Conan live Worker ownership and terminal capture, 2026-11-17
Retained offline evidence identifies the previous owner unavailable cause:
the scratch server selected only JS/HTML Git files, then refused the required
lib/host-import-sigs.generated.json. Its console explicitly reports
worker start failed (sigs HTTP 404) and cooperative scheduling. This was a
local preparation omission, not evidence of delayed Worker readiness. That
original remains closed; no query was sent to its endpoint.
The new scratch helper serves JSON from the same exact
b063d11ee8b53ca6ee4239f9c56b3f7691d4e9c2 source pin, with only published
da9ac26cc8c56a161b044e3705873420aecadaed host-imports overlaid. The retained
module SHA-256 is
87ef39648a61ed6f54ca4ab8de1926de02c651eb2d99655015dc7e5de5407009.
The pinned region map and existing media were reused without copies. Current
generated WIP was excluded. Earlier trusted key/mouse-UP, shell, custody,
display regression and canonical-build proofs were reused.
A disposable real Notepad guest used that module and map. Delaying the real signature response 2500 ms produced 16 pending readiness samples before a valid running app/process/visible window/instance/memory/module/main slot and endpoint binding. CDP independently recovered the executing guest Worker's matching module bytes. Nine changed owner fields failed the same deep identity comparison used for input admission; normal host retirement made the owner predicate reject. No fake owner or guest export call was used. Its whole shell and real Notepad window images were personally reviewed. This qualifies guest ownership; it does not qualify Conan's separate render startup.
The ONE authorized fresh Conan original then launched after personal review of its whole prepared shell. Eleven passive pending samples preceded a valid guest owner receipt: process 1000, HWND 65537, main slot 0, endpoint 1 and the matching executing Worker module. The first capture was premature for the renderer: its before/after identity comparison caught the new D3D render Worker and render endpoint appearing during the PNG operation. The helper waited for guest/window readiness but omitted render-endpoint readiness. Ordinary renderer startup is not a game failure. The strict capture rejection was preserved and is terminal UNKNOWN. There was no retry, subsequent runtime query, guest input or replay. The retained whole PNG was personally inspected offline: it shows only the shell/startup log, and has no accepted lifetime receipt. Menu, New Game/Warrior, movement, second response and native Quit remain unproved. Published da9ac's visible result remains unvalidated here.
Both independent watchdogs were armed before Chrome with actual deadlines and the unchanged 1,500,000,000-byte disk floor; neither fired. Both scopes closed cleanly with empty children. Chrome, driver, watchdog, supervisor and keeper have actual waits and both EOFs. Unique custody files were retrieved and SHA-256 read back before removing only owned tmpfs directories. No box was created: ID/expiry null, state NOT_CREATED, stop receipt N/A.
Evidence, exact relevant patches, served-pin audit and lifecycle receipts:
scratch/runs/20261117-conan-live-worker-owner-menu-gameplay/result.json.
No production defect or product-code change was established, so no new
production build was needed. Further work must qualify passive render-endpoint
readiness on a disposable real rendering guest before any newly authorized
Conan original. This original is consumed and must stay closed.