Hexen II bounded native input observer: first implementation blocker

Evidence: scratch/runs/20261120-hexen2-bounded-native-poll-causal-observer/. This is a source audit, not a qualified observer or an affected-original test.

The prior sealed analysis remains authoritative: final capture and retention used 381 ms, leaving 55 ms after physical DOWN. Token zero cannot distinguish busy execution, a host/render wait, publication delay, or PeekMessage cache and filter behavior. The original observed PM_NOREMOVE; the calibration observed removing GetMessage. Neither is substituted for the other.

The existing hooks were inspected before proposing another observer:

An OVERLAP_REQUEST was appended to the board for the actual consumer lifecycle hook in lib/guest-worker.js. MCM released that file unchanged at 11:47:48 UTC; this resolves its ownership conflict. Host/render producers remain held and untouched. The consumer hook must bind the observer to the real process/thread/endpoint epoch, establish prearm before the exact edge, retain execution/wait/error transitions without a new busy loop, and terminate its own observation on cancellation, retirement, or error. Native USER cache and filter observations must also be retained at their true producer boundaries; the present latest-value export is insufficient for that history.

The remaining blocker is the required exact module identity. The retained ordered-candidate.wasm was compiled for export inspection only, never instantiated: its SHA-256 is 10bdae2c56c79eeb5d0b4ebbc602c8a233e36351380da274a70f7f5e25363564, and its length is 2,418,815 bytes. It exports the latest-value USER getter but no shared native USER cache-transition feed. JS hooks cannot observe a cached Peek path while Wasm runs without entering an import. Extra post-UNKNOWN commands cannot repair that gap. Adding a native publication producer changes this exact module hash; a complete observer therefore needs a revised module pin for qualification and for the affected original. The task requires the exact old pin, so this audit does not substitute the current build or silently patch the module. This is an instrumentation boundary, not a diagnosed cause of the original failure.

No disconnected observer module was added as a substitute for provenance. No actual Worker qualification, PM_REMOVE/PM_NOREMOVE equivalence control, busy/blocked/delayed capture control, affected original, New Game, movement, enemy consequence, or Quit is claimed. All remain unproved. The next worker has the exact consumer-file release, but needs a revised native module pin before implementing and qualifying the complete observation path. It must preserve the existing deadlines and first-UNKNOWN closure. There were zero runtime endpoint requests, no replay, and no test box: ID null, state NOT_CREATED, expiry null, stop receipt null.