Warcraft III fresh repaired-custody assignment, 0558

The single authorized fresh assignment stopped during provisioning. No Chrome or original game was launched, and no objective/control PNG was produced. The concrete blocker is an HTTPS upload write timeout and an incomplete remote closure: cleanup found 628 of 629 providers with exact SHA-256 and size, with only test/binaries/candidates/warcraft3-demo/war3.mpq absent. The local original is 101,868,957 bytes, SHA-256 9e19d7ffb65054e4bdd3add7e26d70e16f063cdc20fa5b57639f3b9b7a196f9c.

Evidence: scratch/runs/20261109-warcraft-repaired-custody-objective-0558/. causal-diagnosis.json, the provision command/process/wait and separate output logs, and retrieved-runtime/cleanup-only.json.gz retain the actual failure and independent cleanup inventory. The provision process actually returned 1, with both pipe EOFs observed, by 05:59:48.308574 UTC. Its untimed traceback records urllib.error.URLError: <urlopen error The write operation timed out> from SSL socket send inside the upload PUT, using the existing 90-second transport socket timeout. This is the terminal observation bound, not an invented time for the failed syscall. The helper did not log individual PUT identities or exception timestamps; exact failed-request attribution and exception time remain UNKNOWN. The absent MPQ establishes incomplete staging, without proving that its PUT produced the traceback. No later upload, closure-barrier, browser, input, capture, gameplay command or retry was attempted after the provision caller returned. Only independent no-input cleanup, evidence recovery and archive followed. Concurrent transfer tasks had already settled before that actual provision wait; their individual ordering is UNKNOWN.

Offline closure and reused proofs

Read CLAUDE/AGENTS, WC0545 repair and WC0535 failure notes. Prepared a fresh 626-provider production closure and 629-provider upload manifest. Actual composed DRIVER_PREFLIGHT=1, staging omission/SHA/path/duplicate/symlink negatives and local closure fingerprint passed before box creation. The fresh remote closure barrier was never reached, so there is no claim of successful remote preflight or original execution.

Reopened final0545's ten successful validation logs and 306 selected indexed proof files against their exact SHA-256 and size, including final composed and startup-control gzip CRC checks. Verified every production provider and the repair's tested source hashes against current bytes, and confirmed no tracked changes in src/lib/host/tools. Reused the exact final0545 locked build, 40 custody tests, 13 composed startup/race cases plus direct controls, qualification composition negatives, 38 renderer controls and contracts. The actual old-fail/ new-pass cancellation proof remains offline evidence; it is not a new Chrome acceptance result. Prior incomplete preliminary evidence was not substituted for a successful final proof or modified.

Production remains c4b4cd337d5c21ef27441451d56596ffc96c46b6, with Warcraft custody SHA-256 5ec810c4d4704ff86afc367b38f7335edf1214e3450ebb98a62ce3df6386b5fe and CPU a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295. Strict lineage, all caps and ACK checks, pre-CPU/pre-DLL-main zero-entry barrier, immutable 90-second case and 14/16/18-second close schedule are unchanged. Historical missing waits, creator identities and startup endpoint readiness timing remain UNKNOWN. No deadline increase, late-endpoint acceptance, Q2/Super provider edit or production repair was made: this failure demonstrates an incomplete upload, not a Warcraft production defect.

Lifecycle and recovery

Observed exact Super0542 ARCHIVED PHYSICAL_DRAINED, stop_5684743e5f0bb8aa, before creation. Independent observable watchdog was armed at 05:57:39.266 UTC, before create, against immutable UUID a984fdd2-42c2-46ec-902a-f4584ed5452d and deadline 06:22:39.112881 UTC. The one box was bx_cy9256js, TTL 1488 seconds, initial expiry 06:22:38.982 UTC. Cleanup at 06:00:31.671 UTC found no owned runtime/Chrome process, case directory or Chrome profile. One unique remote cleanup gzip was retrieved and independently verified for stored/decoded SHA-256, size and GNU gzip CRC before delegated normal archive.

Final state: archived. Stop stop_16b968537401602e completed 06:00:46.062 UTC, forced=false, error=null; archiveAfter=null and final expiresAt was not returned. Local lifecycle and watchdog actual waits returned 0 with separate pipe EOF; runtime lock was released. Posted physical drain promptly at 06:00:58 UTC. No box was deleted or resumed.

Recovery records exact Git base, relevant dirty patch, small existing untracked dependency patch and new run-input/note patches. Media and modules are referenced by existing paths and hashes; no source/media/module archive or duplicate was created. Free /home stayed above 1.5 GB, the only resource gate. No agents, workers, local browser or deployment.

Next: diagnose the provider upload transport with per-request timed evidence before any separately authorized fresh attempt. This assignment schedules no further runtime. Single task STOP.