Hexen II page-task attribution attempt, 2026-11-20

No function attribution or source fix was established. The fresh original scope closed on the actual /home floor before the bounded profiler started or any native game key was prepared. Class, difficulty and level remain unrun. Evidence: scratch/runs/20261120-hexen2-attributed-page-task-source-repair/.

The historical cancellation still establishes only timing: final whole capture plus retention took 170 ms, socket to cancellation took 157 ms, and an overlapping page task lasted 234 ms. The task's long-animation-frame entry has an empty scripts array; no retained named function or CPU stack identifies it. All fourteen structural predicates passed when freshness failed at 351 ms, with zero consumed tokens. This cannot establish a cached-sample defect or identify a host function.

The new graph used the root-frozen pre-W3 producer-scheduling bridge, SHA-256 91069c0446876b6528c72db09517e45a1fe3a7122a1444bc3dc5408f8ed22c2b. It was not the concurrently edited working-tree bridge. original/served-graph.json pins every served source; frozen-bridge.patch and frozen-bridge-recovery.json prove recovery of the exact bridge from the recorded Git commit. Both actual CPU/render Workers loaded d3c95bc28f49bd5e1bd05e6c7536a240ebdfffd803a98c6f3d0372afec3f36d2; all eighteen media inputs were verified by existing path/hash without copying them.

The separately connected profiling recipe was prepared before launch. It permits only target attachment, CPU-profiler setup/start/stop/disable, and tracing start/end; trace events write directly to tmpfs, with an 8-second duration and 48 MiB size bound. Its cleanup has no application, input or frame queries. CPU profiling was configured for a 1 ms sample interval, but neither profiling nor tracing started: measured sampling overhead and a named stack are therefore unavailable. Setup/cleanup wire timings remain recorded. Profiling-only cleanup after UNKNOWN sent one Profiler.disable and took 1105 ms; it is not a runtime measurement of the 234 ms task. The broad existing method/producer/retention probes remain instrumentation, not an independently measured causal effect.

The entire 800x600 pre-focus host captures were personally viewed. Trusted host-button DOWN/independent UP focused the original; native binding completed at 16:05:30.343 UTC. At 16:05:34.131 UTC, Lifecycle.checkpoint('manual-command-wait') observed 1,499,791,360 free bytes and permanently closed the scope. No profiler start, trace prearm, strict prepare, game DOWN/UP or native USER consumption followed. The next local capture command was refused before writing a command or reaching an endpoint. There was no replay, state write, synthetic timestamp, forced presentation, relaxed status or widened 250/500 ms guard.

All twenty supervisor-tracked PIDs were nonlive; browser, private display, focus helper, release helper and messageboard watcher have actual joins/EOF. The finite supervisor and separate cleanup deadline remain recorded. Owned profiles, sockets, lock/auth files and the empty task tmpfs child were removed; root Node cache and all proofs were preserved. The disk floor was breached and later recovered; continuous headroom is not claimed. Box ID is null, state NOT_CREATED, expiry and stop receipt null.

The prior 51 Quake and 42 Hex JavaScript controls are reused against matching source hashes; no new native calibration or causal old-fail/new-pass success is claimed. Production code and inherited WIP remain unchanged, so another full build is unnecessary for this documentation-only change.

The handoff blocker is precise: the fresh original closed before a profile could associate the capture/socket interval with a named function and a genuine producer publication. The historical task remains FUNCTION_UNKNOWN. The retained, unexecuted profiler recipe can support a separately authorized fresh scope once adequate disk headroom is maintained; the closed scope must never be resumed. Only after actual stack attribution should a source owner claim the narrow function, repair it, prove old-fail/new-pass with the unchanged guards and controls, and exercise the affected original.