Blood II original enemy-consequence attempt, 2026-11-18
The enemy-consequence goal remains unproved. The single fresh original
process had no renderer because this task's immutable source preparation
omitted the external pngjs package. Its first PNG request was dropped at
batch 5330 with no renderer. The controller permanently closed UNKNOWN
on the missing PNG. No original input was submitted, no original screenshot
was produced, and no later endpoint request or replay occurred.
This is a concrete preparation/capture blocker, not evidence of an enemy
reaction, damage, defeat, meaningful movement/fire, ordinary native Quit,
or an emulator crash. The earlier accepted movement and ammo 50→48 evidence
in blood2-demo.md remains separate and does not prove enemy damage.
Inputs and controls
The owner-designated local approximately 58 MB DEMO V1.0A BUILD109 was
unchanged. All 98 file sizes/SHA256 values and the manifest pin matched
before dispatch and after cleanup. The original October inventory is absent
locally; the retained identical November inventory in
20261111-blood2-current-targeted-enemy-objective-1000/media-pin-before.json
supplied the complete comparison. Client.exe remains
1fea9e1400b948153341bd38f8b66beba885d45d21d202cab3c39bcc173b53b2;
the manifest remains
a46bb10441eeb182ff5a28effd8112f507d6a510a25fba48d06870b8fea19519.
No compilation archive, media copy, configuration change, guest-state edit,
cheat, level selection argument, health edit or ammo edit was used.
The new app-specific tools/blood2-reviewed-input.cjs was published as
70064d010. Its authored native tests were written before implementation.
They exercise actual forward/turn/Ctrl USER return, independent physical UP,
deliberate reply loss after native consumption, sticky UNKNOWN and zero later
endpoint calls. Both native invocations passed; their native and release
children were actually waited with separate stdout/stderr EOFs. The canonical
locked build passed before the code commit, followed by strict pull/push.
The selected authored whole-frame control image was personally inspected;
it is a synthetic window and provides no Blood II visual evidence.
The unchanged controller contracts use 500 ms confirmation, 1750 ms physical UP, 3000 ms observation/completion, 5-second game and 30-second menu freshness, one finite key gesture, independent prearmed release, permanent owner identity, whole-frame review, and a reserved ordinary Quit route. The original never bound an input owner. The synthetic controls used the production native path without the original's frozen source inventory, so they did not validate its external codec closure. That preparation gap caused this attempt's blocker.
One original and terminal cleanup
Dispatch HEAD was 0233be5360309cd51578bac6d23da146cb2ba38c on main;
module SHA256 was
7d347a6a631ab77335a14935a533feb896be71aa2508fc5aa19b1e5df8f096ce.
The actual native module read was observed and matched. Source hashes, relevant
binary diffs and recovery patches record the exact dirty shared tree, including
foreign WIP; none of that WIP was edited or adopted into this task's commits.
Source/module descriptors were made immutable before the runtime clock.
The local original PID 2111157 ran 00:56:28–00:56:47 UTC with registered
blood2_demo, unchanged registered arguments/media, real ticks, threads,
default supported software D3DIM, and a 1024×712 whole desktop. Its prearmed
bound was 600 seconds with a 180-second Quit reserve. The first capture failed
before any review or DOWN. Owned cancellation then sent TERM; the native
actual wait returned 0 with both EOFs. This is orderly host cancellation,
not ordinary game Quit. Held-critical-section diagnostics appeared during
cancellation; no trap or post-exit crash was observed and no speculative API
repair was made.
An earlier supervisory preparation refusal occurred before creating an
original PID or starting its runtime clock: the source preload was applied
to the wrapper and rejected its mutable prepared.json. That refusal was
retained separately; applying the preload only to the native child allowed
the one original above to start. There was no second original attempt.
native-supervise.cjs actually waited its owner, exit 0, leaked:false.
The independent finite watchdog actually exited on the native wait without
firing; minimum observed free /home was 2,453,798,912 bytes. Native, wrapper
and watchdog PID/start identities and the listener were checked absent after
cleanup. No box was created: ID null, state NOT_CREATED, expiry null, stop
receipt null. No workers, Wine, deployment or foreign process shutdown was used.
Evidence is scratch/runs/20261118-blood2-original-reviewed-enemy-consequence/:
result.json is published last; media comparisons, source/module identities,
recovery patches, actual wait/EOF receipts, terminal closure and cleanup are
retained there. No original before/after PNG exists to select. Only one named
authored control PNG is retained; redundant fixture images were removed.
This runtime scope is closed, ownership is released, and work stops here.