Quake II: shattered window to lower passage, 0537

One fresh original Outer Base run visibly progressed through the shattered window, around the crates inside its room, and down a floor hatch into a distinct red-lit lower passage. I personally viewed each fresh decision PNG before issuing the next command. The final PNG shows the blaster and HP85 in that lower passage. No guard encounter, operated door, kill, or damage cause is established.

Evidence: scratch/runs/20261109-quake2-shattered-pane-onward-0537/ (ignored).

Decisive visual evidence

HP fell from 100 to 99, 95 and 85 during traversal; damage cause is UNKNOWN. Command names describing attempted doorways or ramps are intentions, not proof of those events. There were no later input commands, captures or runtime retries after the final reviewed success image.

Exact production preparation and input

Preparation Git base: 1e0b662f93dab635fa9f17f68e9488c9a6ea3b4f. Published sparse reader 6e18a9865db6764b10e6f0becc1b9b951a4dac79, persistent caller 1473c5952a52d5d2b4546b93a0d462b64fe5c6cd, and custody 4341b87723a7d56bc98223215cca4cdc878fcae2 were pinned explicitly. Later shared custody edits were not staged into this box. CPU module: 2185596 bytes, SHA-256 a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295.

runtime-pins.json covers 722 staged providers (110089205 bytes); its SHA-256 is a0f1d9d5ff39dbc10cf90bc145acec5122059741dad2093ccb7484d94482604a. The production compiled driver passed offline composition/preflight before box creation. Both strict flags, canonical seven EXEs, owning-memory identity, pre-CPU/pre-DLLmain zero-entry barriers, all work/ACK/hold/freshness/close caps were unchanged. Native lifetime remains REVOKED; all-writers and edict ABA remain UNKNOWN. No guest writes or rebinding were used.

The first actual sparse read before gameplay found Enter13 +attack at 150936080, Control133 +movedown at 150937392, and Mouse1200 +attack at 150935088. All 354 samples were SNAPSHOT on one memory identity c2043b68-3b75-43af-8ac8-f5bc2aca0ad2:9/c2043b68-3b75-43af-8ac8-f5bc2aca0ad2:10. These latch snapshots are not event-consumption proof. The visible consequences come from the personally reviewed images. The run used 47 native holds and two capture-only loading waits, with 50 capture receipts including startup (51 PNGs including the additional held-fire image).

The persistent caller retained one stdin PIPE, started concurrent stdout/stderr drains before start, and kept stdin open throughout gameplay. Transport spawn was not treated as browser readiness. Actual DevTools readiness took 31975 ms; Chrome was 151.0.7922.108, JS 15.1.206.16. Original startup was 05:41:10.500Z, ready 05:41:26.253Z.

Tests and recovery

No production source was changed. As requested, the exact SHA/source-matched 0517 test/build proofs were reused rather than rebuilding different shared sources: offline-preparation.json records 19 verified size/SHA references, including the 0505 focused, composed, comparison and locked-build logs and the 0158/0415/0435 supporting proofs. prior-0517-verified.json verifies 523 prior evidence rows, including 279 gzip CRC checks. Fresh controller tests passed the actual prepared launch/request branches for open-command-then-EOF and EOF while pending, with retained PIPE, actual waits and distinct stdout/stderr EOF.

verify.py passed: all command ACK releases, unchanged barriers, sparse identity, all 51 PNG chunk CRCs and decompression, capture hashes, command/capture ordering, retrieval hashes, actual waits and kernel journal reap records were checked. gameplay-and-drain-verified.json records the result. Free /home stayed above 1.5 GB (1985085440 bytes at verification); no RAM gate or old-run deletion.

Recovery uses Git base plus small binary patches and existing media/module path/SHA references. There are no source, media or module archives. All 667 remote regular-file rows (26292639 stored row bytes, 366 newly stored unique contents) were retrieved and independently size/SHA checked; gzip streams passed CRC/decompression before delegated archive. retrieval-verified.json.gz maps every remote path to stored content, including contents already fetched for personal PNG review.

Runtime release and actual drain

Super0525's ARCHIVED PHYSICAL_DRAINED release preceded Warcraft0535. The exact Warcraft0535 release was observed at 05:38:57Z: bx_92fqhutz, UUID 1bcb6c26-4bd9-48e9-8970-b2ccc39a9734, stop_4c1cd00a31228a2e, completed 05:38:39.775Z. No peer runtime was modified.

An observable independent watchdog was armed at 05:39:36.983Z, before create, against immutable UUID ccedbe56-cc5a-4dbd-a816-69cc1d4b5e5d and a 25-minute deadline. Exactly one box was created: bx_fyfyrenn, finite TTL1491 seconds, initial expiry/archiveAfter 2026-11-09T06:04:36.934Z.

After recording visual success, intentional stdin EOF revoked input. The first failure receipt is caller input revocation during this cleanup. No subsequent browser/original input or capture was attempted. This was independent no-input cleanup, not a successful normal browser-close trial. The kernel-cancelled launcher exited -9; local caller and remote parent/supervisor actually waited with status1 and distinct stdout/stderr EOF. No absent terminal receipt is promoted into a close proof.

Kernel custody records 353 owned / 353 actual reaps, 3955 journal events, complete true and command-log EOF true, with no remaining, unreaped or unresolved ptrace/lineage/resume entries. Each reap was matched to its before-reap record, PID/start identity and status. Tracked runtime processes, ports42339/42577, descriptor writers and runtime/input locks drained; the profile directory itself remained present. Historical all-writers UNKNOWN is not upgraded by a census.

Archive was requested only after retrieval verification, at 05:53:25.895Z. Final box state ARCHIVED, stopped 05:53:26.254Z; receipt stop_e280fc7496b4f011, completed 05:53:26.273Z, forced false, error null. Final archiveAfter is null; expiresAt was not returned. Watchdog and lifecycle supervisor both actually waited status0 with separate stdout/stderr EOF, and the local lifecycle lock was released. PHYSICAL_DRAINED was posted at 05:53:35Z; the 05:53:56Z board correction fixes only the two capture subsecond transcriptions.

The auxiliary local messageboard watcher returned status143 without its inner tail wait/EOF receipt. Its stdout gzip lacks the closing trailer; the original bytes are preserved and a partial decompressed text recovery is marked unqualified. No tail wait or separate EOF is inferred from process absence. This does not affect the independently verified remote runtime retrieval and actual custody/caller/watchdog/lifecycle waits above.

Done: visible onward passage beyond the pane verified. No further task or runtime attempt is scheduled; single-task STOP.