Published asynchronous desktop trace: genuine agreement reached

Evidence: scratch/runs/20261117-halflife2-published-async-trace-normal-install/result.json.

Pure preparation was corrected locally before any original process existed. The published controller's durable completion contains elapsedMs and limitsMs; aggregateGuardPassed belongs to its actual stdout. The new run-specific verifier checks both actual outputs, DOWN consumption, independent physical UP, complete-image hashes, current-page reviews, one binding and actual native/outer/owner/watchdog waits with both EOFs. It does not invent a completion field or skip the adapter's source checks.

A fresh published dfc48cba661bb50a831ba19851bfd7e2c0309c0b authored workflow passed through Welcome, Internet Required and SubscriberAgreement. Each whole image was personally viewed and its metadata separately inspected. Next HWNDs changed from 10004 to 10006 to 10007. The complete workflow receipt was verified and written before original preparation. Host cleanup ended the authored fixture; this is not native Quit.

The unchanged published Edit/component proofs and retained module 53c2791da77f4e751c92b77a02f1503ea23741b695db1073230f1b4116eaba2e were reused with all WAT/compiler input hashes checked. Exact Git and relevant binary patches, sealed source inventory, module-before-execution and unchanged 787,701,212-byte original media SHA-256 are retained. No source tree archive, box, worker, duplicate original EXE or public deployment was created.

One authorized ordinary original installer invocation followed. Fresh public metadata and personally viewed whole images preceded each control. Genuine Welcome Next was 10011/generation17; Internet Required Next was 10017/generation23. Both normal inputs passed the unchanged native ACK500, independent UP1750 and aggregate3000 guards, with aggregate completion about 461 ms and 412 ms respectively. There was one input binding.

The subsequent agreement capture timed out on its original 3-second HTTP deadline and permanently latched UNKNOWN. Its PNG appeared later, after the latch. I personally viewed that whole late image: the title reads Steam Subscriber Agreement and the native Edit displays the opening agreement and registration/activation text. The button captions are blank in that frame. This establishes visible genuine agreement progress, but not a successful guarded complete capture, a reviewed input target, acceptance or installation. No later HTTP/query/rebind/replay/retry occurred.

Read-only independent process samples show the main thread accumulating CPU and scheduler wait while broker threads slept during most of the delay. All retained desktop trace slots have valid checksums and no first-error record; there is no demonstrated ProcessDesktop IPC timeout in this invocation. The exact capture delay's source remains unresolved: these observations do not identify snapshot validation, presentation, guest initialization or scheduling as the cause. No speculative runtime repair is claimed, and the earlier historical stall attribution remains unresolved as well.

UNKNOWN triggered orderly host SIGTERM cleanup. Native, outer, persistent owner, process observer and independent watchdogs have actual waits and both EOFs; the owned scopes are empty. The late PNG cannot reopen the endpoint. Further installation needs diagnosis of this capture delay and a separately authorized fresh original invocation. Agreement acceptance, installation, authentication, gameplay and native Quit remain unproved.

Validation also passed the canonical locked build and the published native observer-stall regression: old baseline timeout at about 3009 ms; repaired 500-poll runs completed in about 502/451/435 ms for slow sink/EIO/short write. This task changes no production code and claims no repair of the new capture delay. The local preparation correction and its actual schema checks are preserved as run-specific evidence.