicit-capture-native-onward/personally-reviewed/
(RUN), separate from the preserved preceding execution in its parent directory.
This follows the repaired explicit personal-review author
at60f22ffb5, the explicit capture repair
at9449784c6 and the historical Lua/Level replay.
Historical Level/queue evidence remains historical.
The current author is byte-identical to the repaired60f22ffb5 author. Its explicit capture selection carries forward the9449784c6 repair; its CLI additionally requires independently supplied current personal-review JSON. The actual main operator calls that actual author with mode, capture and review in every one of seven supported modes. Offline integration passes all seven modes, missing-review rejections and stale capture/hash/step, inspection, inspector, classification and mode controls. Five actual-author controls pass against real PNG/raw/owner sidecars. All54 local classifier/controller tests, six assignment routing tests, actual generated-shot positive, old-suffix negative and two read-only observer tests pass before Chrome.
Preparation encountered only offline fixture/manifests issues: the new nested RUN initially pointed negative tests at its absent historical capture fixture; the inherited explicit-input inventory held a stale release-JSON hash; the old wrapper test checked the already populated predecessor OUT. The final tests use real retained fixtures, regenerate hashes and compile the unchanged wrapper with the actual fresh RUN/OUT. All final checks passed before the sole launch. The failed logs are retained. Remote wrapper/shot/observer checks pass; the54 tests are local coverage, not claimed as remotely rerun.
Qualification verifies432 original runtime pins,30 existing media path/hashes,
explicit small dependency closure and runtime JSON, actual RUN/OUT,
Puppeteer24.31.0 CJS/ESM imports, Chrome executable, cache graph, registered
MDK2/WebGL selection and isolated HTTP module hashes. Only small inputs were
staged. Source remains 1329b1ba6eb00144ff77d4cd92a8513d4fc953ad, canonical
module cffca869b541e6eff12ae3a6b74c4476977e0a7e23cd43d8ec5ba8faecd4bcee
and companion 580f49f3c78d26e6855efa0beb37046d03ea24690ff98cfcb3e94ce5e6cb1ee3.
The actual fixed wrapper remains SHA
b5f85284acebec7339786c655f56929954cd8212e64bce3ff71fcc65f071e787.
It was compiled through its existing sourceForRun interface to the nested RUN;
its bytes and original preload are unchanged.
The allocation record was appended before dispatch: fresh bx_gw6xp8cz, expiry
2026-11-04T14:00:53.351Z, guest cutoff13:47, recovery13:52, coordinator stop13:56.
Actual state was idle/healthy/fully hydrated before launch. The full470-second
interval and recovery fit. closeBy was narrowed once to13:32:00.037Z and never
extended. Chrome151.0.7922.108 PID11146 identity time13:24:15.147Z closed normally
at13:32:00.634Z, elapsed465.487 seconds. The deadline timer initiated closure;
recorded normal-close completion is597ms after closeBy. No second Chrome,
reload, retry, deadline extension or counter reset occurred.
All22 PNGs were personally viewed separately. Every capture saved the current
logical canvas RGBA and actual owner921600 readback before PNG. Owner availability
is false at startup0 and true at every later capture. Startup was gray at512/1024,
black at1536, BioWare logo at2048, black at2560, title/legal at3072. The original
balanced Escape followed the3072 review. Title/legal persisted at3120–3152;
legal text disappeared by3168. Controller artwork appeared at3184 without readable
New Game or a selection ring, so no Enter was sent there. Every title review
explicitly sets transitioning:false and openingAnimationVisible:false.
Changes between static title captures were not labelled visible animation.
| Current step | Personally viewed pixels and separate native facts |
|---|---|
| 3200 | Readable New Game and bright red controller ring; separate capture-bound review and native preflight precede balanced Enter. |
| 3248 | Select Difficulty dialog, orange Medium highlight; separate current review/native preflight precede balanced Enter. |
| 3296,3312 | Darkening menu after Medium; no further input. |
| 3328 | Black; menu/startscreen native state still retained. |
| 3344–3392 | Black; game/game, Level absent, level1/checkpoint9, type10 request6174784 size192320, completion0, completedBytes982692. |
| 3408 | Black; completion1, active cleared, completedBytes1175012, exact192320 delta; request/prepared/queue1 retained, Level still absent. |
The resource was pending longer than in the predecessor; no historical step was
used as a completion deadline. Current owner sidecars retain registers,48 stack
words, PC/previousPC, wait/yield/sleep fields and scheduler counters. At3344–3392,
PC is 0x4515cd, stack top 0x45774a, worker slices3277→3325, state active,
suspendCount/waitPolls/sleepUntil/lastYield all0. The retained original disassembly
in the predecessor's adjacent owner-disassembly.txt.gz identifies the zero-delay
wrapper return and archive output callback. At3408, stack top changes to
0x451f62, worker slices3341, ECX192320 and completion is directly observed.
The pending samples do not measure a monotonic decode cursor; stable stack and
aggregate bytes alone were not called a stall. No scheduler events were recorded
in these owner snapshots. owner-resource-correlation.json preserves the current
correlation, distinct from the historical disassembly.
The original finite timer closed Chrome after the last capture117. The subsequent quit command118 was written after closure and received no ACK; its timeout is retained as a handback helper failure. It was the only attempted command after the last shot; no progression or retry followed. All118 commands actually executed (0–117) have successful sequential ACKs and exact requested step counts. The menu ledger ends224 slices/42 commands, commandLimit113, all three one-shot input flags true. The continuation ledger ends112 slices/21 commands, seven loading quanta, three pending quanta, skipAttempted false. Those cumulative counters and absolute deadlines were never reset or extended.
The same worker repaired only the ignored local quit helper offline: next.py quit now checks the existing closure record and returns without submitting a
command when Chrome has already closed normally with the box left running. It
rejects abnormal closure; an open session retains the original single quit path.
Four focused tests execute the actual entrypoint with inert peers, reproduce the
old post-close submission and verify the repaired normal/open/abnormal branches.
All six routing tests and seven-mode author/operator integration still pass.
This repair was not used for a further runtime command. Executed helper bytes,
the repair patch and tests remain separate in RUN. It does not promise to eliminate
a close racing a later command; the next replay should hand back before the timer.
All488 closed remote files/37138066 bytes were recovered with full size/SHA checks; a second remote inventory is identical. Focused checks pass22 PNG/raw pairs, 200 served-file hashes, disabled cache/isolation, balanced Escape/Enter/Enter, drained input and the Lua completion. No game input follows Medium. Chrome and the owned driver are absent. Custody returned13:34 with the box still RUNNING (API idle, healthy, hydrated), expiry14:00:53.351; normal shutdown remains the coordinator's13:56 responsibility. No lifecycle operation, other box, subworker, guest call/write, forced callback, repaint, fault clear, emulator repair, source archive or media/module copy occurred.
Only this note is committed. Preparation Git SHA, relevant binary diff and new
small helper patches preserve provenance; no production code changed and no full
build is claimed. The next original progression requires a fresh authorized run,
the same independently viewed capture guards, and shorter review/transfer latency
within the unchanged470-second interval. The tested routes remain ordinary
16-slice pending/construction observations, then separately reviewed loading
observations only when current native Level l1 and queue qualify. Do not resume
this closed run or infer loading/tutorial from black pixels. This assignment ends
with one bounded original progression, Lua completion and the remaining Level/queue
boundary explicitly unobserved.