Drakan qualified flight 0635: live identity gate stopped before T
The one separately authorized fresh browser original reached a reviewed living mounted scene with full red health, ridge, conifers, river and bridge. Its sole T command stopped before physical down at the caller's fresh live-identity assertion. There was no observer arm, T, Up, gameplay heading, attack, rearm or retry, and no later capture or input. Takeoff and onward flight are unqualified. The terminal is a concrete caller gate failure, not a module admission failure or a native-completion UNKNOWN.
Evidence is scratch/runs/20261109-drakan-qualified-flight-0635/. Git base
ecb95f4a8eb91140ed58ccca1f1cb90f9048c99b, the relevant uncommitted binary diff,
and small run-input patches provide source recovery. Media and modules remain
existing path/SHA references. No source-tree archive, media/module duplicate,
production edit, Quake2/Warcraft provider edit, worker agent, local browser or
deployment was made.
Exact closure and offline checks
The regenerated closure has 673 providers, 126,063,451 bytes, with 16 required
physical helpers. All uploads joined before the final remote HASH barrier,
manifest SHA256
124a612df35602f52c26169dee191ab67fcf2df75c5d48abea83734a5c55c5ce.
The released custody provider remained exactly
5ec810c4d4704ff86afc367b38f7335edf1214e3450ebb98a62ce3df6386b5fe,
including the dec33253c/4341 repairs. Concurrent offline custody edits were
excluded by selecting published Git blobs. No provider changed during staging.
CPU is the actually selected current module
a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295,
2,185,596 bytes. Published admission a331104e9, opt-in 0e2bd19d8 and release
repair 296eb1015 remain unchanged. The document preload sets boolean opt-in
before compilation. Actual owning Worker bytecode, ordinary HTTP artifact and
browser retained-compile owner checks authenticate a320; no supplied Module,
canonical refetch substitution, historical module or historical actor pin was
used. Four selected production providers match the exact current admission
proof's hashes. The source-compiler boundary remains unproved by synthetic
source-loader fixtures.
541 retained evidence files were independently reread by SHA/size, including 114 gzip CRC/decode checks. Current a320 native54/browser20/Worker60/import11 and locked-build proofs retain their original scope; legacy4b45 proofs remain separate. Identical accepted tests/builds were reused, not rerun.
Changed run-specific composition was checked before allocation:
- The actual caller and page bridge pass synthetic COMPLETE, UNKNOWN, physical release rejection and wrong-actual-owner controls. Current actual module bytes and independently relocated installed-media code pins are used. Original action start, buffered stamp/sequence and actual release-Promise settlement remain observable; wrong ownership dispatches zero down events.
- The separate Up caller passes delayed release and rejected-release controls, including actual completion time. It was prepared but never used in gameplay.
- The pinned actual native custodian supervises and reaps a Node-only child; complete=true, without Chrome/game. Regenerated helper closure, syntax and actual composed driver preflight pass. These are offline controls, not a physical release or original native-completion qualification.
No exact32, finite observer, CALL/RET, identity, preCPU/preDLLmain, deadline, workACK or freshness cap was changed.
Original and first terminal
Native warning/startup presentations, pointer capture, main-menu New Game,
Normal difficulty and story Enter lead to the living mounted scene. Some native
menu clicks selected only a hover before a later reviewed selection. Four stale
requests were explicitly refused before dispatch with effects=0; only fresh
captures followed those refusals. They are not UNKNOWN retries. The early
warning-fresh review text described the warning from startup; its actual
pixels had already become a teal desktop. Preserve that review-description
error rather than treating it as a second warning image. Subsequent screenshots
were personally inspected before their corresponding input.
The final reviewed pre-T PNG is review/webgl/living-before-t.png,
06:49:17.499Z, SHA256
0060aa88db509cf5e3dafa065a4836a26127acda97748775e81d392484894ec8.
It shows living mounted gameplay, not takeoff. Its exact owning CPU/renderer
bindings and complete capture metadata are retained. It does not imply physical
GPU qualification, FPS, worldZ or sustained flight.
sole-t requests 1900 ms, retaining the original-start/release contract. After
fresh-capture admission, selected-byte/private-owner verification and a read-only
fresh actor/code sample, the caller fails at 06:49:29.396Z, before its bridge
probe, release timer or down dispatch:
assert(live.base === base && live.back === live.object &&
live.vtable === base + 0x150b20 && live.endpoint === identity.endpoint &&
live.slot === 0 && live.process === identity.processId &&
live.epoch > 0 && live.root > 0 && live.token > 0 && live.thread > 0);
The live tuple was not saved before that assertion. Which individual predicate failed is UNKNOWN. No post-terminal guest read or capture was made to replace that missing evidence. The exact failing stack, command/admission, absence of all post-gate action artifacts and independent no-input shutdown are retained. The synthetic positive tuple did not qualify this real mounted actor. This failure does not authorize another T or another original. Next offline work is to make each refusal preserve its actual operand values and qualify that caller against fresh original actor identity; do not weaken the gate or guess a pointer.
Recovery, custody and runtime release
Chrome receives ordinary Browser.close ACK and exits 0, signal=null, with no cleanup signals, in 342 ms. Both browser streams drain before profile removal. Kernel custody is complete=true: all 367 owned PID/start identities have 367 actual terminal reaps. Remaining, unreaped, unresolved ptrace/resume and lineage sets are empty. The outer custodian wait is 1, preserving the caller failure; it is an actual wait with separate stdout/stderr EOF, not a claimed successful game command. Ports 44781/39931 are closed, profile and writer lock absent, no own writable FDs remain, and the runtime lock is released. Absence supplements actual waits. CLONE_UNTRACED and nonleader-exec coverage limits remain explicit.
All 398 regular remote artifacts (11,552,279 original bytes) were recovered
through 248 new unique gzip transports, 459,145 bytes, plus 87 existing local
references. Original/compressed SHA and size plus independent Python/GNU gzip
CRC/decode checks passed before the delegated normal-archive request. There
were no remote symlinks or special files. A later all-reference reread also
passes, including 181 exact successful complete HTTP body pins. Logs are gzip
from first stored byte; source/module/media providers are references, not
archives. Local commands retain actual waits and separate stream EOF. Local
lifecycle roles and supervisor also have actual waits. Free /home exceeded
1.5 GB at sampled checks; no RAM gate was applied.
The observable independent immutable 25-minute watchdog was prearmed before the
single UUID-idempotent finite-TTL create. UUID
1cf87c53-ac3c-48ff-a96b-a4ae4e82fe05; box bx_yuspm6ar;
initial expiry 2026-11-09T07:04:01.762Z; watchdog deadline
07:04:02.230Z. Delegated normal stop stop_0c5ed0ef9ded78e9 completed
06:50:46.557Z, forced=false, error=null. Final state ARCHIVED,
archiveAfter=null; the API does not return a separate expiresAt field. The
watchdog did not fire. Prompt ARCHIVED PHYSICAL_DRAINED release was posted
at 06:51:17Z, allowing Quake2 to take the next runtime. No box was resumed
or deleted. Single-task STOP after explicit own-note commit and immediate
authenticated main pull/push.