Hexen II valid API envelope and paired original traversal — 2026-11-12

Both current backends now have reviewed original courtyard, ramp, skull-door and chamber traversal, native input progress, and an attack pose while alive. WebGL's separate 1,000 ms attack ended at health 69; SoftwareGL's second-approach 600 ms attack ended at health 50 with the spider in the foreground. Incoming player damage is proved on both. SoftwareGL also reached native death and then restarted normally. Enemy damage, an enemy kill and a kill-counter change remain unproved. No emulator defect was isolated or changed.

Run: scratch/runs/20261112-hexen2-api-envelope-paired-progress-1040/. Parent 1025 RESULT/RECOVERY and published note commit 81e4c9b27, plus the 0953 report, were read and preserved. This replaces their current paired-gameplay limit without erasing either failed caller or its evidence.

Caller repair and qualification

The actual run caller now separates the 55-minute lease/observation deadline from the command API's integer execution timeout, which is capped at 600 seconds. control.py, envelope.py and the shared command builder submission.py are ignored run helpers. The original production request.cjs/file-ACK client is unchanged. Each detached API command is submitted once. Its acknowledged process ID, exact command and start time are retained; a status-read timeout reobserves that same GET route through the original deadline. It never publishes another browser command. An unknown POST response or changed identity closes input admission. No unknown request from either parent was replayed.

Before allocation, the full current caller path was exercised against a local HTTP endpoint: actual serialized POST, detached subprocess, persistent Node request.cjs, actual numbered command consumer, and exact ACK file. The result was delayed 21 seconds. A forced status-read timeout retained the same operation; one browser command, one consumption and ACK number 0 / ID delayed completed. The endpoint executed both legal timeout boundaries 1 and 600. Invalid values were rejected before submission. A separate known HTTP 400 pre-submission rejection produced no operation handle. An accepted command whose POST response was lost performed its unique marker effect once, then closed admission without replay. Complete wire bodies and operation events are retained. The local test uses the installed local Node path; the remote caller uses its pinned remote Node path, through the same command builder. The initial absent-local-binary attempt and earlier boundary-control revision are explicitly recorded rather than counted as the final qualification.

The parent-qualified browser, retained CDP helper, native dialog/bootstrap, strict identity binding, input, schema and cleanup helpers are byte-identical; native-bootstrap-preflight.json records their hashes and inherited qualification. The native MessageBox-before-GL/first-birth and replacement negatives were preserved, not rerun as a new fixture here. Current retained-screenshot, unknown-input, hard-deadline, stale/cleanup and 15-form regressions pass, as do five focused input/Worker tests. No production source changed, so no new build was required.

Reviewed native progress

WebGL dismissed the complete native notice, traversed Single Player → New Game → Paladin → Apprentice, naturally meshed models, and reached the courtyard. review/webgl/world-wait.png shows the statue, ramp and health 81. Subsequent turns, backoff and ramp corrections are native, reviewed movement, not parity measurements. ramp-climb.png shows the skull doorway. An early 1,800 ms Up+Control approach crossed into the chamber at 78. early-native-melee.png shows the extended punch after a 1,000 ms Control attack at 69, with spider legs right. The visible spider was then faced, followed by another 1,000 ms forward attack at 36. Native pause completed at 1 health. These frames prove living attack and incoming damage, not a hit on the enemy.

SoftwareGL independently followed the same class/difficulty route and reached the courtyard at 84. A passive follow-up confirmed the initially deferred turn; completed captures were not automatically treated as presented movement. ramp-top.png shows the skull door and door-approach-attack.png the chamber at 84 after the early Up+Control approach. The 1,000 ms attack's completed frame shows resting fists at 42, so it alone is not visual attack-animation proof. The following 1,500 ms attack ended in native death at 0, with the explicit “Player is a meal for the spiders” message. Space did not visibly restart it; a subsequent normal attack input initiated native loading and a courtyard restart at 83. No save/config/media patch or forced guest call was used.

A 600 ms courtyard Control pulse shows a distinct attack arm pose while alive. After native reapproach/backoff, return-backoff.png shows a complete spider directly ahead at 63. doorway-spider-melee.png then shows an extended punch, foreground spider legs and 50 health after a separate 600 ms attack. final-chamber-pause.png confirms the living native pause at 19. This is stronger current living-encounter evidence than the first attempt, without qualifying an enemy hit from the punch pose alone.

All 48 original commands have exact successful ACKs; 40 balanced pulses and 86 trusted native events were recovered. Seven command-level ACK waits crossed 20 seconds and retained their original numbered files. Original CDP reads settled within 20 seconds in this run; delayed CDP retention is established by the preserved parent fixture and the current retained-result regression, not by a new overdue original screenshot.

Identity, recovery and shutdown

Runtime source remains 8a834a3dc7affc8c6e53245beab56e87a32b81e8 and both recovered actual guest modules match SHA-256 04c534192ee3fe39e388536c2ce710c12c47a5cd2ca4d95c19f9d00b1e8f0a17. Original root glh2.exe remains SHA-256 76d96731e3076441e5c70bce2089cd8f36f25b373214f656aa9d5cd8d532e9dd. The joined provision process and 619 remote hashes/import checks passed before browser launch. SoftwareGL's owner is actual GL endpoint 1, software bridge and matching exported memory; legacy endpoint 2 is excluded. WebGL's owner is GL endpoint 1 with an actual WebGL1 context. Native SystemInfo reports ANGLE SwiftShader/Subzero; JavaScript Intel strings are not physical-GPU evidence. No FPS, raw-draw performance, physical-GPU, parity or audio claim is made. Licence scope stayed runtime/API observation and emulator source only: no guest reverse engineering, disassembly, extraction, patch or substitute DLL.

566 remote artifact references / 440 unique gzip blobs were recovered and checked for encoded/raw hashes and sizes before shutdown. check-evidence.py passes module, exact-envelope/ACK, balanced trusted input, cleanup and actual-join checks; visual enemy damage is not automated. The new dashboard entry has no invalid, missing or escaping references; the repository-wide checker still reports 23 older containment errors outside this run. Git SHA and relevant binary diffs/new-file patches are the source recovery mechanism; there is no source/media archive.

Own box bx_44e57mew was created 10:39:01Z with initial expiry 11:34:00.161Z. The independent deadline watchdog was observably armed before allocation. Both runtime parents joined 0 with both EOFs; Chrome exited 0, exact controller and Chrome PIDs were absent, and owned profiles were removed. Local manager and watchdog have actual parent waits 0 with both EOFs. Hash verification preceded normal stop stop_ccb1466db3c5cdfc, completed unforced without error. Final state: archived 11:01:14.814Z, expiry null. No task box remains running.

Next gameplay work is visible enemy damage/kill through native aiming and attack. Current paired traversal and living-attack objectives are qualified; no new causal emulator blocker, broader compatibility count or enemy-hit acceptance is claimed.