Original LBreakout bounded paddle completion

Evidence: scratch/runs/20261110-lbreakout-paddle-consumption-0300/ (ignored).

The generated caller retains the 080064edf default refusal of unobserved Left/Right and all combined play commands. The explicit completion:paddle path requires one direction, observation, a command ID, the prior reviewed image and an 80–300 ms ordinary balanced native pulse. Completion binds the immutable owning instance, command token, image SHA/label, original EXE and SDL mapping and lifecycle generation. It observes original SDL key delivery, original paddle entry, its original SDL key-state poll/return, paddle return with displacement in the requested direction, and release before page presentation and capture. The existing two-second deadline and instruction budgets are unchanged.

All boundaries use the existing single-target breakpoint driver. A validation-only mode checks original media and loaded instruction anchors without enabling the older broad range diagnostic. No shared CRT, renderer, startup, Q3 or Nox code is changed. No guest writes or artificial calls occur in the native observer. Offline compiled controls use the existing isolated test objects and original SDL/paddle routines; those test objects are never used in native gameplay.

The prior accepted fixture 4fc62b7d7 is reused after independent hash/size verification; no native fixture rerun is authorized or performed. The single authorized native run ended at the first visual nonresponse below.

Offline validation and exact runtime

Published observer commit: b1dc83b64febca461b806a77e1f66dbeff2f51c3. The locked build passed; its canonical module remained 4f2c69a1528e01a9ab5127201d5e8a222da2a35236377a1351f444f55fa5a7d9, matching the module actually captured from the owning native Worker. All 832 staged file pins remained unchanged. Exact Git SHA, tracked binary diff, relevant untracked source patches and run-specific helper patches are retained; Git is the source archive.

Twelve inspection/handoff tests, generated paddle completion controls for both directions, pause and level-start controls, and original compiled SDL controls passed. The new targeted observer also passed against the original compiled SDL and paddle handler using isolated offline objects. Nine full generated caller controls exercised the actual numbered transport consumer: success, wrong command, wrong module, wrong image, incomplete work, observer failure, presentation refusal, unobserved paddle refusal and combined-play refusal. The success control returns individually gated Right, Left, P pause, P resume and Space images through a boundary adapter; it makes no native visual claim. Negative controls require no first image and sticky rejection of all later key, capture and quit operations before protocol. Initial offline adapter failures and corrections are retained separately.

Native blocker and strict terminal action

Each of three returned images was individually model-reviewed. Startup showed Loading. A separate capture after ordinary loading showed the original green main menu with Local Game selected. One independently released ordinary Enter 80 ms pulse returned the same main menu. The original game rectangle (31,186)-(672,668) is pixel-identical in those last two returned PNGs, confirmed by a full offline PNG decode and SHA-256 comparison. The two full PNG hashes differ because pixels outside that game rectangle differ.

The first unchanged expected menu transition was treated as visual nonresponse. Immediate owning SIGTERM cancelled Chrome PID 7165 and Node PID 7156 at 03:07:58Z. There was zero later browser protocol, input, capture, retry, quit or extension and no second native run. No native paddle command or observer arm was reached; native Left/Right movement, separate pause/resume/Space and score progress are not established by this run. The prior score-200 run remains explicitly noncompliant and is not substituted for this result.

Only native CDP down/up acknowledgements are available for this Enter. They do not establish original SDL/menu consumption. The returned-image nonresponse is proven; whether the guest missed the pulse or presentation preceded processing is not established. No speculative guest/provider repair was made. Next work is an offline original menu-consumption/presentation completion gate, before a separately authorized native attempt. This task's sole native authorization is exhausted.

Final evidence and lifecycle

All 55 remote artifact references resolve to 50 unique local artifacts, 3049104 bytes. Captures already retrieved for review were reused by SHA/size; none was downloaded or copied twice. Independent rereading verified every unique hash and size. All three PNGs passed chunk CRC, decompression and row-size checks; no gzip artifact was present. Actual launcher wait returned -15 with separate stdout and stderr EOF records before the sole delegated unforced stop.

Only new box bx_99yye6ek, UUID ae5b1ec5-1d5c-47f2-a54d-b3c0eddaa8bc, was operated. The independent watchdog was observed before allocation. Its immutable deadline was 2026-11-10T03:31:33.682336Z; the initial API expiry was 2026-11-10T03:31:33.250Z. After finalized evidence verification, sole stop stop_0fe7d955ecbd44bb completed at 2026-11-10T03:08:55.756Z, unforced. Final state is archived, expiry null. The watchdog was actually waited and reaped with exit 0 and separate stdout/stderr EOF records. Other boxes were untouched. No archives, repeated media copies, local Chrome, nested workers, guest-state/clock writes, artificial native calls or deadline changes occurred.