Original LBreakout score and brick progress with native CRT memcpy

Evidence: scratch/runs/20261110-lbreakout-native-copy-score-0241/ (ignored). The single fresh original runtime visibly advanced Michael's score from 0 to 200 and removed bricks from the lower-right of the original level. The final returned image, review/original/paused-return.png, shows the yellow Pause overlay, score 200 and openings where the initial board had light bricks. This establishes score/brick progress, not a level clear or sustained play.

Protocol deviation: after the repeated Right pulse returned an unchanged paddle position, I continued ordinary play rather than cancelling on that possible visual nonresponse. This does not satisfy the requested strict first-nonresponse rule. The score-200 image is retained as visual evidence, but the run is not reported as fully compliant acceptance. The box was stopped before the offline correction below; no further native run was attempted.

Exact published runtime and offline checks

The runtime was sealed from published main 9deab9f96e3a46e20ea1fdb8f0c2fdde84ba2ac2, containing the CRT repair e7e0312bd. Its actually captured owning Worker module is SHA-256 4f2c69a1528e01a9ab5127201d5e8a222da2a35236377a1351f444f55fa5a7d9. All 830 staged file pins remained unchanged. Exact Git SHA, relevant tracked binary diff, relevant untracked source patches and small run-specific helper patches are retained. Git remains the source archive. No source archive, evidence media copy, local Chrome or nested worker was created.

The live run used the unchanged observer. CRT, other shared providers, Q3 and Nox remained read-only throughout. Reused accepted unchanged fixture 4fc62b7d76540d0c937cae8cfb28d025ea1d6371; its retained completion artifacts were independently hash/size verified. There was no native fixture rerun.

The locked build passed with the repaired canonical hash. Eleven inspection tests, generated pause/resume and level-start observer ordering tests, and original compiled SDL input controls passed. The actual generated browser consumer ran offline against current module bytes with success, wrong-command binding, wrong-module binding, incomplete work, observer failure, presentation refusal and foreign immutable owner controls. Completion used zero full-module transfers within its presentation budget.

Additional run-specific controls execute separate P pause, separate P resume, and balanced Space delivery with individual returned-image gates. A wrong resume binding makes closure sticky, with no later protocol/input/capture/quit. These boundary adapters never launch local Chrome and make no visual gameplay claim. Their source is retained as small patches, including an initial offline adapter syntax error and its correction.

Reviewed native sequence

Each of 20 returned images was individually model-reviewed before dependent input. Ordinary menu keys reached Start Original Set, then Late At Night, level 1/25, Michael score 0, the original brick board, paddle and attached ball. Ordinary P returned a usable dimmed board with Pause. Separate ordinary P removed Pause and restored board colors. Ordinary Space returned a detached, airborne ball.

The first progress pause succeeded but still showed score 0 and an attached ball after a life loss. Ordinary further play stayed in this same runtime: separate resume, short Right inputs and a second Space launch. A provisional reading of that launch as score 20/brick removal was rejected by the next paused image, which still showed 0. The correction is explicitly retained; neither that provisional reading nor transient dark pixels support acceptance. One repeated Right pulse's image did not establish additional movement, and none is claimed.

After another separate resume, ordinary Right held for 300 ms visibly moved the attached ball/paddle from about x351 to x507. A third ordinary Space launch in the same runtime returned an airborne ball. The next ordinary P returned the accepted paused score-200 image with multiple lower-right bricks missing. The initial/resumed score-0 images and this final image are the sole basis for the score/brick conclusion; no guest-memory score or brick values are used.

The caller admitted direct Left/Right pulses with CDP down/up acknowledgement but no original paddle-consumption completion observer. That admission gap is demonstrated by the generated caller; the reason for the unchanged paddle image itself is not established. After stopping the box, the same worker repaired the admission gap offline: direct Left/Right and the legacy combined play command now refuse before any native protocol/input. Menu Up/Down, Enter, separate P and ordinary Space remain available. Paddle input requires a future original guest-completion observer. A twelfth inspection test covers both directions, observe flags and combined commands. The full newly generated consumer and real transport-consumption path also verify the refusal, owning terminal closure, zero paddle edges/images and zero later protocol; separate pause/resume/Space success and wrong-resume-binding controls remain covered.

The original two-second completion/presentation deadline was unchanged:

Command Guest complete ms Presentation complete ms
Original level start 1302 1328
Initial P pause 1429 1461
Separate P resume 101 157
First progress P pause, score 0 1378 1421
Second progress P pause, score 0 1302 1328
Final progress P pause, score 200 1408 1449

There was no transport refusal, timeout, browser death, unusable image or second native runtime. The visual-nonresponse deviation above remains explicit. No guest state/clock writes, artificial handler calls, instruction-budget changes or deadline changes occurred. Normal successful closure followed the final image review. Browser exit and actual launcher wait both returned 0, with separate stdout and stderr EOF records.

Evidence verification and final lifecycle

All 267 remote artifact references resolve to 226 unique local files, 7,724,314 bytes. Previously retrieved captures were reused by SHA/size instead of downloading or copying them again. An independent reread verified all unique hashes/sizes; all 20 PNGs passed chunk CRC, decompression and row-size checks. No gzip artifact was present. Verification and actual launcher wait/separate EOF preceded the sole delegated stop request.

Only new box bx_hq83bkwe, immutable UUID 0d93a23a-b1ba-436d-90c2-4916ad5a9578, was operated. The independent watchdog was observed before allocation. Its immutable deadline was 2026-11-10T03:07:30.100705Z; initial API expiry was 2026-11-10T03:07:29.468Z.

The sole normal unforced stop, stop_6c8acc8b88ede5b1, completed at 2026-11-10T02:52:05.619Z. Final state is archived, expiry null. The watchdog was actually waited and reaped with exit 0 and separate stdout and stderr EOF records. Other boxes were untouched. The allocation helper's messageboard line retained the old task label; the appended correction names this task and the same exact box/UUID. It does not indicate another allocation.

Next: retain this verified score/brick baseline. Level completion, sustained paddle play, audio and performance acceptance remain outside this result.