Quake II persistent caller engagement attempt, 0450
Guard/door objective not achieved. One fresh original reached its main menu using the persistent caller. Its first passive binding sample cannot decode any of the three actual binding strings. This unexpected UNKNOWN was terminal: there was no gameplay input, later capture, second original or retry. The WIP and first-failure evidence are preserved.
Evidence: scratch/runs/20261109-quake2-persistent-native-combat-0450/ (R).
Read the 0435 caller, 0425 failure, 0415 observer and 0350/0325 gameplay notes.
Personally inspected the retained armor100HP/25armor, outdoor46HP and death
PNGs; paths, SHA-256 and sizes are in prior-reviewed.json. They are prior
evidence and establish no new gameplay for this attempt.
The 722-provider closure uses published caller/composition
1473c5952a52d5d2b4546b93a0d462b64fe5c6cd, the exact passive reader from
c58857772, and custody directly from published
4341b87723a7d56bc98223215cca4cdc878fcae2. CPU SHA-256 remains
a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295
(2,185,596 bytes). Both all-caps observation flags, pre-CPU/pre-DllMain
zero-entry barriers, and work/ACK/close/hold/freshness limits are unchanged.
Fresh actual composed-driver preflight passed without Chrome or an original. Fifteen prior focused/build log paths were rechecked by exact stored size and SHA before reuse. The prepared run-specific controller starts concurrent stdout and stderr drains, awaits OWNED_INPUT_READY, flushes the explicit start on one retained PIPE, then retains that same writer for commands. Its launch-started record proves transport spawn only.
The fresh offline lifecycle regression uses this actual prepared controller and task's launch/request branches with the production caller. The explicit Boat adapter runs real local interpreters and children instead of a remote browser. Open-pipe setup plus a numbered command/ACK passes; EOF during a pending command is sticky terminal and produces no ACK or later command. Both cases retain actual waits and distinct stdout/stderr EOFs, including the adapted remote parent and interpreter. No box, network or game is used by these fixtures. Closing the writer is terminal; it is never reopened.
Source recovery is the exact Git base plus closure-recovery.patch.gz
(84,797 uncompressed bytes), covering relevant uncommitted hunks and the small
untracked providers. Media/modules are referenced by existing path, size and
SHA in the manifest. No source/media/module archive is made. Native source
lifetime remains REVOKED, all-writers and edict ABA UNKNOWN, and historical
shutdown causes UNKNOWN. No production code change, peer edit, agent/worker,
local browser, rebinding, guest write or deployment was used.
Concrete first failure
The caller passed OWNED_INPUT_READY and the flushed explicit start, and retained
the same writer through readiness and its sole no-input cleanup command.
original-controller/transport-started.json proves transport spawn only.
The actual remote strict startup receipts separately establish both all-caps
flags and zero execution entries before release of the parked DLL load, with
trace [1,0x400000,0x500000]. The startup capture's Worker and renderer modules
both match CPUa320. The startup PNG was personally inspected: main menu, no
guard, operated door or gameplay progress.
review/startup-binding.json retains the exact point sample. The overall reader
status is SNAPSHOT and its five-word attack reads are equal, all zero. But the
individual live binding results for engine keys13/133/200 are all explicitly
UNKNOWN sparse/outside direct window, text=null:
| Engine key | Actual string pointer | Live binding |
|---|---|---|
| 13 (Enter) | 0x08ff1a10 |
UNKNOWN |
| 133 (Control) | 0x08ff1f30 |
UNKNOWN |
| 200 (Mouse1) | 0x08ff1630 |
UNKNOWN |
The concrete cause is the reader's supported-address boundary:
tools/quake2-passive-attack.cjs requires va+n <= 0x04000000 in direct();
binding() immediately returns UNKNOWN for these pointers instead of reading
or translating them. Thus the published reader cannot establish this run's
actual bindings. Config defaults Enter=+attack, Control=+movedown do not
resolve the live result, and no rebinding or guest write was attempted. Physical
keys are all released in this sample. Neither the zero latch nor the menu PNG
proves attack consumption or weapon effect. This is a specific reader coverage
blocker; no attack delivery or game simulation defect is inferred.
The sample was captured at 04:55:05Z and its unknown binding results were
personally inspected before the local terminal record at 04:55:46Z. The only
numbered command is cleanup-only, action=quit, on the same retained writer.
No hold, keyDown, later PNG or capture command followed the UNKNOWN. The existing
automatic native keyboard preflight predates original launch and is not gameplay.
The retained startup PNG SHA-256 is
fbe0e49ee0e81ea3ef9c5bcbe3d938c0e7092b35a5b0b08b1aced89b69fadf5c.
Cleanup, evidence and release
The no-input quit failed at 04:56:06.596Z: Browser.close ACK, TERM and KILL, terminal=null, drained=false, elapsed19,867ms under the nested19,862ms cap, normal=false. There is no successful numbered quit ACK. The local caller actually waited1 with separate stdout/stderr EOFs; its launch transport waited-9 and its request transport waited1, also with distinct EOFs. Sticky remote revocation was delivered and the existing supervisor drained its owned descendants. These cleanup failures remain failures; historical shutdown causes remain UNKNOWN.
Kernel evidence has 188 owned identities and 188 actual terminal reaps, verified against 1,818 journal events and their before-reap witnesses. Drain is complete, command EOF true, and remaining/unreaped/unresolved sets empty. Actual remote supervisor wait1 and each pipe EOF are retained separately from kernel waits; its receipt still says remoteCustody UNKNOWN. The census records ports36375/ 43765 closed, all tracked runtime identities absent, no profile FD writers, input writer absent and runtime lock free. Untraced/nonleader/all-writers and edict-ABA coverage are not upgraded. Local lifecycle/watchdog actually waited0 with separate EOFs; the local runtime lock was verified free. The board watcher actually waited-15 with separate stdout/stderr EOFs and complete gzip logs.
Both exact prerequisite releases were observed in order: Super0445 at04:49:04Z,
then Warcraft0445 at04:52:44Z. Only then was this box created. Its immutable
UUID is 11a3284d-63fb-42d3-9faa-219a6cad9c36; the independent observable
25-minute watchdog was prearmed before finite-TTL1496 creation. Box
bx_nteu4dds had initial expiry 2026-11-09T05:18:06.327Z. All47 remote
regular artifact rows/640,531 row bytes were retrieved, deduplicated (39 newly
stored contents), and verified by stored SHA/size and GNU gzip CRC/decompression
before delegated normal archive. Final state ARCHIVED, stopped
04:57:13.409Z; receipt stop_479d99a7075aa948, completed
04:57:13.430Z, forced=false, error=null. Final archiveAfter=null; expiresAt
was not returned. The PHYSICAL_DRAINED release was posted at04:57:39Z.
No production code changed. Routine offline inspection errors (ambiguous short custody SHA and a deduplicated receipt's physical filename) were corrected in this assignment; neither was treated as a resource occupancy or runtime failure. The prepared controller's actual offline lifecycle regression passed before creation. Free /home stayed above1.5GB. Final validation covers retrieved SHA/ sizes/gzip CRC, PNG chunk CRC/IDAT, strict startup, pointer diagnosis, zero gameplay commands and no later captures, actual kernel waits, distinct EOFs, archive receipt and local drain. Evidence remains ignored; only this note is committed. Next is offline support for passively reading these real sparse binding pointers with identity/translation validation, followed by a separately authorized fresh attempt. This single task stops after publication.