Q2 reviewed closure: terminal production upload timeout

Guard defeat, onward-door progress and gameplay screenshots remain UNVERIFIED. The sole allocated run stopped during provider provisioning, before a startup barrier, browser, changed native fixture, cancellation fixture or original attempt. No production repair was made. This is a concrete transport blocker, not native qualification.

Evidence: scratch/runs/20261109-quake2-reviewed-launch-guard-route-1638/. The exact Git base is 586f4aa441b4ea402710adbde9515702aa15388e with the relevant uncommitted diff and small run-input patch. Source, media and modules were not archived. Media stays referenced by existing path, size and SHA-256.

Fresh production preparation retained the exact 961-byte pipe-wait.py, SHA-256 5c67aa36ea6d6c5251129309f0cd532937bf855c5cecec46bb71d63c9804e933. The final manifest has 742 providers and SHA-256 76a1109e1f13a360244d85b3b604309e18fa623cb0267328726a801d39e4ec14. Its explicit run inputs are the three retained hooks plus this run's control/controller/input scripts. Primary CPU remains 3e491c0840bb7771328738a0d160b4bbaa3d7ab72074cbd38a7b41e5d7e8ec18; compatibility CPU remains ecb45716b1ee7fc5218b1670dd270b19c7664ee30d000b3360a12786a1c21eaf. There was no historical default PATHHASH substitution, blanket repin, policy waiver, local Chrome, agent, worker or deployment.

Before allocation, the actual generated provision branch passed offline with real pinned provider reads/hashes, actual startup barrier and production manifest-file consumer CLI. Only upload and remote transport were local seams. The actual generated launch passed through the SHA-authenticated production owning caller, retained launch helper, current policy, health/fixture/wire hooks and real kernel supervisor at DRIVER_PREFLIGHT. Actual runpy reads of the declared parent helper were audited. Kernel reaps were complete with empty unresolved sets and released lock; generated parent, supervisor and caller waits retained separate stdout/stderr EOF. A direct read/import audit found no provider outside the declared closure and run manifest/barrier metadata.

Inspected focused tests passed: staged provider damage, forged repins, version/authority/default refusal, missing/damaged parent helper and consumer hook controls; current browser-policy tests with inert compile/attach; actual caller signal/exit/separate-pipe terminal controls, pending requests, no later commands and normal completion; retained controller ready/start and early-refusal waits. flock /tmp/wine-build.lock bash tools/build.sh passed; both CPU hashes remained exact. The prohibited browser-owning-binding test was not run. These are offline controls only.

An initial preparation command supplied an incorrect version string and refused before creating a manifest or child. Preparation then used production exports. An initial audit omitted its own run barrier; its ENOENT and the corrected metadata setup are retained. Neither was a native attempt.

Runtime followed Redline1634's final published RELEASE STOP at 16:45:16Z, published commit 5295f2058e98619ac813a25c7d48fcfd17d576fe. Independent rereads verified its no-box outcome, artifact identities/CRCs, seven offline kernel reaps, two actual parent waits and separately joined watchdog. Earlier Q2 bx_y37jnhc6 and Q3 bx_t2ubq3yn were independently exact-API ARCHIVED with completed unforced normal stops, recovered artifact identities/CRCs, physical drain and actual watchdog joins. Their recorded historical UNKNOWNs remain UNKNOWN. No observation fallback ended this task or authorized allocation. Q3's following repair did not become a circular runtime gate.

An independent immutable 25-minute UUID watchdog was armed and independently observed before the only allocation. Box bx_8p7rprn8, UUID f1ab242d-30d3-4643-a43c-4b1259612c6a, initially reported archiveAfter=2026-11-09T17:10:59.649Z. Actual production provisioning failed with urllib.error.URLError: <urlopen error The write operation timed out>; the underlying stack is SSL write/sendall during a provider PUT using the retained 90-second transport timeout. The provision process exited one. No provision retry, native launch, later gameplay input/capture or deadline extension followed.

Read-only recovery verified 741 of 742 providers, with no changed present provider. The sole absent file was the canonical baseq2/pak0.pak. The retained threaded uploader does not record per-provider attempts, so the exact first failed PUT path and network/service cause remain UNKNOWN; the inventory alone does not establish that attribution. An initial recovery assertion returned no census and failed before compression; its exact wait/EOF and stderr are retained, and its failed predicate remains UNKNOWN. A subsequent read-only observation preserved the actual census without discarding its fields. Neither observation launched a runtime or retried provisioning.

One unique compressed filesystem census was recovered and SHA/size/gzip-CRC verified before delegated normal shutdown. No staged runtime argv or root FD holder remained; profile/writer lock were absent and the runtime lock was free. Runtime/kernel tasks and parent waits were NOT_CREATED, not empty reap proofs. Ports/endpoints were NOT_CREATED, with no numbered connect proof. Browser EOF, all-writers, aliases and ABA remain UNKNOWN. Empty or invalid artifacts never supply execution, wait or EOF proof.

Final exact API state: ARCHIVED, archiveAfter=null, final expiresAt NOT_RETURNED. Normal unforced stop stop_a6ca07d03158e881 completed at 2026-11-09T16:50:48.928Z without error. The actual watchdog join returned zero with separate EOF; lifecycle tool exit was zero, its lock was released, and the immutable deadline hash stayed unchanged. No box was resumed/deleted or left billing idle. Observed free /home stayed above 1.5 GB.

Next work is offline upload/transport diagnosis with per-provider provenance; a new native attempt requires separate authorization and fresh gates. This worker makes no further attempt and ends with final published RELEASE STOP.