Q2 received-frame UNKNOWN causal boundary, 1240
No justified code repair. OFFLINE diagnosis narrows actual1205 to the motion sample's consistency/validation branch, but the retained result cannot identify which predicate rejected it. Production behavior and health policy are unchanged.
Evidence: scratch/runs/20261109-quake2-received-frame-read-repair-1240/.
git-base.txt records this task's exact base; relevant-uncommitted.patch is
empty. diagnose.py verifies SHA/size of the original retained health trace,
failed hold, terminal and four numbered replies, without copying them.
retained-boundary.json names their existing paths and hashes and authenticates
the actual1205 providers against its runtime pins. Historical host/Worker files
come from Git 1e0b662f93dab635fa9f17f68e9488c9a6ea3b4f; current files differ
and were not substituted for that lifecycle analysis. No archives/media/module
copies, browser, guest, Box, workers or deployment.
The last valid sample at 11:43:25.783Z has frame263/time26300/HP100; all411 valid health rows retain HP100 and no recorded decreases. The rejected page sample is timestamped .811Z and is retained/terminal at .817Z during-action. Its process/endpoint/Worker/module/renderer/canvas/epoch and memory identities, all12 binding pointers/texts and both attack-latch reads equal the prior valid sample. Top-level attackInput remains SNAPSHOT. Thus the reader's owner, memory, canonical-code, bounds and binding/PTE rejection returns did not fire. This is not evidence of native frame lifetime, all writers, or absence of ABA.
The final hold sample spans the independent release boundary
(whollyDuringHold:false); its page physical ArrowRight is zero, while the
earlier progress observation is high. Those are separate sequential observations,
not an atomic contradiction. Physical release ACK is retained. Four numbered
ACKs precede the failed command; reply4 and post-terminal captures are absent.
No HP after UNKNOWN, damage-cause inference or fifth ACK is supplied.
The actual generated health provider is the1205 health-preload.cjs, SHA
6a2c9ca3ee868c5a4bcddef2cf84e202b6915f090bb11e98282a7efd60f3046a.
It wraps the actual production composition: capture-after, before-action,
during-action and100ms passive progress all use the existing readPage chain.
Positive signed health arms it; first HP loss/death or UNKNOWN throws terminal.
The action check runs before sample persistence. Independent UP remains allowed;
close clears the passive timer. Existing command/hold caps and terminal refusal
were not changed. The generated source hash in consumer-controls.json is a
fresh local composition identity, not the differently staged native1205 hash.
Canonical EXE SHA 2fa987c9390663e230c5b6479285e7c55511e35f80091bfed96f43062c1c810b
and parse-frame-disassembly.txt establish the native lifecycle directly:
0x40333d/342 clears60 dwords at0xa9ce98; 0x4033d3 or403445 sets valid before
playerstate parse4034be and entity parse4034fe; 0x403503-527 copies60 dwords to
0xa9cf8c+(frameNumber&15)240. Native time is frameNumber100. Signed HUD
stats[1] is consumed at0x40d4a5 from0xa9cf42.
The fixed direct addresses,240-byte size,16-slot ring and signed HP offset match
the reader. Current-frame clearing/parsing and ring publication are multiple
guest instructions, not an atomic publication protocol.
Historical host allocates shared memory with initial==maximum, passes the same memory/module/endpoint in Worker init, and retains memory until runtime/render retirement before clearing references. The page reader pins memory+buffer and checks those objects again before return. There is no evidence here supporting a grow/detach, stale address, replacement memory, or dynamic frame pointer fix.
The reader compares two whole current/published240-byte snapshots around its other observations. Accepted motion also requires active state4, valid1, positive frame, matching server time, bounded pmType and current==published. Damage additionally requires two equal3840-byte ring snapshots and a matching decoded current/history row. Since motion is UNKNOWN, ring-only health failure cannot alone explain this terminal. None of motionFirst/motionSecond, their state/slot/current/published bytes, or damageFirst/damageSecond is returned. The retained "raw" reply/hold contains decoded results, not these raw frames. Stable parse-in-progress, between-read mutation, inactive state and malformed frame all map to the same recorded reason. Which occurred in1205 is UNKNOWN.
consumer-controls.cjs executes the actual serialized reader and extracts health
functions from the actual generated1205-preloaded composition. Four distinct
authored counterexamples produce the identical recorded motion/health UNKNOWN
shape and terminal. NormalHP100, damage98, death0, signed death-1, unreadable
memory, owner rebind, and memory rebind controls pass; passive ticker starts zero
new reads when terminal/closing; real input.hold attempts independent UP and
preserves the first failure even if persistence/UP fails. Existing passive-attack
and passive-damage suites and locked build pass. These controls demonstrate
non-identifiability, not reproduction of1205 or an oldFAIL/newPASS repair.
Chrome-launching startup-failure tests were not run.
Next causal experiment, requiring separate authorization for any native run: retain the already-read local motionFirst/motionSecond and damageFirst/ damageSecond values in the existing readPage result on its first rejection, plus the exact predicate booleans (including owner checks). Do this inside the same reader invocation and existing evidence sink: no extra observation layer, new read, retry, delayed acceptance, terminal follow-up read or health waiver. First validate that return-only diagnostic change offline with a read-count control and unchanged acceptance/terminal results. A separately authorized single run would stop on the same first loss/death/UNKNOWN and retrieve those bytes. Offline decoding can then separate stable current/ring mismatch from between-read change or malformed/inactive data. Until that boundary is retained, changing timing, lifetime or selecting an older ring frame is unjustified.
No new Box. Prior bx_cq8zqp74, UUID179a8a60-1e58-4425-9a22-15f4fedc2187,
remains UNCONTACTED; inherited ARCHIVED/PHYSICAL_DRAINED with normal stop
stop_fad5e3232932c556 completed11:44:31.856Z and initial expiry12:02:52.623Z.
Native REVOKED; all-writers/edict ABA UNKNOWN. SingleTaskSTOP.