Disciples II public mouse route: static boundary

Run: scratch/runs/20261118-disciples2-public-mouse-consumer-route-quest/. Actual original hover consumption remains unproved. No producer/readiness repair, New Quest, map action or ordinary Quit is qualified. No runtime was launched, and no old endpoint was contacted. The predecessor's external SIGTERM at 16:17:05.779 remains sender UNKNOWN and its scope permanently closed.

Newly established static bindings

The registered original EXE is adba242a0a3563651baffa6ec7d9a6c0978b7ae844a1cccedb2a1f4e8eb7b1f8. The EXE and all five registered DLLs were hashed and their imports parsed. None has a DirectInput import or matching ASCII DirectInput/device-query name. This does not exclude constructed names or GUID-based dynamic use. No acquired device, state query or buffered query is established; DirectInput is not assumed.

Original RTTI identifies vtable 0x6b4094 as CMqUIKernelSimple and 0x6b4164 as CMqUIControllerSimple. These are static bindings, not live object identities. Kernel slot +8 is loop 0x545cf3; +24 is HWND accessor 0x545f07; +38/+3c are sampled client/screen POINT getters 0x545f37/0x545f4e; +48 is async-button helper 0x545fb1.

The loop calls cursor helper 0x545e33 at 0x545d95. Its sole EXE GetCursorPos call is 0x545e3d, returning to 0x545e43. It copies the screen POINT, obtains ClientToScreen(hwnd,{0,0}) at 0x545e5f, and subtracts that origin. It does not test the cursor API BOOL before using the POINT. The sample precedes an individual unfiltered PM_REMOVE PeekMessageA at 0x545dca, or alternate GetMessageA at 0x545db8. DispatchMessageA is loaded at 0x545d68 and called through EBP at 0x545de0. The loop can continue without queue exhaustion. This is a concrete original-code difference from the earlier authored drain-then-sample fixture, not a readiness waiver.

After dispatch, controller +4c at 0x545e0c statically binds 0x5475f8, which walks render subscribers. Rendering alone establishes no hover use.

The window creation path at 0x54a9ba initially uses DefWindowProcA. The kernel constructor installs subclass 0x546150 with SetWindowLongA(GWL_WNDPROC) at 0x545bb3, then its object as GWL_USERDATA at 0x545bc6. The separately registered 0x54a16a callback in older notes must not simply be assumed to be this subclass. The subclass obtains the target HWND's GWL_USERDATA at 0x54616f and has distinct routes:

Public input Static forwarding route Simple-controller binding
WM_MOUSEMOVE 0x54639c -> 0x546595 -> +68 at 0x5465c4 0x5478cf -> subscriber call 0x54795c
Mouse buttons 0x546409 -> 0x546505 -> +64 at 0x546537 0x54782c -> subscriber call 0x5478bc

Subscriber lists can be empty. Reaching a forwarder is not successful application consumption. The async helper separately calls GetAsyncKeyState at 0x545fd8 and tests the physical high bit at 0x545fde; live use is unmeasured here.

Bink is a competing public cursor caller. Registered binkw32.dll imports GetCursorPos at original IAT 0x30036138, with calls at 0x300015a8, 0x300017f1 and 0x30002fc3. Their returns are 0x300015ae, 0x300017f7 and 0x30002fc9. Future observations must identify the actual loaded module relocation and callsite. An EXE-only caller list or generic cursor-read count cannot exclude this source. No Bink runtime call is claimed.

The retained audit passes 44 checks of actual PE bytes, whole-file hashes, query ordering, vtable bindings and competing calls. These are static checks, not native execution or consumer controls. Exploratory unaligned listing starts are excluded from instruction inference; exact byte assertions and aligned query listings support the findings above.

Remaining public-consumer boundary

The existing observer records the host get_mouse_position result before WAT GetCursorPos stores the public POINT and returns BOOL. That packed host result and one call-frame word do not establish a successfully completed public query, or the controller subscribers' use of its data. No actual target HWND, process, thread, lifetime, device, returned-query data or visible hover consequence is established in this task.

The next passive instrumentation needs a completed public query and its output, or exact public MOVE dispatch and caller continuation, bound to actual process, memory, thread, HWND generation and lifetime. It must separate Bink, nested, sibling, nonconsuming, failed and stale activity, then correlate with a personally reviewed whole-frame hover consequence. Existing nonempty refusals remain. No generic callback/Peek/frame completion supplies that proof. No production defect has been demonstrated to repair.

Admission, evidence and release

Initial /home free space was 1,497,309,184 bytes, below the required 1.5 GB; a later sample was 1,492,959,232. Native controls and a diagnostic original were not attempted. The newly authorized diagnostic has not been consumed. Headroom recovered to 1,548,255,232 during sealing, permitting compact durable retrieval and this documentation. No continuous whole-task floor is claimed.

Git, relevant clean-path diff, unique audit inputs, original binary hashes, disassemblies, recovery instructions and validation are retained. No source tree, media, tested module or frame series was copied. There were no source changes, native controls, build, runtime inputs, private runtime reads/writes, clock changes, workers, Python, Wine reference or deployment. All runtime, browser, display, watchdog and actual-parent wait fields are null because no runtime was created. Box ID null; state NOT_CREATED; expiry null; stop receipt null. Evidence retrieval and final publication receipts are in the run result.