Warcraft III native right mouse — 20 November 2026
Standard recipe original, 2026-11-20 15:28 UTC
Run 20261120-warcraft3-standard-recipe-original-prologue applies the exact
hash-bound next recipe, removing the added profile-argv predicate. Chrome151
rewrote proc cmdline into one space-separated string containing the correct
profile flag; Puppeteer's spawnargs retained the flag as an individual argument.
The requested temp/cache and physical profile realpath were recorded without
adding an input-authority gate. The actual original launched, the software
producer advanced, and both actual workers matched module10bdae. The whole
private800x600 consent frame was personally reviewed; trusted consent DOWN/UP
completed, followed by native owner binding and a reviewed fullscreen main menu.
The first Single Player click at page(665,198) was published15:29:33.084, applied15:29:33.092 and remained DELIVERY_PENDING until independent trusted UP at15:29:33.176. Native status then returned UNKNOWN: physical UP preceded native USER return. No native consumption token or removal was produced. This is a real input-consumption refusal, not a setup assertion or a500ms timeout. It does not establish why the guest failed to consume the click. The scope closed15:29:33.179 after91 ordinary requests, with zero later requests or replay. Profile, Prologue, Thrall movement/objective and normal Quit remain unproved; no gameplay credit or speculative runtime fix is claimed.
Eight schema checks, real helper READY/cancel/joins/EOF and31 focused checks pass. The existing250/500ms and private one-use guards are unchanged; historical native controls are referenced by hash without calibration. All22 supervised PIDs are nonlive, supervisor leaked=false, browser/display/helpers have actual joins and the transient profile was removed. Runtime minimum home free space was1,618,276,352 bytes. No box was created: IDnull/NOT_CREATED, expiry and stop receipt null. Exact source diffs, named consent/menu/fence images, media/proof references and final build/publication status are retained in the run, with result.json written last.
Follow-up: 20261120-warcraft3-restored-floor-mouse2-player-progress.
This separately authorized fresh scope reached actual native owner binding on
an independently checked whole private 800×600 display. CPU and render Worker
bytes each matched the retained 10bdae module once; actual imports were
retained. Physical mouse, key and host release subprocesses passed exact launch
path READY/cancellation before DOWN/actual exit and both EOF checks against a
local transport fixture, issuing zero input commands. This is helper lifecycle
evidence, not native consumption. The real reviewed host consent click and
independent trusted host UP also passed before native creation and binding.
The first manual guest command failed at 13:03:17.074 UTC before proof creation,
release arming, prepare or DOWN: it supplied intent.observation, whereas the
private authority requires intent.review. This authoring error permanently
closed the scope after 26 ordinary requests, with zero later requests or replay.
The delayed-capture/slow-ACK command never reached those controls. No new guest
Mouse2, left or key consumption or independent guest UP is claimed. All remaining
production negatives and original registered profile/Prologue/Continue/Thrall
movement/objective/Quit remain unrun. Previous genuine Mouse2 proof is referenced
by exact SHA only. No Warcraft game progress was achieved.
Offline repair adds manualIntent to the owned W3 controller: it requires the
actual personal review field, validates the exact edge and current source, and
creates fresh nonce/source identity before proof creation. The next-driver
substitution is retained as a hash-bound recipe; the executed driver is retained
unchanged and is permanently closed. The owned independent release helper now
handles exact key edges with their DOM code through the pinned Git connection
loader, avoiding the held GL controller in the child. This key route has lifecycle
and JS coverage only. Thirty-two focused checks pass; see the run's checks for
the final full-build result. Required production and affected-original
qualification is incomplete, so these repairs remain released uncommitted WIP.
Browser, mouse/key/host release helpers and display have actual join receipts;
all 21 supervisor-tracked processes are nonlive and the browser profile is gone.
Runtime minimum /home free bytes was 1,622,253,568. No box was created: ID,
expiry and stop receipt null, state NOT_CREATED. Root shutdown delegates only
owned cleanup. No foreign held source was changed.
Follow-up: 20261120-warcraft3-proven-mouse2-manual-player-progress.
Runtime admission closed during preflight on a genuine current /home reading
of 1,205,202,944 bytes, before any browser, physical helper or endpoint request.
The independent shared report at 12:53:34.038 UTC measured 1,180,213,248 bytes.
No later free-space recovery reopens this scope. Original Warcraft III,
physical-helper READY/cancel preflights and remaining production input
controls were not run. Thrall movement, objectives and normal Quit remain
unproved. The prior Mouse2 native consumption and trusted UP are reused by
exact evidence hashes, without claiming a new production qualification.
The run-local floor watcher now attributes each byte count to its nearest
measurement timestamp, falling back to its board-line timestamp only when no
separate measurement time is supplied. The actual MCM completion quoting
12:37:28 remains historical for a 12:39:29 launch; a genuine new low sample in
a completion still closes admission. Explicit measuredAt and
actualHomeFreeBytes shared reports are recognized. Unknown timestamps,
truncated boards and incomplete new low reports fail closed. Its derived
driver is syntax checked, integrates the watcher, checks actual 800×600 root
dimensions before Chrome, and refuses startup in this permanently closed run.
It was not executed. Thirty focused checks and the literal guarded full build
pass. Later free-space recovery did not reopen the scope. Production qualification
remains a precise disk-floor blocker, so the released native repair stays WIP.
Latest run: 20261120-warcraft3-rightmouse-final-fence-delivery-progress.
Uncommitted WIP now separates genuine socket publication from dispatch ACK in
the opt-in mouse controller. sendPublished signals only after the actual
WebSocket send returns; before that it completes the whole-image socket
capture, synchronous evidence retention, private intent validation and release
fence. Status waits for that signal. ACK and native status must both finish
inside the original 500 ms wall/monotonic window; every started task is joined,
primary failures survive, and no observer starts after closure. Default key
and left controllers were not edited. Full pre-DOWN capture/retention still
finishes before native preparation. No capture or owner check was removed.
The display was actually created at 800×600 before browser/fixture launch;
the guest client remained 480×360. A newly reviewed entire fullscreen image and
current client map selected page (400,320) → client (240,192). Actual CPU
and render Workers independently matched the retained 10bdae module; the
owning native USER GetMessage return consumed exact WM_RBUTTONDOWN 0x204,
wParam 2, lParam 12583152, token 1, HWND 65537, thread 1/slot 0. Independent
trusted right UP matched nonce and owner and joined. No new WAT was needed.
Measured same-process monotonic phases: pre-DOWN capture/retention 234.6 ms, all before prepare; preparation request/return 28.6 ms; final complete socket capture/retention 155.7 ms; socket publication 216.5 ms after prepare start; first status reply 249.6 ms, before ACK at 251.9 ms; CONSUMED status 271.4 ms. The ACK is transport evidence only. This positive production run establishes real right consumption plus UP, not all requested delayed-capture/slow-ACK production controls. Twenty-five JS tests pass, including slow ACK with a real transport signal, omitted publication, capture/send error, late status, monotonic expiry, nonconsumer, UP races and existing focus/right/left/key controls. These JS controls do not establish production compatibility.
The first blocker was the inherited run-local messageboard parser at
12:43:30.594 UTC. It treated MCM's new completion line quoting its older
12:37:28 actualhome1250816000 sample as an in-scope new floor observation.
Own actual runtime minimum was 1,610,350,592 bytes. This is a false admission
closure caused by historical-text parsing, not proof of a new measured floor.
The scope nevertheless stayed permanently closed: 33 ordinary requests at
closure, zero later requests, no replay. A subsequently written capture
command was never read or executed. Production left/key, remaining negatives,
old-fail/new-pass slow-ACK qualification, and registered Warcraft III profile,
Prologue/Continue, Thrall movement/objective/Quit were not run. No game
progress is claimed; no commit/pull/push because required qualification remains
incomplete. A final offline change adds the same monotonic limit to the
pre-send callback; it is separately recoverable and has JS/build validation,
but no browser rerun after closure. Browser/display/helpers have actual joins;
supervisor reports no leaks. Box null/NOT_CREATED, expiry/stop receipt null;
root shutdown delegates owned cleanup.
Latest follow-up: 20261120-warcraft3-rightmouse-host-focus-native-progress.
The released WIP remains uncommitted. The qualified 5465c599d host flow now
precedes session creation as well as observation/binding. The actual whole
private display was personally reviewed, the consent button received one
trusted click with independent trusted UP, and one document read confirmed
focus, visibility and fullscreen before native creation. The real owner then
bound successfully. No page focus override or skipped guard was used.
The actual authored production CPU and render Workers each matched 10bdae
once; their identical bytes are retained once. All 305 actual import entries
were recovered. The pinned graph, complete served manifest, local import
closure and 12 original-media references passed preflight. The inherited
unchanged bridge/release controls were reused by exact SHA. Thirteen mouse
tests plus nine qualified focus-order tests pass, including actual bootstrap
call ordering and first-refusal same-read scalar retention. A new full literal
flock /tmp/wine-build.lock bash tools/build.sh passed with actual exit 0.
After a personally reviewed whole fullscreen frame, one exact DOM2/VK2 DOWN
was published at page coordinates (512,450). The independent process
confirmed trusted right UP, matching nonce and owner, at 12:26:52.689 UTC and
joined successfully. This does not prove a native WM_RBUTTONDOWN return:
the fixed native 500 ms delivery deadline closed at 12:26:52.646 before any
native status reply was recorded. The final socket capture and evidence work
consumed much of that interval; publication occurred at 12:26:52.607. The
evidence does not distinguish a delayed transport ACK from delayed native
consumption. No later query was made to diagnose it. Focus is established at
bootstrap, not guessed to be the cause of this new deadline refusal.
The scope permanently closed with 29 ordinary requests, zero later ordinary requests and no replay. The already armed independent UP completed after closure as required. Actual native left/key compatibility and native negative controls remain unqualified; the passing JS controls are not substituted for them. Warcraft III, its profile/Prologue/Continue, Thrall displacement, objective/combat and Quit were not attempted. No commit/pull/push occurred because required native/original qualification did not complete.
All 21 supervisor-tracked processes are nonlive; browser, release helpers,
window manager and private display have actual wait/EOF receipts. The browser
profile was removed. Runtime free-space samples stayed above 1.5 GB (minimum
1,670,115,328 bytes); an offline post-closure read at 12:27 found 1,219,964,928
bytes and permanently closed admission too. No historical floor marker was
reused as a new runtime observation. Box ID/expiry/stop receipt are null,
NOT_CREATED; root shutdown delegates completed owned cleanup.
The following records describe the earlier, separate failed scope.
Work in progress, uncommitted. Original Warcraft III was not launched.
The only browser qualification closed at its first native session observation
with NATIVE_MOUSE_UNKNOWN actual import/owner/focus unavailable, before bind,
review capture, release-participant creation or any guest DOWN. There were 19
ordinary requests through that failure, zero later endpoint requests and no
replay. The combined refusal does not establish which prerequisite failed.
The final offline source adds diagnostic fields without weakening the gate;
it has not been rerun in a browser.
Admission deviation: the shared messageboard's 12:07:55 worker closeout reports an actual free-space sample of 1,499,365,376 bytes, below the 1,500,000,000-byte floor. This task failed to drain that watcher update before its 12:09:11 browser supervisor launch. Its own runtime samples stayed above the floor (minimum 1,616,154,624), but the shared report should have closed admission. Thus 19 ordinary requests occurred after that shared floor report. Zero requests occurred after this task's first UNKNOWN at 12:09:17.912. Do not treat this run as satisfying the required zero-after-floor admission policy.
The opt-in bridge extension accepts DOM button 2 / VK_RBUTTON 2, expects
WM_RBUTTONDOWN (0x204), exact native wParam=2, and derives client
coordinates from the reviewed viewport point through the real renderer mapping.
The ticket retains its actor, session token, native owner and lifetime fences.
Native MSG coordinates are rechecked against that derived point, including
after dequeue. Nonremoving Peek proof remains separate from later removal.
Right UP retires pending DOWN and prevents a cached DOWN returning after UP.
The existing default key and left-button routes retain their contracts.
tools/native-mouse-consumption-input.cjs composes the existing private review
authority with this exact mouse edge. Source/review/fresh/native/hold limits
remain 30 seconds / 15 seconds / 5 seconds / 500 ms / 1500 ms. Whole-frame
equivalence is the default capability; the authored qualification explicitly
selected the existing manual-context capability. One-use proofs, immutable
identity, synchronous final socket fence and complete-frame evidence sink
remain required. No successful review or native gesture occurred in this run.
The new independent helper is tools/native-mouse-release.cjs, announced and
claimed before editing. It owns only trusted physical UP delivery on a separate
process/connection, with exact target/PID-start-token/owner/button/nonce
matching, cancellation before DOWN, bounded release and actual child joins.
It reuses the released child lifecycle helper rather than editing Hexen's
controller or release helper. Its real trusted right UP remains unqualified.
Eleven focused JS tests pass, including the actual released 7bca bridge's right-intent refusal and the new guard's acceptance, exact coordinates/button, mutable-edge rejection, native-message matching, cached/removing receipts, left/key defaults, owner/focus/modal/capture/retirement/viewport fences, no-consumer/stale/deadline, replay/cancel/UP races, and final publication order. Existing Hexen left-button/manual-context and Quake II capture JS regressions also pass. These are guard/controller tests, not trusted DOM or native guest execution proof. Full literal guarded builds pass; no WAT, compiler, GL, producer or held-region source was edited. No optimization variant was made.
The sole authored production browser used the coherent existing 34c/72dc/cf181
graph and retained 10bdae module with the owned input/bootstrap overlay.
Actual CPU and render Worker bytes each hash once to
10bdae2c56c79eeb5d0b4ebbc602c8a233e36351380da274a70f7f5e25363564
(2,418,815 bytes). Actual import tables were recovered before the refusal.
This establishes module identity and producer readiness, not right-button
consumption or a real legacy-endpoint transition. No complete image was
captured before failure; no visual review, Thrall selection, displacement,
objective/combat consequence or ordinary Quit is claimed.
Evidence: scratch/runs/20261120-warcraft3-native-rightmouse-player-progress/.
Both the exact executed overlay and final WIP are recoverable from Git plus
patches. Registered Warcraft media and the retained module are referenced with
hashes, not copied. The native supervisor reports no leaked processes; Chrome,
display and window-manager close/EOF receipts are retained, all tracked PIDs
were verified nonlive, and the transient profile was removed. Local box ID,
expiry and stop receipt are null (NOT_CREATED); owned cleanup is complete.
Native right consumption plus trusted UP and all production negative controls must qualify before a newly authorized ordinary original run. This failed scope is permanently closed. The feature stays uncommitted WIP because the required native and affected-original qualification did not complete.
Schema-correct production run, 2026-11-20 13:12 UTC
Run 20261120-warcraft3-schema-correct-native-player-progress replaces the
stale driver constructor with the exact C.manualIntent(command, sequence, latest.source) call. Eight offline menu/game right/left/Enter/Escape command
and physical-UP variants passed the actual production bridge preparation
schema before browser startup. The adapter's GL support dependency now comes
from the pinned Git loader, avoiding another worker's mutable helper source.
The actual private root was verified as 800x600 before Chrome. CPU and render
Workers each hashed once to the retained 10bdae module; all 305 imports were
observed. A personally reviewed host consent click and independent trusted UP
preceded native owner binding. A personally reviewed right click at (400,320),
with an added 40 ms capture delay and 120 ms ACK delay, returned native
WM_RBUTTONDOWN (0x204), wParam 2, token 1, client (240,192). Socket publication
at 13:12:48.719 preceded native consumption at .739 and observation at .781;
ACK completed at .883. The independent child recorded trusted right UP at
.807. This is new production evidence, not reuse of an earlier receipt.
The next personally reviewed left control hit the original 500 ms deadline at its final full-frame socket fence. Preparation started 13:13:08.822, the final capture began .09.026 and completed .09.373, and the final publication guard refused before DOWN. The first permanent close was 13:13:09.612. Its armed independent helper cancelled with no physical edge. There was exactly one guest wire DOWN (the successful right control), 41 ordinary requests, and zero requests after closure. Whole-root capture, decoding, focus checks and synchronous evidence retention all contribute to the fence duration; the precise source of the delay is not established. No clock was widened.
Left consumption, key consumption, wrong-target/late/UP-race negatives and registered Warcraft profile/Prologue/Continue/Thrall/objective/Quit remain unqualified. A later offline df observed 1,441,304,576 free bytes; its exact measurement time was not retained. A subsequent timestamped sample above the floor does not reopen this permanently closed scope. All 22 supervised processes are nonlive with no supervisor leak, all physical helper/browser/ display joins have exit and both EOF receipts, and profile/tmpfs are absent. Box ID/state/expiry/stop are null/NOT_CREATED/null/null. All 32 focused checks and the full flock build pass. Preserve uncommitted WIP; do not publish the feature before the remaining production and affected-original checks pass.
Full-frame fence investigation, 2026-11-20 13:19 UTC
Run 20261120-warcraft3-fullframe-fence-cause-native-progress found a concrete
duplicate blocking operation in the run-local admission watcher. Every drain
read the entire 35,053,146-byte messageboard before discarding its already-read
prefix. Twelve drains on an unchanged real board read 420,637,752 bytes and
took 13.5–22.4 ms each. The replacement reads only bytes after its retained
offset, with short-read completion, truncation detection, unchanged partial
report parsing and permanent floor closure. The matching twelve unchanged-board
drains read zero bytes and took 0.005–0.253 ms. Historical reports are still
processed at startup; no admission check or input guard was removed.
The fresh actual production native fixture used this replacement and the
unchanged whole-image authority on the complete private 800x600 display.
CPU and render Workers each matched the retained 10bdae module once; actual
305-import observations, reviewed trusted host consent and independent host UP
were retained. A personally reviewed left click at page (400,320) consumed
native WM_LBUTTONDOWN 0x201, wParam 1, token 1, client (240,192), with trusted
independent UP. Preparation-to-observed-consumption was 401.5 ms. Enter then
consumed WM_KEYDOWN 0x100, wParam 13, token 2 and received exact trusted Enter
UP. The earlier right40ms-capture/120ms-ACK proof is referenced by exact SHA;
it was not replayed.
New same-process phases separate checkpoints, page queue waits, xwd and focus subprocesses, XWD-to-PNG encoding, PNG decoding, retention, final capture, publication and ACK. The successful left final fence took 290.8 ms, including 60.8 ms xwd, 71.2 ms encoding, 53.1 ms decoding and 62.1 ms combined focus subprocesses. Its complete-image retention took 0.52 ms. Source30/review15/ fresh5 seconds, wall and monotonic native500 ms and hold1500 ms remain intact. These measurements prove redundant board I/O was removed and the new left path passed. They do not isolate the cause of every delay in the previous run or establish constant capture cost under different machine load.
The supervisor received SIGTERM at 13:19:59.236 UTC, then SIGKILLed its remaining driver after the three-second grace. Sender and cause are UNKNOWN. The last page response was 13:19:42.476; zero later endpoint requests were made. A later offline-authored close command was never read. The runtime is permanently closed, with no replay or replacement. The later shared GL floor measurement at 13:20:09.630 was after this closure. Nonconsumer/late/wrong-target/UP-race production controls and registered Warcraft profile/Prologue/Continue/Thrall movement/objective/Quit remain unrun. This is a blocker, not game progress.
All 34 focused tests and a fresh literal full flock build passed. The watcher fix and instrumentation are retained as run-local inputs; inherited Mouse2 code remains uncommitted WIP. No commit/pull/push was performed because the remaining production and affected-original qualification did not complete. The supervisor actually joined the driver and reports no leaks; all 22 tracked PIDs are nonlive. Display and host helper have individual wait/EOF receipts; browser and mouse/key release individual wait/EOF receipts were not written before the driver died and remain unavailable. The owned profile had no unique downloads and was removed after verification. Box ID/expiry/stop receipt are null, state NOT_CREATED; delegated owned cleanup is complete.
Remaining native controls and original startup, 2026-11-20 13:28–13:34 UTC
Run 20261120-warcraft3-incremental-fence-native-prologue-progress reused the
unchanged incremental board reader, including timestamp attribution, split
reports, positioned short reads and truncation refusal. The prior supervisor
SIGTERM preceded its allocated deadline by 171.993 seconds. The receipts prove
the received signal, subsequent escalation and actual child join; they do not
identify its sender or upstream cause. No cancellation/expiry signal was
suppressed. This run's driver closes its gate on TERM/INT/HUP and then runs its
ordinary finite cleanup and actual joins, with supervisor escalation unchanged.
Fresh complete physical-helper/schema/exact-argv READY, arm/CANCEL before DOWN,
zero-input and actual wait/both-EOF preflights passed before browser startup.
All native fixture endpoints used a private root verified as 800×600 before
Chrome, complete root images and personal reviews. Actual CPU/render modules
each matched 10bdae once per endpoint. Earlier right delayed-capture/slow-ACK,
left and Enter positives are referenced by exact SHA, not recalibrated.
Four remaining production controls passed on separate endpoints. A right DOWN at viewport (410,320), signed for (400,320), maps to client (246,192) rather than (240,192); it never applied or obtained a native token, and trusted independent UP ended the ticket UNKNOWN. The first attempted one-pixel offset was an ineffective setup control because both points mapped to client x=240; its actual consumption is retained and is not counted as a negative pass. The corrected control used a fresh endpoint after normal cleanup.
A real 510 ms dispatch delay after native preparation hit the unchanged 500 ms
deadline before publication, with zero DOWN and independently cancelled UP.
A separate PREPARED-state race delivered trusted independent right UP before
any DOWN: actual native status was UNKNOWN, token zero, with the exact owner
and nonce in physicalUp. Cancellation explicitly closed the actual native
guard after PREPARED and before DOWN; status was UNKNOWN with the cancellation
reason and no token, and its release helper cancelled without input. These are
production guard controls, not a claim of a cached Peek/removal UP-race test.
Each negative endpoint closed permanently with zero later ordinary requests;
physical helpers, browser and display have actual wait/EOF receipts and the
supervisors report no leaks.
The registered warcraft3_demo original was launched but trapped before a
menu at _clearfp. Offline inspection found a run-local serving error:
requests for the existing shared binaries/dlls/msvcrt.dll returned 404, so
the loader fell back to an unimplemented built-in stub. This does not establish
an emulator defect. The server allowlist was repaired to serve its existing
shared DLL/TLB dependencies; their paths and hashes are references, with no
media copies. The qualified host-review helper also had a fixture/Hexen-only
app allowlist; the run-local derivation adds only the registered Warcraft app
identifier and has an offline old-reject/new-pass/other-app-refusal preflight.
Neither repair was exercised by a subsequent original browser.
The initial readiness loop also failed to treat the fatal console event as terminal: it recorded 123 further readiness responses before the operator noticed the fatal and explicitly sent TERM to the exact owned driver. No input authority had been established. A run-local fatal-event closure now stops that loop immediately; actual old-handler/new-handler offline controls pass, but the new handler has no original-browser qualification. Do not claim zero post-fatal requests for the first original attempt. Its zero-after-closure count starts at the later explicit TERM gate closure.
The original endpoint was stopped and joined without reopening it. The fresh
original attempt then closed at its first admission checkpoint on the genuine
shared sample actualHomeFreeBytes=1498824704, measured at
2026-11-20T13:34:07.804Z. This was before its browser/display/helper creation
and made zero endpoint requests. Its generic originalLaunched:true mode
field is incorrect; the separate launch correction records actual browser
creation false. Later disk recovery cannot reopen this scope. Profile,
Prologue/Continue, Thrall selection/destination/movement, objective and ordinary
Quit remain unproved. Mouse2 stays released, uncommitted WIP because affected
original qualification is incomplete. Box ID/expiry/stop receipt are null,
state NOT_CREATED; all owned cleanup is complete.
Fresh 34 focused tests and the literal full flock /tmp/wine-build.lock bash tools/build.sh passed with actual exit 0. The resulting main-tree module is
recorded separately from the retained 10bdae module used by the browsers.
Exact Git/diff/new-file patches, executable driver inverse recipes, original
media/shared-dependency references and native supervisor/helper receipts are
retained in the run; no source-tree snapshot or media copy was made.
Follow-up: 20261120-warcraft3-qualified-mouse2-original-prologue.
The registered original launched with the repaired run-local server. Its actual
HTTP receipt served binaries/dlls/msvcrt.dll with status 200, 278581 bytes and
SHA-256 887eb5ce93edb7192ca3e9220f07f9ca0f94db02af5862ebcbdfcb852db99fd1.
DLL/TLB bytes, original media, frozen graph/module/ABI, eight command schemas,
actual helper argv/READY/cancel/joins and the actual server/fatal callbacks
passed before browser launch. Prior native positive/negative proofs were reused
by exact source and evidence hashes, without repeating calibration. The prior
123 post-fatal readiness responses remain a recorded deviation; the repaired
fatal callback has offline closure coverage, and this launch had no fatal event.
The private display was verified at 800×600 before Chrome. CPU and render
Workers each attested 10bdae once; all 305 imports were retained. The complete
initial host image was personally reviewed: Warcraft III Demo logo below the
host fullscreen consent strip. Trusted host consent DOWN/click and the
independent trusted UP completed, document focus/visibility/fullscreen passed,
and the actual original's native owner bound successfully.
The first manual capture failed at 2026-11-20T13:46:05.253Z, before completing
any guest review capture, private gesture proof, prepare or DOWN. The actual
same-read refusal reports healthy/nonterminal bridge, bound owner, unchanged
scene and slot, focused visible document, but ready:false. Its scene sample
was at 13:46:04.992Z, approximately 260 ms before the context response, compared
with the unchanged 250 ms freshness predicate. This supports investigating
sampling/freshness; it does not isolate every predicate contributing to ready.
The endpoint permanently closed after 46 ordinary requests, with zero later
requests and no replay. Profile, Prologue/Continue, Thrall selection/destination,
movement/objective and native Quit remain unproved. No original progress credit.
All 22 supervisor-tracked processes were nonlive; supervisor leak flag false. Browser, host/mouse/key release helpers and both display processes have actual join and EOF receipts. The browser profile and owned tmpfs directory are gone. Runtime minimum home free bytes was 1518821376. Box ID/expiry/stop receipt null, state NOT_CREATED; delegated owned cleanup complete. Seven run-local admission tests passed; unchanged prior 34 focused/full-build checks are historical reuse. No production code changed or was committed/pulled/pushed; inherited source WIP remains uncommitted because original qualification is incomplete. The held browser-native-input and foreign lanes were read only.
Capture context timing, 2026-11-20
Run 20261120-warcraft3-capture-ready-timing-original-prologue adds narrowly
authorized bridge snapshot readiness diagnostics and permits the mouse adapter's
read-only context() when every structural predicate passes. Input ready(),
prepare(), valid(), the 250 ms dual-clock sample guard, and the native 500 ms
deadline retain their prior implementations. Missing diagnostics retain strict
legacy readiness. No observation timestamp is renewed and no capture prepares
an input ticket.
The registered original's first complete whole-display capture now passes and was personally reviewed as the Single Player main menu. The after-capture sample was 332–341 ms old; only freshness failed, with all fourteen other predicates true. Exact capture phase timings include 24.36 ms XWD acquisition, 106.87 ms encoding and 103.93 ms PNG decoding. Context queue responses took about 4 ms. This demonstrates why requiring input admission after synchronous capture work rejects a truthful observation. The prior original's real failure and this real original pass are retained; a paired deterministic test of the exact previous adapter and current production bridge isolates the overbroad context check. Twenty mouse tests, combined focused checks, eight updated schema checks, physical helper READY/CANCEL/join/EOF preflight, and the full locked build pass. Stale preparation and replaced/unready context/input refusals remain covered. Those deterministic negatives are JavaScript guard tests; no new physical negative-input campaign or gameplay claim is made.
The subsequent menu command was issued 31.822 seconds after its source capture. The operator's delay exceeded the unchanged 30-second source limit, producing the first UNKNOWN at 14:00:52.883Z, before proof, prepare, DOWN or release arming. The runtime permanently closed after 74 requests; no later endpoint request or replay occurred. Profile, Prologue/Continue, Thrall selection/destination, movement/objective and native Quit remain unproved. The attempted screen-to-CDP coordinate conversion was never exercised and has no qualification credit.
The diagnostic wrapper's extra publication fields incorrectly looked up lexical
runningApps through globalThis, reporting zero/false. These fields are
non-authoritative and are not producer-failure evidence. Actual producer
observations, module attestations and capture/queue/readiness traces are retained;
repair that diagnostic lookup before a future fresh scope. No closed endpoint
was queried to repair the trace.
Both actual Workers attest 10bdae once, all 305 imports and real shared CRT
887eb5/HTTP 200 pass. All 22 supervised processes are nonlive, leak flag false;
browser, display and helpers have actual joins and both EOF, profile/tmpfs gone.
The supervisor records runtime exit 1; the enclosing execution tool returned
143 with sender/cause unknown. Board watcher was explicitly interrupted, exit
130. Minimum runtime home free bytes: 1542447104. Box null/NOT_CREATED, expiry
and stop receipt null, delegated actual owned cleanup complete.
The tested capture fix remains uncommitted with inherited native-pointer WIP; the required original progression qualification stopped at a genuine UNKNOWN. No commit/pull/push, workers, Python, Wine, deployment or foreign-source changes.
Live preparation observation, 2026-11-20 14:28 UTC
Run 20261120-warcraft3-live-observation-prepare-prologue identifies two linked
adapter/driver defects in the previous fresh-menu attempt. The initial
prepareEquivalent check admitted only a fresh sample, so a structurally valid
but expired first observation permanently closed the session before the bounded
wait could see another producer observation. The run-local Runtime.evaluate
wrapper nested the async result in {value} without awaiting it. CDP serialized
the Promise as {}, hiding its rejection; subsequent calls retried the closed
session and retained its first diagnostic snapshot. Those twenty identical
diagnostics are not evidence that the native producer remained frozen.
The adapter now checks structural identity before and after proof verification,
then samples the actual live bridge on each attempt through the unchanged
observeAndPrepareNativeIntent. Only a new owning native observation can admit
preparation. The controller supplies an awaited diagnostic expression, used by
the derived run-local driver, so asynchronous refusal reaches the permanent
scope gate immediately. Intent, bound identity, one-use proofs, source/review
ages, default input, 250ms native freshness, prepare/send/500ms delivery and
independent release guards remain unchanged. No producer timestamps, repaint,
guest execution or clocks are manufactured by the fix.
Twenty-seven focused checks pass, including new native-message-wait samples,
frozen/stale/replaced/late samples, immutable intent, release and default input.
The exact prior adapter SHA bf0b963a9bc89a786102c7093733e51bf64ed0a946aa1b09dbfb85586819bd92
fails the two new freshness regressions. A VM execution of the actual prior
diagnostic wrapper demonstrates Promise-to-empty-object serialization; the
awaited wrapper returns PREPARED and exposes async rejection. These are unit
results, not a real original-producer old/new qualification.
Actual /home free space fell to 1,499,635,712 bytes during setup, before any
browser or authority existed. This scope is permanently closed with zero
endpoint requests. The locked full build exited 1 at the foreign host.js
lifetime-review census gate; held files were untouched. No code commit or
publication is justified. The fix remains uncommitted WIP. Single Player,
profile, Prologue, Continue, Thrall movement/objective and Quit are unproved in
this run. Prepared command recipes and the derived awaited driver are retained
for a future fresh scope; do not reopen this one. Box ID null, NOT_CREATED,
expiry null, stop receipt null; no owned runtime cleanup was required.
Fresh original live sample, 2026-11-20 14:40 UTC
Run 20261120-warcraft3-live-sample-original-prologue completed the fresh
schema/helper/media/module preflight and all 27 focused checks. The original
ran on the whole private 800x600 display; both Workers attest module 10bdae.
The new main menu was personally reviewed and immediately followed by the
prepared Single Player gesture, without intervening setup or diagnostics.
The retained live reads now demonstrate the adapter fix on the original:
context at 14:42:43.061 saw the native sample from 14:42:42.586, age 475 ms;
preparation at 14:42:43.248 read a genuine newer check_input sample from
14:42:43.212, age 36 ms, with the same owner and scene. Preparation returned;
DOWN was published at 14:42:43.485. No clock or producer timestamp was changed.
The next status read at 14:42:43.519 refused that sample at age 307 ms under
the unchanged 250 ms freshness guard. The awaited wrapper exposed the refusal
and closed the scope after 82 ordinary requests, with zero later requests.
Independent UP physically fired at 14:42:43.569 and its transport acknowledged at 14:42:43.579, but no matching trusted receipt arrived; the helper exited 1. This is a genuine UNKNOWN, not input or progression success. Profile, Prologue, Continue, Thrall selection/movement/objective and ordinary Quit are unproved. No retry or post-closure capture was made. Supervisor cleanup joined the owned browser/display/helpers with no leaked processes. Minimum runtime home free bytes were 1,501,360,128. Box null/NOT_CREATED, expiry and stop receipt null. The inherited fix remains WIP; no commit/push or foreign edits. Main build repair remained separately owned, so no new full-build success is claimed.
Post-DOWN observation cause, 2026-11-20
Run 20261120-warcraft3-post-down-live-observation-cause identifies the exact
failed predicate as fresh: all fourteen structural predicates were true,
while the unchanged native scene sample was307ms old. Preparation read age36,
then the final synchronous capture occupied216ms before socket publication.
The physical UP fired and received a transport ACK; no renderer receipt exists.
A bridge control reproduces the retained307ms timeline: stale DOWN is refused,
then closing the ticket as UNKNOWN leaves UP without an applied renderer edge
and therefore without a receipt. The retained original lacks direct appliedAt
evidence, so this application/refusal explanation remains an inference rather
than a retrospective claim that DOWN was consumed.
The owned Warcraft controller now supplies the native authority with pixels decoded directly from each new whole-root XWD. The exact PNG is still encoded, retained and hashed, and every source/proof/pre-DOWN/socket comparison remains. The expiring path no longer decodes that PNG back to the same pixels. The shared bridge/host are unchanged;250ms/500ms, timestamps, identity checks, release authority, synchronous evidence retention and one-use handles remain. The run-local authority wrapper uses this codec. Neither current original scopes nor historical proof scopes are reopened.
31 focused tests pass, including both byte orders, padded rows, all-pixel equality against the original encoder/PNG decoder, new/replaced frames, malformed headers/masks/bounds, one-use pixel buffers, and the real authority through final socket fence. A fault control makes the old redundant PNG decode fail while the corrected authority succeeds without invoking that decoder; this is an offline observation test, not native gameplay qualification. The retained whole800x600 image, synthesized as XWD only in memory, produced six alternating codec pairs: old median47.48ms, new23.90ms, every RGBA byte equal. This timing does not prove a future original will stay below250ms. Existing frozen/stale/replaced/late sample and release controls remain covered.
The mandatory full flock build passed. Before any browser, authority or runtime endpoint existed, actual home free space fell to1498783744 bytes. That native scope was permanently closed with zero requests; later capacity recovery did not reopen it. Original Single Player/Prologue/Thrall/objective/Quit validation remains blocked. No new box/runtime was created: ID null, NOT_CREATED, expiry null, stop receipt null. Exact module10bdae/CRT887eb5/media/proof references were SHA-verified without copying. Publication receipts and exact source patches are retained with the run.
Decode-fixed original and launch environment, 2026-11-20
Run 20261120-warcraft3-decode-fixed-original-prologue executed the affected
original with committed 84ce0e7b1, module10bdae and real CRT887eb5. The software
producer advanced and both actual worker modules matched the retained module.
The retained whole800x600 host frame shows Warcraft's menu within the browser
and the fullscreen-consent strip. Host consent and game input were not sent;
profile/Prologue/Thrall/objective/Quit remain unproved.
The physical environment check could not verify the required TMPDIR/cache:
Chrome's /proc environment exposed neither variable. Independently, the
qualified transient helper's launch-environment construction replaces TMPDIR
with its own tmpfs child. This is a task-environment failure, not evidence of a
game or decoder failure. The driver was signalled to close before consent;
45 ordinary requests preceded closure, with no subsequent endpoint requests
and no replay. Finite supervision joined all owned children without leaks.
The run-local driver now reapplies and checks the exact requested environment before browser launch, retaining its original executed source as a patch. A local Node child verifies the corrected paths; the negative child confirms the old TMPDIR override. That negative child also used the old TMPDIR and thus did not meet the task's exact-environment requirement. These offline controls do not establish a corrected Chrome environment or native gameplay. The closed original is not relaunched. No shared bridge, host, timing guard or game state was changed. Box ID null, NOT_CREATED; expiry and stop receipt null.
Corrected local launch recipe, 2026-11-20
Run 20261120-warcraft3-corrected-local-env-original-prologue seals the
previous run's staged evidence after actual home headroom recovered. Its own
actual tool Node reports the requested TMPDIR and compile-cache directory.
The local launch recipe reapplies TMPDIR, TMP, TEMP and NODE_COMPILE_CACHE
after the qualified transient helper's environment construction. A fresh
Node child confirms the corrected effective temp/cache paths. The old helper's
TMPDIR override is demonstrated in an ordinary JavaScript object only; no
child was launched with the wrong environment. Module10bdae, CRT887eb5,
original media and nativeMouse2 control proofs were SHA-verified and referenced
without copies or repeating calibration. Eight schemas, real physical-helper
argv/READY/cancel-before-DOWN/joins/EOF, all31 focused tests and the full locked
build pass. The helper transport fixture dispatches zero inputs.
The new browser scope stopped at an added operator profile-argument assertion
before navigation, Warcraft launch, consent or input, with zero ordinary
endpoint requests and zero later requests. The assertion inspected
Puppeteer's reported PID's proc cmdline for the expected user-data-dir flag;
it failed before retaining that argv. Consequently this is an operator-added
launch-observation blocker, not proof of an incorrect Chrome profile,
environment, module or game failure. The exact reason the flag was absent
remains unknown. Mode-only originalLaunched:true fields are explicitly
corrected by the audit: a browser was created, but the original game was not.
No actual worker/module or engine attestation is claimed for this scope.
The executed driver remains unchanged and permanently closed. A hash-bound next-driver substitution retains path observations first and removes the extra argv assertion; it was not executed. Temp/profile observations are operational evidence, not new native input-authority predicates. No shared bridge/host/Hex/GL/WAT change, guard widening, state write, replay or later endpoint request occurred. The original250ms freshness and500ms delivery guards remain. Profile, Prologue, Thrall movement/objective and normal Quit remain unproved. The launch-environment correction has local-child validation only; corrected original-browser/gameplay qualification remains outstanding.
The whole private root was verified800x600 before Chrome. All13 tracked PIDs are nonlive; browser/display have actual join and EOF receipts, supervisor leaked=false, and the owned transient profile was cleaned. Runtime home free space stayed above the floor (minimum1,640,960,000 bytes). No box was created: ID null, NOT_CREATED, expiry null, stopReceipt null. Exact Git/diffs, run-local new-file patches, prior-proof/media references and next-driver recipe are retained in the run; result.json is published last.
Deliberate 300 ms first press, 2026-11-20
Run 20261120-warcraft3-deliberate-press-original-prologue reused the released
standard recipe, changing only the requested local temp/cache paths, fresh
scope timestamp and personally chosen first press duration. All31 focused
tests, eight schemas and physical helper READY/cancel/join/EOF checks passed;
36 module/media/control references matched their hashes. No production code,
shared bridge, native guard or game state was changed, and no calibration was
repeated. No full build was needed for this run-local/documentation task.
The actual original reached the personally reviewed whole800x600 main menu after trusted host consent DOWN/UP. Both actual worker modules matched 10bdae; the loaded CRT matched887eb5. The first Single Player press at665,198 used300ms within the existing1500ms hold and500ms native delivery bounds. Independent UP was prearmed at15:36:05.094Z. DOWN was published at05.239Z, logged on wire at05.240Z and applied at05.247Z. Three status results remained DELIVERY_PENDING without native USER consumption. At05.283Z the status guard refused a258ms-old check_input observation (monotonic257.425ms); fresh=false was the only false readiness predicate. All14 structural predicates, owner, focus, modal, capture and viewport remained valid. This establishes an expired observation, not a structural identity change or a proved game failure.
The scope permanently closed at05.286Z, before its original500ms deadline (05.597Z) and before trusted independent UP at05.541Z. The release fired at 05.538Z and acknowledged at05.544Z. Thus this run's terminal cause differs from the previous90ms release-before-USER failure; increasing the intended press duration did not establish native consumption. No status relaxation, timeout widening, replay or later ordinary endpoint request was attempted. There were69 ordinary requests and zero after closure. Profile, Prologue, Thrall movement/objective and normal Quit remain unproved.
Finite native supervision joined the browser/display/helpers with EOF and reported leaked=false; all21 tracked PIDs were nonlive. The physical profile, private display socket/lock/auth and owned empty temp child were removed. Runtime minimum home free space was1,584,328,704 bytes. Box ID null, NOT_CREATED, expiry null, stopReceipt null. Exact Git/diffs, owned run-local patches, four selected host/menu/pre-DOWN/fence images and media references are retained; result.json is written last. The shared checkout already had MERGE_HEAD041a7830 with HEADc3fa9a44e before these documentation edits. The coordinator cleared the merge bookkeeping and released its Git hold at 15:38:06Z without changing held sources. This worker leaves foreign files untouched; its own explicit documentation commit/publication receipts are retained with the run.