LBreakout2 actual SDL consumer controls; causal blocker, 2026-11-09

The first concrete blocker is missing observation of the original SDL/game consumer during the retained Right pulse. Trusted browser DOWN/UP and unchanged paddle geometry establish nonresponse, but do not distinguish state sampling after release from the game's own input-inhibit branch or another interruption before paddle handling. No historical runtime cause is demonstrated, so there is no justified production repair or changed live-fixture prerequisite. No box was allocated, no pulse replayed, and no new gameplay progress is claimed.

Evidence is in scratch/runs/20261109-lbreakout-sdl-input-cause-2350/. Starting main SHA: ec04b2b9eec031348a90b56851b95f49ecb31432. Exact final source SHA and small relevant patches are retained there; Git remains the source archive. No source tree or game-media archive was made.

Actual original consumers

node tools/lbreakout-sdl-input-controls.cjs loads the canonical original EXE and SDL.dll into the production interpreter, patches their actual imports, and executes their original x86 routines. It implements no paddle movement. Both media hashes are asserted before using native addresses. This is an offline test with isolated paddle/surface objects and explicit test config; it does not initialize a complete game or claim native browser evidence.

The controls execute SDL WinDIB key translation/message consumption (RVAs 0x10c20, 0x10a10), DX5 buffered keyboard decoding (0x138a0, 0x13070), and SDL_GetKeyState (0x19810). They then call the original EXE paddle handler at 0x405904, which polls SDL keyboard state and relative mouse state. Left, Right and Space bindings use the original key symbols 276, 275 and 32.

Both original SDL keyboard paths produce these actual-game-handler results:

Control Observed result
Right DOWN then UP before handler x remains 320
Handler while Right is DOWN x becomes 328
Handler after Right release x remains 328
Handler while Left is DOWN x returns to 320
Space DOWN / UP right-fire request 1 / 0
Paddle field +0x3c set, Right DOWN x remains 320

The last branch is an observed input inhibit; its historical value was not retained. The Space check demonstrates a native launch/fire request, not actual ball detachment. Mouse polling is present in the native handler; these controls use zero relative deltas and do not establish historical mouse input. WinDIB's native repeat-bit rejection also passes. An initial diagnostic supplied four arguments to its five-argument consumer and failed; the retained corrected control supplies _this, hwnd, message, wParam, lParam and passes.

Retained route comparison

The accepted November 8 route used ordinary keyboard control and separate Space launch pulses (including 250 ms), then Right 176 ms, and eventually showed score 2000. It supplies no proof that increasing the failed pulse duration would repair its cause. The previous generated native fixture implemented its own keyboard movement, so it did not qualify the original SDL consumer.

In the November 9 original, trusted Right DOWN was recorded at 23:45:02.552Z and UP at 23:45:02.626Z; independent release was armed before DOWN. The 23:45:02.630Z state still contains Right UP in the renderer queue and one buffered DirectInput event. DOWN is absent from that page queue, which does not prove arrival at SDL or a game state read while held. Guest focus is 0x10001 both before and after, so a missing guest focus is not established. The final capture began at 23:45:02.787Z; no later queue snapshot was retained.

The retained ready and final images were separately inspected again: score 0, attached ball, unchanged paddle position, intact brick field. No new capture or dependent input followed. Neither SDL state during DOWN, a DX5 decoder receipt, the paddle handler invocation nor its +0x3c field is recorded. These are the specific missing observations that prevent choosing a causal patch.

Both retained execution snapshots end at 0x603783. In the offline load SDL starts at 0x5f0000, making this SDL RVA 0x13783, following its DX5 input wait. That mapping is an inference for the historical process, not a verified historical load address or proof of a permanent loop. The corresponding zero-timeout message-wait yield code also exists in the accepted source; this comparison establishes no new regression.

Validation and preserved identity

The original-consumer controls pass after the locked build: 40 native calls, bounded returns, both keyboard routes and the original paddle handler. test/test-directinput-buffered-keyboard.js --rpc passes 14 checks, including rapid DOWN/UP ordering through production renderer, GuestRpc and WAT consumer. test/test-keyboard-focus-seed.js passes; native-key controls pass all five tests. tools/verify-lbreakout2-251.js verifies 220 original files, authentic CRT and 219 companion mappings. flock /tmp/wine-build.lock bash tools/build.sh passes. Logs and independently SHA/size-verified selected historical receipts are retained in the run directory.

The module remains SHA-256 3e491c0840bb7771328738a0d160b4bbaa3d7ab72074cbd38a7b41e5d7e8ec18 (2186068 bytes), matching the failed original's module. SDL and EXE media hashes are recorded in the controls output. The registered lbreakout2_251, original DLLs/CRT, empty arguments, C:\LBREAKOUT, and repaired Canvas2D/noD3D identity chain remain untouched. No production SDL/input patch was made.

No live box, watchdog or stop receipt exists for this task: box ID/state/expiry and receipt are not applicable. The prior archived box was not resumed, modified, or deleted. /home remained above 1.5 GB free. The next prerequisite is observation of the bound original SDL/game consumer during an input interval, sufficient to distinguish these concrete controls before choosing a repair or another live attempt. Historical score 2000 remains the high-water mark; this task stops at the precise causal blocker.