LTris keyboard original attempt, 2026-11-17
The original line clear and changed score/line counter remain unproved. One fresh LTris process reached one Modern/Normal, starting-level-0 board through ordinary keyboard menus. Three local P Pause intents expired before arming or DOWN, at capture ages 6.593, 5.033 and 6.374 seconds. No gameplay gesture was sent. This is an operator timing blocker, not a demonstrated runtime fault; the five-second gameplay review bound was not changed.
The board subsequently ended through the visible native Escape → “End Game?
y/n” → Y route, showing Game Over, Pieces Placed 4, Score 0 and Lines 0.
Escape returned to setup; reviewed Up selected Back, and Enter returned to
the main menu. The main menu initially needed a keyboard selection before
activation. Its separately reviewed Quit selection and Enter produced native
[Exit] code=0. All thirteen consumed keyboard gestures have independent
physical-release ACKs and actual release-child waits with both EOFs. Native,
watchdog, supervisor-child and outer waits all completed with exit 0 and both
EOFs. No host signal, cleanup command or watchdog firing ended the original.
Native Quit is proved; successful gameplay and original Pause are not.
Evidence: scratch/runs/20261117-ltris-keyboard-original-line-clear/result.json.
The initial already-reviewed capture name was refused locally before a menu
gesture because its review file existed. That setup error and the three
NOT_SENT Pause records are retained. Captures following menu consumption can
precede the final drawn transition; later full captures, rather than ACKs,
established that the board and setup had actually appeared. No second process
or board was started, and no terminal UNKNOWN endpoint was contacted.
Native keyboard qualification
The new test/test-ltris-sdl-keyboard.cjs authors a small PE using the existing
SDL fixture's PE/import header and a replacement keyboard-only program. It
calls the bundled authentic SDL_Init, SetVideoMode, PollEvent, FillRect, Flip,
Delay and Quit. Enter advances its black menu surface to a green board; P
changes it to red Pause and back to green. A matching actual SDL_KEYUP must
unlock every subsequent keydown. Left and Right change distinct public colors;
Escape exits through SDL_Quit and ExitProcess(0). These are authored fixture
surfaces, not original LTris gameplay.
test/test-ltris-sdl-keyboard.py drives that program through the released sealed
runtime and native guard, requiring the expected rendered colors, native-input
return receipts, independent releases and actual lifecycle waits. Intermediate
releases have physical-UP ACKs; the fixture's final Escape instead retains the
actual exited-owner wait, with physical UP explicitly unacknowledged. It is
not counted as a release ACK. The complete fixture route passed before the
original was started. Policy tests, actual-native wrong-owner/late-ACK/UNKNOWN/
parent-stall guards, and the existing real SDL consumer test also pass. The
canonical locked build passed before publication.
Exact inputs and boundary
Fixture and original used identical runtime provider inventories from Git
c2a980c5fd491cd66fd17d5c972cf88826c3da06, with foreign WIP excluded and recorded
as a binary patch. The new test bodies are also retained as patches. Both used
the existing pre-execution-retained module at
scratch/runs/20261117-ltris-current-original-visible-line-clear/tested.wasm,
SHA-256 533c599a144837890b6b45f6cbace2e52e9b4f48a3a8dd08b4261cd20223b1a1.
Its exact compiled Git/patch proof is reused by hash. Sealed module records and
the original's two real guest Worker receives/instantiations match that hash.
The explicit original batch size 1000 matches the fixture's default 1000;
real ticks, real guest Workers and a 1024×768 canvas were used throughout.
All 43 execution-media/manifest/CRT/ZIP records were verified unchanged before and after execution. Default controls were personally read from retained prior whole original Controls and Player1 screenshots on the same unchanged media: arrows, Up, PageDown, Down, End, Delete and P. Those references are explicit; they are not described as current-run Controls views. No Controls/Player1 pointer gesture, settings change, guest state read, solver, autoplay, forced pump, clock edit or deadline change was used. The historical 507.729 ms Player1 pointer failure remains unreproduced; no pointer repair is claimed.
The original's 1200-second deadline, independent prearmed 1220-second watchdog and irreversible 180-second Quit reserve remained fixed. All owned native groups are empty and their listeners retired. Selected whole PNGs, exact source identities, module/media references, actual waits and EOFs are named in the result; there is no source archive, frame archive or copied media.
No box or worker was created: box ID null, state NOT_CREATED, expiry null,
stop receipt NOT_APPLICABLE. Free /home remained above 1.5 GB. The task stops
with the single original allowance spent. Further original gameplay requires
a separate assignment after addressing operator capture-to-input timing
without relaxing the existing freshness predicate.