Quake II live attack attempt, 0425
Guard/door objective not achieved. I launched the owned caller without a
persistent input stream. Its stdin reached EOF immediately, and the caller
latched terminal UNKNOWN (session closed) at 04:25:13.166589Z. The local
launch transport actually waited with status -9 and separate stdout/stderr
EOFs. There was no command, native hold, PNG capture, retry or second original.
This is a launcher setup mistake, not evidence of a game attack defect.
Evidence: scratch/runs/20261109-quake2-live-attack-combat-0425/ (R).
The 0415 repair and 0350/0325 notes were read. I personally inspected the prior
0350 death PNG (b58a7645…), its preceding outdoor HP46 PNG, and the 0325
Jacket Armor25/HP100 PNG (80c5d2c0…). These are prior evidence, not new
progress. Enter is the shipped +attack, Control is +movedown; this run
never reached a live reader sample, so actual live bindings, attack latch,
consumption and weapon effect remain UNKNOWN. No shot, damage, engagement,
kill, operated door or level transition is claimed.
Before allocation, the new 722-provider integration closure and actual composed
driver preflight passed without starting Chrome or an original. The CPU remained
2,185,596 bytes, SHA-256
a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295.
The composition and passive observer were exact published
c58857772013727cee0527553e82c9ddcf1c31ef bytes. Custody was explicitly read
from published 9ab247857d0f5b341d2502d8b7191a20b4dfd762, SHA-256
a2d9ca6f74bdb193e017c3b340a450ee0b0b13736030e0d512ff500b26fda9b3;
concurrent 0420 WIP was excluded. Local and remote manifest barriers matched,
SHA-256 73d9c2aeaa57c3537d417cc6c61dbab1d1a2d8246eb83d43a06ac34845f7405a.
Both all-caps WINE_QUAKE2_LOAD_OBSERVATION and
WINE_QUAKE2_NATIVE_WRITE_OBSERVATION remained true in the staged manifest.
The strict before-CPU/pre-DllMain code and zero-entry barrier were unchanged,
as were work/ACK/close/hold/freshness limits. There is no actual strict startup
barrier receipt, launch receipt, ready file, or guest execution proof in this
attempt. Automatic browser bootstrap continued after local transport loss:
six HTTP rows and a DevTools-ready record are retained. No PNG or input followed
the local terminal. Original guest execution is NOT ESTABLISHED.
After detecting the terminal, I used only identity-checked SIGTERM to the owned
remote supervisor. Kernel cancellation escalated to KILL; this was not normal
browser close. All 126 owned identities have 126 actual terminal reaps, with
before-reap witnesses in 1,355 journal events. Kernel drain is complete, command
log EOF true, and remaining/unreaped/unresolved-lineage sets empty. The remote
supervisor's parent actually waited with status 1 and separate stdout/stderr
EOFs. Ports39133/39373 were closed, tracked runtime identities absent, profile
FD writers empty, and input/runtime locks drained. Kernel waits and stream EOFs
are separate evidence; remoteCustody: UNKNOWN remains in caller receipts.
No all-writers, nonleader or historical shutdown qualification is inferred.
The exact WC0410 box bx_s55bpw94, UUID
574f1f83-256b-443b-9e86-260c217f9e65, ARCHIVED PHYSICAL_DRAINED release at
04:14:54Z (stop_652973db00dab7ea) preceded creation. Other workers were
offline. This task's sole box was bx_998p4xja, immutable UUID
be9bb475-1344-4344-8a65-640ad4e00744. An independently observable immutable
25-minute watchdog was prearmed before finite-TTL1494 creation. Initial expiry:
2026-11-09T04:49:29.726Z. All 23 unique remote artifact files / 148,356
stored bytes were retrieved and verified by SHA/size and gzip CRC/decompression
before delegated normal archive. Final state ARCHIVED, stopped
04:26:13.139Z; receipt stop_045211242240670a, completed
04:26:13.159Z, forced=false, error=null. Final archiveAfter=null;
expiresAt was not returned. Local watchdog and lifecycle actually waited zero
with separate EOFs; the board tail actually waited -15 with separate EOFs.
Physical-drain release was posted at 04:26:43Z.
No production code changed. Accepted caller/shared-Q3 checks plus 0415 passive reader, composed-driver, keyboard-delivery and locked-build logs were reused only after exact path/size/SHA checks. Fresh validation covers closure, retrieval, reaps, EOFs, zero commands/captures, and local drain. A preparation-script syntax typo was corrected before the passing preflight and before allocation. Source recovery is exact Git base plus small relevant/owned-input patches; media/module inputs are referenced by existing paths and hashes, without source archives or repeated local media/module copies. Free /home stayed above 1.5 GB. No agents/workers, local browser, peer box, rebinding, guest writes or deployment were used. Native source lifetime remains REVOKED, all-writers and edict ABA UNKNOWN, historical cleanup causes UNKNOWN, full27/parity/audio/FPS open.
Next requires a separately authorized attempt with a persistent owned-caller input stream established before launch. This task stops with the blocker and evidence preserved; it does not retry the original.