Hexen II page function attribution original, 2026-11-20

The 234 ms historical page task remains FUNCTION_UNKNOWN. No causal source repair or class/difficulty/level progress is established. Evidence is in scratch/runs/20261120-hexen2-page-function-attribution-original/.

The single fresh original used the frozen pre-W3 bridge 91069c0446876b6528c72db09517e45a1fe3a7122a1444bc3dc5408f8ed22c2b. Both actual CPU/render Workers were verified as d3c95bc28f49bd5e1bd05e6c7536a240ebdfffd803a98c6f3d0372afec3f36d2, with the exact eighteen media inputs verified at their existing paths. Shared bridge, host, renderer and other lanes were read-only.

Before launching the browser, the local manifest was corrected to use the actual three existing focused test files. A failed local manifest attempt did not launch anything. All sixteen recipe/test paths then passed syntax and identity checks. Optional trace flags propagated before launch. The local profiler lifecycle was also corrected to start only for the explicitly traced intent and stop in that intent's finally block before another capture. These are run-local recipe changes, not a production repair or validated sampling result.

The prepared independent profiler permits profiling methods only, uses a 1 ms CPU sampling interval and bounds the profiled interval to eight seconds and 48 MiB. Raw trace destinations are in the assigned tmpfs. Its start/stop sampling path was not reached. Sampling overhead is unknown; method wrappers and retained native snapshots also impose instrumentation overhead. The 2614 ms cleanup flush was an unarmed Profiler.disable, not evidence about the historical page task.

Both selected whole 800x600 images were personally viewed: the pre-focus host showed the Hexen II menu and fullscreen consent button; the focused image showed Single Player selected. Trusted host-button DOWN with independent UP succeeded. Native binding completed at 16:29:28.692 UTC. At 16:29:45.037 UTC, Lifecycle.checkpoint('manual-command-wait') observed 1,498,906,624 actual free bytes on /home and permanently closed the scope. The intended Enter command was subsequently refused locally before a command file or endpoint request. No profiler start, trace prearm, strict preparation, game DOWN/UP or native USER consumption occurred. No application/input/frame requests followed UNKNOWN, and there was no replay.

The historical evidence still contains an empty scripts array for the 234.525 ms long-animation-frame entry. The 170 ms capture/retention interval and 157 ms socket/cancel interval overlap it, but do not identify a function. No newer producer before cancellation, cached-sample defect or adapter-only cause is proved. Default 250/500 ms guards and frozen/replaced/retired/epoch policies remain unchanged; no timestamps, state, presentation or status authority were altered.

The handoff remains the missing named stack: a separately authorized fresh original must correlate a bounded page CPU profile and trace with final capture, genuine producer publication, dispatch, status, native USER and independent UP. The closed scope cannot supply that evidence and must not be resumed. This task therefore does not claim to have identified or fixed the function.

The existing 51 Quake and 42 Hex control results were reused against matching source and log hashes, with historical native proofs referenced rather than recalibrated. Production code was unchanged, so no redundant full build was run. Cleanup evidence records all supervisor-tracked PIDs nonlive, actual browser/display/helper joins and EOF, finite runtime and cleanup deadlines, and removal of owned profiles, display socket/lock/auth and empty tmpfs task child. Proofs and root Node cache were preserved. The floor breach is retained; continuous headroom is not claimed. Box ID, expiry and stop receipt are null, state NOT_CREATED.