Stronghold first-script native attempt, 2026-11-04
The sole fresh attempt on bx_8kfdtfgw proved that published a2b2717a1
executes its preload before the actual Chrome Worker's first script. It then
stopped at a demonstrated URL binding prerequisite. No title, menu, preset
activation, settings application or gameplay was reached. The earlier closed
attempt's TargetClose cause remains unknown.
Evidence is in scratch/runs/20261104-stronghold-first-script-native/.
The coordinator allocation identifies this box as the fork of stopped,
snapshot-verified bx_9eqxpvx4; the predecessor was never accessed or resumed.
The fresh box was running, healthy and hydrated before launch. All 2,415
inherited pins matched. Runtime source candidates matched the inherited pins,
so no compiler run, module copy or source/media bundle was necessary.
Only the small published successor and fresh run inputs were staged.
The source checkout pin was
bc7555e5629c30c40779d1e1c33db36c9f92e55a. The exact original normal module
was 4417305d70322467441aef51e888b0fe66451d9c60da0c734eeb23f5a22c0a6b;
compatibility remained
07fb4e75d05177271097dcd67d84e2629272b50073074bbdea8825ce10b4ed34.
Media stayed at the inherited
test/binaries/candidates/stronghold-crusader-demo/installed/App_Executables/Stronghold Crusader multiplayer demo.exe,
SHA-256 9024c533095f2227b0105b7e8a9e4cf28ec4fb206b165629787b2c97cf99f13f.
The actual composed closure, 30 retained helpers, Puppeteer CJS/ESM, executable
Chrome 151 and 164 HTTP body pins passed before launch. Successful static
responses used isolation headers and no-store; the browser disabled its cache.
The live log also retains five nonfatal 404 responses (build-info, auth,
favicon, font bundle and app icon), with isolation headers and no cache header.
Chrome launched once at 15:55:32.605Z. Actual
browser/worker-bootstrap.jsonl records instrumentation installed before startup
resume, then successful preload at 15:55:35.492Z in script 3, context 1,
target 4DEF694A53708B96B8483A2691ABD0F9, original session
FEFA86F8DFE8157B36FEBE8C2D6FE787. ready.json records frozen step 0,
page compilation of the exact normal module, its original init transfer at
slot 0, and that Worker's receipt and instantiation. The passive attachment
report confirms one pre-instantiation installation, no entry error, no trace,
and an installed observer with empty records. This is real Chrome evidence,
not the earlier fake-CDP controls.
binding-qualification.json records the first rejected prerequisite at
15:56:05.874Z: the original page observation stores the Worker constructor's
relative argument lib/guest-worker.js?v=dev; CDP identifies the same resource
as http://127.0.0.1:41969/lib/guest-worker.js?v=dev. The original attachment's
literal filter has zero matches; resolving against the actual page produces
one. Thus its existing begin gate would reject the original transfer despite
the recorded compile/slot-0 receipt. The gate itself was not armed or bypassed,
and owning thread 1 was not qualified. No guest-state inspection, grant,
navigation command, game input or passive trace followed this prerequisite
failure. Partial native/callback/application history remains unknown.
The separate read-only qualification session successfully wrote its receipt,
then its session.detach() helper returned No session with given id.
The connection disconnected in finally. This secondary helper error is
retained in remote-commands.jsonl; it does not turn the rejected binding into
success. The only mailbox command was normal quit, index 0, acknowledged at
15:56:15.197Z. No observation command was queued before readiness.
The owned successor now resolves sent-module URLs against the actual owning page at the attachment boundary. It preserves raw page evidence, origin, path, query, event order and duplicate events. The published attachment, observer and guest Worker remain unchanged. Its focused regression reproduces the original literal failure, accepts equivalent relative/absolute spellings, and rejects twelve foreign, malformed, ambiguous, hash/slot/session/order controls. Existing first-script controls and actual retained composition pass; route and all original budgets remain unchanged. This repair is offline and was never staged into or retried on the closed attempt. A fresh separately allocated original attempt must prove the repaired binding and owning thread before proceeding to personally reviewed title/menu/preset activation.
Both focused suites passed, including the published attachment's shipped-Worker
controls on both module variants, and
flock /tmp/wine-build.lock bash tools/build.sh passed. That later local build
included concurrent peer D3DIM work and produced normal 61a88dd4be80cc2ce216d6a4efd03ec1e223bb82c31a3f930688675184388f92
and compatibility 1e2861ac4bdaa3e5f8cd581b9e94b809bef940fe36b7897a6170a53c26a8c457.
Those local artifacts were never served on this attempt; the native module
identity remains the earlier 4417305d… pin. The URL/bootstrap successor
controls also passed after that build.
Recovery retained 25 unique files, 1,375,588 bytes, with full SHA-256 and size
matches. A second inventory matched all 24 stable files; process sampling is
recorded separately. I personally viewed the only PNG,
selector-before-launch.png: it shows the selected Stronghold demo, frozen
step 0 and the initial blank emulator surface. It proves no game scene.
All served-file pins matched. Git SHA plus relevant binary diff and the new
note patch are the source archive; no source tree archive, repeated media or
module copy was created.
Chrome PID 18720 exited normally with code 0 at 15:56:15.197Z; the PID was
absent and HTTP 41969/CDP 37779 were closed. Custody was returned on the board
at 15:56:33Z, with bx_8kfdtfgw left RUNNING (API state=idle), healthy and
hydrated, expiry 2026-11-04T16:33:27.336Z. No subsequent box access occurred.
Coordinator owns snapshot and shutdown. There was no repeat, reload, closed
attempt resume, lifecycle command, public deployment or subworker.