Original LBreakout visible pause, resume and launch
Evidence: scratch/runs/20261110-lbreakout-visible-pause-score-0208/ (ignored).
Code shipped on main as d33b2fa343767e6873417c560d1d2de75dce0c88 before
the sole original run. Reused targeted initialization 5cbb2a8f4, zero repeated
module transfers 05af3c36e, and unchanged accepted fixture 4fc62b7d7.
No native fixture rerun, second original attempt, or inherited-box operation.
Repair and offline controls
Pause still used range tracing, which globally disables fast execution even outside the selected addresses. Its keyboard entry/return, handler entry, display/state return and caller now use the existing compiled breakpoint driver. The handler entry moves directly to its expected display/state target; there is no broad handler range. The same driver already handles initialization. The unarmed path retains native run and metric function identities. No provider, module, guest instruction, clock, handler invocation or instruction budget changed.
A separate ordinary P command can request resume state 8 only after identifying the reviewed pause image. This permits a returned resume image to be individually reviewed before normal Space launch. The preexisting combined play command was not used in the live run.
Actual generated-consumer success, wrong command binding, wrong module binding, incomplete work, observer failure, presentation refusal and foreign immutable owner controls passed. Completion incurred zero full-module transfers; full byte validation remains before arming and around capture, with immutable owner and DLL binding checked during polling. Eleven inspection tests, pause/level-start observer tests, original SDL input controls and the locked build passed.
The actual compiled-module test now runs original full-screen surface_gray
against isolated original SDL surfaces in alternating disabled/range/target
modes. Every output pixel matches, and natural entry/end targets are observed.
Three samples in milliseconds were disabled 541/508/540, range 845/636/640,
target 672/607/524. Only range mode enables the global tier guard. These isolated
samples establish the instrumentation mechanism, not a native two-second promise.
Original background work, complete 25-level parsing, moving targets, conflicts,
restoration, and full generated preload validation also passed. Both pause and
resume preload arms preserve tier guard zero against the actual compiled instance.
Sole native result and first terminal boundary
Each of ten returned images was individually model-reviewed before dependent input. The original menu led to Late At Night, level 1/25, Michael score 0, full brick board, paddle, attached ball and lives. Initial guest presentation completed at +1383 ms. Ordinary P completed at +1760 ms and returned a usable image showing the dimmed original board and yellow Pause text.
The separate ordinary P resume completed at +113 ms and returned the colored board with no Pause overlay. After that image was reviewed, ordinary Space (300 ms held, independently released) returned an image with the detached ball airborne above and left of the paddle. Its displayed score was still 0.
The next ordinary P attempted to retain current progress after natural gameplay
during model review. Original handler entry occurred at +26 ms, display/handler
return at +1984 ms, and SDL P release/guest completion at +1986 ms. The unchanged
two-second deadline left 14 ms for presentation; the browser returned
presentation deadline, withholding the image. The caller immediately closed
and the worker issued owning cancellation. No later guest protocol, input,
capture, retry, quit or extension occurred. Score and brick progress are
unconfirmed. This is visible pause/resume/launch evidence, not score or clear
acceptance. No remaining live authorization is claimed.
Evidence and lifecycle
Runtime used exact code commit above and module SHA-256
3e491c0840bb7771328738a0d160b4bbaa3d7ab72074cbd38a7b41e5d7e8ec18.
All 826 staged runtime pins remained unchanged. Exact Git SHA and relevant
binary diffs, including small owned helper inputs, are retained; no source
archive or game-media copy was made in evidence. Shared providers, Q3 and Nox stayed
read-only; no nested workers or local Chrome were launched.
Only new box bx_sradnt45, immutable UUID
0d11c083-7500-49b8-ae41-f9c9bb55e1f2, was operated. Its independent watchdog
was observed before allocation, with immutable deadline
2026-11-10T02:38:30.795687Z; initial API expiry was
2026-11-10T02:38:30.174Z. All 152 artifact references (131 unique) were
retrieved and independently reread for SHA/size; ten PNGs passed CRC and
decompression, with no gzip artifact present. Actual launcher wait (-15) and
separate stdout/stderr EOF were verified before requesting stop.
Exactly one delegated normal unforced stop, stop_e90213fd85a7f77f, completed
at 2026-11-10T02:17:53.342Z. Final state is archived with expiry null. The
watchdog was actually waited and reaped with exit 0 and separate EOF records.
The preexisting bx_chfab9w3 was untouched.
Next: offline investigation of the original pause drawing cost and remaining presentation margin, using retained evidence and compiled controls. Another native attempt requires separate authorization; do not extend this deadline or infer score from a withheld image.