Warcraft III qualified production attempt, 0535

The sole authorized attempt failed during Chrome startup. No version, renderer qualification, original game navigation, ready marker or objective/control PNG was produced. Evidence is retained in scratch/runs/20261109-warcraft-qualified-native-objective-0535/.

The first failure at 05:37:54.962 UTC is DevTools startup deadline; no endpoint. The actual startup observer began at 05:36:54.961 and reached its unchanged 60-second startup bound at 05:37:54.961. The overall browser window was 05:36:51.624–05:38:18.682 within the immutable 90-second case, followed by the existing five-second custody extension. No command, input, capture or retry followed the failure; only independent cleanup and evidence retrieval ran.

chrome.stderr contains a DevTools listening URL for port 42411 and DBus errors timestamped 05:38:06.304. The listening line has no timestamp. It does not prove when the endpoint file became available or that startup could have observed it before its deadline. The cause of slow startup or endpoint observation remains UNKNOWN. No concrete Warcraft production defect was isolated, so no production code or cap was changed.

Preparation and unchanged proofs

Read AGENTS/CLAUDE and WC0512, WC0500 and WC0445. Reopened WC0512's final qualification, startup/cancellation, renderer, contracts, staging, production preflight and locked-build proofs against their evidence-index SHA-256 and sizes, including gzip CRC, actual waits and separate stdout/stderr EOF. Verified the tested source hashes and all 626 final repaired production providers against current bytes. These exact proofs were reused; no code changed.

The fresh prepared closure contains the published 838d4f0df private bounded qualification metadata/trace channel and 16c496ca2 startup cancellation repair. Actual composed production preflight ran offline in this assignment; the fresh 629-provider upload and staging omission/SHA/path/symlink negatives passed. The remote barrier then matched all 161,568,829 staged bytes. These checks started neither Chrome nor the original. The sole actual Chrome run did not reach qualification and therefore does not validate the bootstrap fix in Chrome.

CPU remains SHA-256 a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295; custody remains published 4341b87723a7d56bc98223215cca4cdc878fcae2. Strict lineage, all observation/ACK flags, pre-CPU/pre-DLL-main requirements, 90-second case, work/ACK limits and 14/16/18-second cleanup schedule remain unchanged. Historical WC0445's 53 missing waits, WC0410's creator lineage and WC0512's creator10418/start99598 remain UNKNOWN. No claim that 18 seconds would have passed a historical case is made.

Actual termination and waits

Independent Browser.close was unavailable without a CDP connection. Cleanup sent TERM after 14 seconds and KILL after 16 seconds. At the full 18-second bound, the browser receipt still had terminal=null, closed=false, stdoutEOF=false, stderrEOF=false, drained=false, cancelled=true and normal=false. The startup failure remained the first reason. This is an observed unsuccessful full 18-second cleanup in this case, not a prediction about earlier cases.

The unchanged custody receipt records 83 owned identities, 15 actual terminal waits matched to before-reap sequences in 643 events, 68 missing waits, commandEOF=false and complete=false. Known fork lineage is complete with no unresolved lineage/resume/ptrace rows. The creator witness retains 7,071 actual scans and 201 paired actual CONT-before/return records. The missing 68 waits and browser terminal/EOF remain UNKNOWN. Parent and caller waits and their separate pipe EOF receipts are retained separately; they do not supply Chrome EOF.

A later cleanup census found all 83 recorded identities absent, service port 37755 closed and no profile FD writers. This supplies physical drain only. Port 42411 was recovered from stderr; its closure was not independently socket-tested in that census. The owned profile cache was inventoried before removal. No shared custody, Q2 or Super files were edited.

Runtime lifecycle and recovery

Observed the exact Super0525 ARCHIVED PHYSICAL_DRAINED release, stop_6966241a5392ff24, before creation. An observable independent watchdog was armed before create against immutable UUID 1bcb6c26-4bd9-48e9-8970-b2ccc39a9734 and a 25-minute deadline. Before creation, a routine offline wrapper error was corrected: the lifecycle supervisor was removed from a 180-second test wrapper and restarted independently against the same immutable plan. The earlier actual wait/EOF and correction are retained. Caller stdin stayed open with concurrent separate stdout/stderr drains.

Box bx_92fqhutz used TTL 1466 with initial expiry 06:00:50.082 UTC. All 27 unique remote evidence files, 336,185 stored bytes, were retrieved and independently verified for SHA-256, size, decoded gzip hashes and GNU gzip CRC before normal delegated archive. Stop stop_4c1cd00a31228a2e completed 05:38:39.775 UTC, forced=false, error=null. Final state is archived, archiveAfter=null; final expiresAt was not returned. PHYSICAL_DRAINED was posted 05:38:57. The local lifecycle and watchdog have actual wait 0 and separate pipe EOF; the runtime lock is released.

Recovery records exact Git base plus small relevant dirty/new-input patches. Media and CPU modules remain at their existing SHA-pinned paths. No tree, media or module archive or duplicate was created. Free/home above 1.5 GB was the only resource gate. No agents, workers, local browser or deployment was used.

Next: investigate the retained startup/endpoint timing and missing waits offline. No further runtime or original retry is scheduled. Single task STOP.