Airstrike: dispatched Space → SDL → startup wait
Historical source last updated, 2026-10-27. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.
Result: the next boundary is DirectInput keyboard collection, not another Win32 message dequeue. Actual SDL consumption remains unresolved. No source defect or gameplay acceptance is established. This task executed no guest, build, Node, Boat or scheduler operation.
The accepted predecessor is commit 4da44d928030034dc6ff91357db7d47d6724e489, with canonical receipt receipt-manifest.json and coordinator verification air-postdeadline-verified.json. Its four reports and 1802 verified payloads were reused without archive rehash or test replay. Ordinary decision21 at5005 admitted the main thread, dequeued/routed Space down/up to HWND 0x10001, and recorded both DispatchMessageA calls. Frozen5006 has zero credits and an empty renderer FIFO; the inspected frame still says “Press any key to start”.
Exact binary path
Addresses below are original SDL preferred VAs. The retained runner loaded SDL at 0x536000; runtime VA = original VA − 0x10000000 + 0x536000. The EXE remains at its preferred 0x400000 base. Full hashes, byte excerpts and source line ranges are in static-evidence.json (historical artifact: airstrike-dispatched-space-application-cause-oct27/static-evidence.json).
- SDL registers its common Win32 procedure
0x10022480(runtime0x558480): the WNDCLASS construction stores that pointer before RegisterClassA at0x10022f03. The external-window setup also installs that same procedure with SetWindowLongA at0x1001442e. The common procedure delegates unhandled messages through pointer0x10039f0c. DirectX setup assigns this pointer0x100139a0and the video pump callback0x10013eb0at0x10014a2d..0x10014a55. - The saved final main EIP
0x549d7fis exactly0x10013d7fin the DirectX pump0x10013c70, immediately after WaitForMultipleObjects. This identifies the pump executing at the snapshot. It does not record the actual HWND procedure lookup, callback entry/return, backend-pointer value or earlier path through that pump. - The DirectX backend handler's byte-index/jump table maps WM_KEYDOWN, WM_KEYUP, WM_SYSKEYDOWN and WM_SYSKEYUP (
0x100/101/104/105) to0x10013a51: pop registers, zero EAX, return. These messages do not call SDL's keyboard-event producer on this backend. All four table entries and the exact return bytes were checked directly in the pinned DLL. Dispatching Space therefore cannot by itself prove an SDL_KEYDOWN was created. - The DirectX pump first drains Win32 messages (
0x10013c85..0x10013cbf), waits, then reads device data through vtable offset0x28at0x10013dae. A successful read, nonzero event count and enabled processing lead to the device callback at0x10013e24. Keyboard descriptor initialization selects0x10013420; that routine walks 16-byte device records, translates scan codes at0x10014320, and calls the private keyboard producer0x1001a230. The producer updates key state, checks event enable/filter gates and calls SDL_PushEvent (0x10018970) at0x1001a436. - SDL_PollEvent (
0x10018900) pumps events and then calls SDL_PeepEvents (0x10018670) for one event, action2, mask0xffffffff. Its internal queue head/tail are0x10035760/764, records start at0x10035768with 20-byte stride. The caller needs a positive poll result and event type2, not merely an empty host FIFO.
Application exit from the startup screen
The release source's main draws the startup message in message_mode, which draws one frame and calls wait_for_key. That loop returns on any SDL_KEYDOWN, ignores SDL_KEYUP, exits on SDL_QUIT and delays when no event is available. No Space-specific startup filter appears.
The actual EXE independently corroborates this: 0x413a39 loads the startup text at 0x42d3b8, calls the message routine through 0x4110d2 → 0x412bf0, and that routine calls 0x412750 at 0x412ca0. The wait loop calls 0x41137f → 0x4292da → SDL_PollEvent at 0x41277b, tests result≥1, compares event type with2 at 0x412791, and returns keysym from [ebp-0x10] at 0x4127b6. Only afterward does main set up objects/players and print “Entering main loop.” See the aligned wait-loop disassembly (historical artifact: airstrike-dispatched-space-application-cause-oct27/wait-loop.objdump.txt). The source archive is corroboration, not proof of a reproducible source-to-EXE build.
What is missing
The accepted runner's dispatch counter is recorded in the API trace hook while reading MSG arguments (test/run.js:3381..3398), before proof of guest callback completion. The compiled revision's DirectInput keyboard collector compares live key state against a previous-state bitmap; it does not consume a keyboard edge FIFO. The runner supplies that state through renderer.peekAsyncKeyState. A down/up pair between samples can therefore leave no new pressed-state difference. Both Space messages were pending before the admitted main run, and the final key state is up: this makes collapsed sampling a concrete candidate, not a demonstrated explanation of this run. No retained GetDeviceData arguments/results, previous bitmap, producer result, SDL queue state or application poll result closes that inference.
The smallest next observable gate, requiring a separately authorized task, is one correlated keyboard-device GetDeviceData completion: device/acquisition identity, requested/returned count and HRESULT, DIK 0x39 records with data 0x80/0, live VK_SPACE and previous bit at collection. Follow any produced record through 0x10013420 → 0x1001a230 and queue insertion, then observe the poll return/event type at EXE 0x412780/0x412791 and exit 0x4127b6. Capture the actual HWND procedure/backend pointer if callback routing is investigated. This separates sampling loss, unavailable device data, SDL filtering/queue failure and an unobserved application return. No new runtime is authorized by this report; the consumed allowance remains CPU79.815637/wall100.206209416 of90/120, with the single invocation consumed.