AVP2 staged executable: first original response-closure blocker

Run: scratch/runs/20261107-avp2-staged-entry-gameplay/ (ignored). The bounded task ended at a reproducible harness blocker, without Marine world, movement, attack, or an original screenshot. No emulator mismatch was established and no production source, guest settings, bindings or clocks changed.

Executable and dependency proof before allocation

The initialized-entry predecessor released its files at 16:38:17Z. Its repaired 619-file runtime manifest was reused without repinning functional files. Every file matched its recorded length and SHA-256 before allocation. The actual CommonJS closure resolves tools/return-fire-body-driver.cjs, tools/browser-response-body-collector.cjs, tools/browser-consumer-completion.cjs and test/static-server.js. staged-closure.test.cjs rejects the predecessor's incomplete manifest and accepts the repaired one. Remote assembly separately reconfirmed this closure and all runtime, helper and 18 original media hashes before launch.

The executable itself was run from /home/user/wine-assembly, the exact runtime root used remotely:

ALLOCATION_DEADLINE=<immutable-future-epoch-ms> \
LOCAL_QUALIFICATION=1 LOCAL_DIRECT=1 PINS_FILE=browser-input-pins3.json \
OUT_LABEL=local-executable \
node scratch/runs/20261107-avp2-staged-entry-gameplay/staged-browser.cjs

Those local fixture flags use the predecessor's Boids SDK control through the same executable and production attachPinned. They were removed for the original. No separate hand-written direct-API test substituted for executing the entry script. This proves Node startup and initialized guest/render composition; it does not qualify the registered AVP2 launcher.

local-executable/identity-admission.json passes complete source receipts, live window/input/process ownership, document/session/context/epoch/freshness, actual owning endpoint, native WebGL2 and completed render fences with no fallback. Actual initialized CPU and render module bytecode both match: e41511fe861363fbe1639d80f20cbcdd7240f46937c93a801eaa256ca094c467 (2,182,493 bytes). The personally reviewed local-executable/boot.png shows cyan SDK terrain and blue objects. It is not an AVP2 screenshot. The accepted admission regression rejects 41 mutated receipt controls; fresh Chromium input tests and keyboard/mouse stalled/rejected controls reconfirm independent UP and sticky UNKNOWN blocking queued input.

One original, no gameplay input

Before allocation, separate normal-stop and independent 25-minute watchdog OS processes were explicitly armed and verified alive. A fresh UUID idempotency key allocated only bx_56ahtghb at 16:42:01Z. Watchdog cap: 17:07:01.345Z; initial platform expiry: 17:17:01.172Z.

The original executable command was node followed by the same staged entry path above, with PINS_FILE=browser-input-pins3.json, OUT_LABEL=remote-webgl, WINE_TEST_BOX=bx_56ahtghb and the immutable allocation deadline. Local fixture flags and NODE_OPTIONS were removed. Actual document route: index.html?debug&candidates&threads&no-log&d3d-renderer=webgl&app=avp2_demo.

The production consumer observer records original AVP2.exe reader EOF: 57,344 bytes, SHA-256 36a772c2340ea7e415d8213d92b7a18d9541ef07c53b028d8a8ec475caf79b7d, bound to the actual initiating request, session, target, document and unique execution context. Actual instantiated bytecode was recovered and matches the module hash above. This is not complete original owning-Worker admission.

The first fault at approximately 16:42:55.972Z was an unclassified HTTP 404 for the spawned child icon, URL path icons/apps/vfs%3Ac%3A%5Clithtech.exe.png (decoded path icons/apps/vfs:c:\lithtech.exe.png). The existing narrow absence policy only allows the registered AVP2 launcher icon, not this dynamic child identity. first-fault.json, HTTP lifecycle logs and collector-final.json preserve it; the collector remains incomplete. No absence policy or guard was broadened.

The console records the original launcher's normal ShellExecute handoff to lithtech.exe -cmdfile avp2cmds.txt and launcher exit0. That command is guest behavior; no command file was staged or altered. Later, owning() failed with TargetCloseError during Worker evaluation before the first boot capture. The exited launcher's Worker is a plausible stale inspection target, not a proven emulator defect. The final state has child process1001 and no render endpoints. No original launcher/menu/world was visually reviewed, no ordinary input was sent, no force capture was used, and the original was not replayed.

Recovery, checks and closure

All 26 unique remote artifacts (3,791,622 bytes) were retrieved and verified by length and SHA-256 before requesting normal stop. The browser exited0. The remote audit proves both owned ports closed, no live owned process, no profile and no held input. Stop stop_25bc3139e2fb6503 completed at 16:44:00.110Z, forced=false; box state ARCHIVED. Its current archiveAfter is null, so archival retention is unknown. Both lifecycle delegates exited after verified normal stop. Local browsers, profiles, ports and watcher are closed; audit.json records free /home space above2GB.

Source base: cd98b939d97f6e115bad35e1d2b38876dfdb023d. Git is the source archive: source-diff.patch, new-files.patch (the relevant untracked lib/wininit-boot.js), owned helper patches, runtime hashes and media path/hash references preserve recovery inputs. No source archive or repeated media copy was retained. A local command-construction error before any remote assembly request is preserved in assemble-boot.log; corrected construction issued the single successful remote assembly/launch, without original replay.

Focused checks: staged-closure.test.cjs, entry-identity.test.cjs, input-controls.test.cjs, input-opposing.test.cjs, node tools/verify-avp2-recovery.js, node test/test-candidate-browser-manifest.js, and run-local audit.py pass. This is a notes-only commit; no production code changed or build claim is made.

Next: address the dynamic child-icon closure boundary without weakening source/module/media/input checks, and make Worker inspection tolerate the normal exited-launcher handoff. Then requalify the executable locally before any separately authorized fresh original attempt. Marine gameplay remains unproven.