Stronghold attached original preset attempt, 2026-11-04

This worker owns only the fresh run scratch/runs/20261104-stronghold-attached-original-preset/ and the assigned bx_9eqxpvx4. Coordinator owns box shutdown. No other box, worker, Start, gameplay gesture, production guest-worker edit, deployment or lifecycle command is authorized here.

The entrypoint is the published tools/stronghold-preset-driver-attachment.cjs from f517b6d45eaa1b77bb209451f98f439f1f4be3b6, with the passive observer from 8574dfa540da10ade7ba7b5799c974d131775a38. Its retained original driver and route hashes remain 230a11947699af0721dc54d58c6b027f6ba21563b59bffc7d2014221ec481a10 and bb38b5787c277c94c2913a9ddba72a8663b9fd2fb706a88e320bcb83419633de.

Preparation verified all 2,415 inherited pins against base 1188e9bf57ef35bd3363fb88b7ca0da442ce696d. The initial running box had no Chrome or game driver. Twenty-four small changed runtime/helper inputs were staged individually against recorded main 0b27067a178c12c1591234c7037fa881a5bdb738. The inherited compiler lacked its entrypoint and four vendored compiler stage files; those exact small build inputs were staged individually, then the remote compiler ran under flock /tmp/wine-build.lock. These preparation failures and the initially stale source entries in the newly authored manifest remain in the raw logs; the corrected actual --check passed before Chrome/game execution.

The old module identities were cffca869b541e6eff12ae3a6b74c4476977e0a7e23cd43d8ec5ba8faecd4bcee and compatibility 580f49f3c78d26e6855efa0beb37046d03ea24690ff98cfcb3e94ce5e6cb1ee3. remote-module-identities.json records the rebuilt exact current artifacts, which matched local artifacts. Fresh runtime-pins.json and runtime-release.json retain the required authorization string, assigned box, expiry 2026-11-04T16:14:21.621Z, closeBy 15:59Z and coordinator lifecycle.

Preflight executed the actual composed declarations without calling main, verified the 30-file retained helper closure, Puppeteer CJS and ESM, executable Chrome 151.0.7922.108 and 164 actual HTTP responses with body hashes, isolation headers and JavaScript/WASM MIME checks. The exact compile/send/slot-0 owning Worker binding remains a mandatory live attachment gate before activation; offline preparation is not that live evidence.

The rebuilt normal module is 4417305d70322467441aef51e888b0fe66451d9c60da0c734eeb23f5a22c0a6b; compatibility is 07fb4e75d05177271097dcd67d84e2629272b50073074bbdea8825ce10b4ed34. The sole original run served the normal module with that exact hash; its 144 served-file records all matched the fresh pins. Neither a transferred Module identity nor its slot-0 instantiation was qualified before failure. The preflight does not establish the later live paused-Worker transport gate.

Actual result: bootstrap failure, no preset activation

The one original attachment process launched at 15:42:12.649Z. Its registered selector/Launch was recorded, but no readiness record was published. The first actual failure was at 15:42:55.389Z, phase worker preload: TargetCloseError: Protocol error (Runtime.evaluate): Target closed. Target 7F4FA222158BE85C078BC90A3EBA4481, session 3A71A59A9D049EB0E8175C460D435CFF, detached at 15:42:55.381Z. No successful preload receipt exists. The retained bootstrap subsequently reported its original 60-second readiness timeout. Cleanup's attempted state collection reported Runtime.callFunctionOn: Target crashed; final state and the cause of target closure remain unknown.

I submitted initial observe(2048) command 0 before ready.json existed. That was premature: it exhausted its own unchanged 40-second reply deadline while bootstrap was still pending. The mailbox never executed it; no reply, ACK, observation PNG or guest input was produced. The earlier Worker preload failure remains the first runtime failure. No title/menu/preset review, game gesture, trace start, native frame, callback/application history or preset completion was reached. Absent native/input/ACK/capture correlations are unavailable, not successful empty histories. There was no reload, repeat or resumed attempt.

I personally inspected the only PNG, selector-before-launch.png: it shows Stronghold Crusader Multiplayer Demo selected in the emulator with frozen step 0 and its initial blank surface. It is no title/menu/preset/gameplay proof.

Owned offline dependency repair

tools/stronghold-attached-original-preset-driver.cjs composes the published attachment without editing it or any peer helper. Its bootstrap installs a beforeScriptExecution instrumentation breakpoint while the exact Worker is still waiting for debugger startup. It releases that startup hold, accepts only the first instrumentation pause whose parsed script has the exact Worker URL and an explicit execution-context ID, and evaluates the unchanged preload in that actual context before resuming the first script. Missing context, wrong script/reason, lost target or preload exception fails closed without executing the script. The existing browser deadline bounds the handshake. A failed preload also requests immediate original browser cleanup rather than waiting for the later readiness timeout. No guest steps, inputs, recognizers, route, observer semantics, module identities or original bounds change.

This repairs an explicit bootstrap transport prerequisite; it does not prove why the original target closed or that real Chrome will now succeed. No second browser/game run was authorized or performed. The concrete next boundary is a separately allocated browser qualification of this exact first-script/context handshake and original compile/send/slot-0 binding. Preset activation, settings application and gameplay remain unproved.

node test/test-stronghold-attached-original-preset-driver.js passes exact transport order and seven negative transport controls, rejects late/expired attachment, compiles the actual retained helper/driver closure and proves route and bounds unchanged. These are fake-CDP/VM controls, not browser qualification. The new test is discovered by the ordinary unit tier. Both the published attachment's focused controls and the serialized flock /tmp/wine-build.lock bash tools/build.sh pass. The new successor's actual --check passed offline against a separately recorded current local manifest. The closed run's remote manifest retains its inherited, unused tools/callback-ownership-census.json; this is the only pin difference from the local successor-check manifest, recorded explicitly rather than rewriting the run's pins. No runtime inputs or original evidence were changed afterward.

Recovery and handback

I requested closure of the existing Chrome via its existing CDP endpoint during bootstrap cleanup; Chrome exited 0 at 15:43:53.798Z. PID 29934 is absent, HTTP 45905 and CDP 36085 are closed, and process evidence contains no live Chrome or game driver. Twenty-two unique raw JSON/JSONL/log/PNG files, 1,832,956 bytes, were recovered with full SHA-256 and size matches. A second inventory matches all 21 stable artifacts; process sampling is separately recorded because its command/PID changes. Chrome profile and duplicate closed snapshot copies were excluded. No source/media/module bundle was recovered.

Custody was returned early at 15:45:10Z. Last observed assigned box state: running (API state=idle), healthy, hydrated, expiry 2026-11-04T16:14:21.621Z. Coordinator owns snapshot/shutdown; this worker issued no lifecycle command and performed no further box access after handback.

Git commit plus relevant binary diff is the source archive; no source tree archives, media or module copies are staged into evidence. Original media stays at its inherited path with SHA-256 9024c533095f2227b0105b7e8a9e4cf28ec4fb206b165629787b2c97cf99f13f.