MCM MENU stable owner projection (uncommitted WIP)
Run: 20261120-mcm-stable-owner-menu-quarry-progress.
The actual registered run 20261120-mcm-atomic-command-registered-quarry
reached GPS PREPARED, then compared its entire page identity across a MENU
capture. slice.seq, EIP, ESP, RPC counts and timing changed as the CPU Worker
executed. That caused a false owner replacement refusal before review or Start.
Its earlier missed shared floor at 12:37:28Z and 206 later requests remain an
admission deviation. That closed scope is never qualified or reopened.
The proposed adapter projects the existing explicit owner fields, guest endpoint/slot and source preparation owner. It excludes execution, presentation and physical-input samples from MENU identity. Current page/Worker reads, module inventory and byte verification, process and endpoint object tokens, renderer guards, session and epoch checks all remain. The MENU context also compares actual native-instance, memory, thread-manager, observer, preparation, canvas, HWND object generation and viewport tokens. A prepared MENU requires its render endpoint to remain null. A replacement or retirement still refuses.
The synchronous image sink retains each identical PNG once and writes separate stage metadata referencing it. It verifies retained bytes again on reuse. Exact whole RGBA equality, private one-use intent, source30/review15/fresh5, direct12 and physical release policy remain unchanged.
Validation: 94 checks passed, including new VM adapter controls executing the production identity expressions through prepare/review/Start with changing samples, and process/native/thread/observer/window/preparation/Worker/endpoint/ module/stopped negatives. The full flock build passed. These VM controls are not executing production Worker MENU proof. A new native production-Worker control and expanded browser negatives were written and syntax checked, but were not successfully executed. The first native invocation failed parsing before any Worker/browser creation; its duplicate declaration was corrected.
Runtime admission permanently closed on the first applicable new shared actual floor, the root measurement at 12:53:34.038Z (1,180,213,248 bytes). The delayed watcher also measured 1,284,403,200 bytes at 12:54:49.087Z. The board cursor had been initialized at task start. Historical completion notes quoting 12:37:28 are distinct from these new actual measurements.
There is a further deviation: the broader focused batch was mistakenly called
offline-only, but test-mcm-synchronous-source.cjs launches a native guest
Worker. It ran after closure and exited successfully with its helper awaiting
link.stop(). Its request count was not logged and is UNKNOWN. This test is
not MENU identity proof or original-game qualification. No registered original
or browser was launched; Start, GPS ARMED/PUBLISHED, quarry, Escape and Quit
are unrun. No runtime admission may reopen on recovered disk space.
The adapter/context/tests/docs remain uncommitted because the required real
executing production Worker identity validation was not completed. No pull or
push was attempted. The atomic channel from 164b44503 is reused unchanged in
the guarded run-local recovery recipe. That recipe was not launched; static
syntax/reconstruction checks are not entrypoint READY or runtime preflight
qualification. Evidence retains exact Git/binary diffs and owned new-file
patches, not source/media snapshots. No box was created: ID null,
NOT_CREATED, expiry null, stop receipt null.
Fresh production control, 13:02–13:05 UTC
Run 20261120-mcm-production-stable-owner-quarry-progress independently
authorized a fresh scope after disk recovery. The corrected native test exited
0. Its actual production guest Worker executed 1000 x86 blocks per identity
sample: the old volatile owner falsely refused, the stable projection passed,
seven owner/module/endpoint/epoch replacement controls refused, and actual
Worker termination refused. This is an executing production Worker with an
authored MENU/owner binding, not registered MCM MENU proof. The Chrome fixture
also passed trusted DOWN/UP, one-use, whole-RGBA, owner replacement, retirement,
source30/review15/fresh5 and unchanged direct12 controls; its process and tmpfs
were absent after cleanup. Prior94 checks and full build are historical reuse
only; the post-closure native-batch deviation above remains disqualified.
Registered MCM launched using the existing f5ae module/frozen runtime and 62d GPS overlay with the current adapter/context. Preflight verified183 source files,305 imports and311 media references, output IO, and current guard hashes without asserting unchanged proof for the modified WIP. Whole private root 800x600 dimensions and a root PNG were verified before Chrome. The sole actual CPU Worker byte recovery matched f5ae; no renderer Worker existed at startup. The recursive local helper syntax/dependency closure was completed after startup, before physical input, rather than before launch. This ordering deviation is retained explicitly; static reconstruction is not runtime proof.
The complete startup and fresh800x600 images were personally viewed. MCM was at its hardware-change OK/Cancel dialog. The attempted OK command hit the unchanged12s direct admission expiry at13:05:11.102Z, before its callback or any input DOWN. The driver permanently closed on this known stale refusal (its generic failure label says FIRST_UNKNOWN). Own first actual floor followed at13:05:14.796Z:1,479,909,376 bytes; GL later reported its slightly earlier 13:05:14.735Z measurement of1,495,015,424 bytes. Both followed closure. No requests may follow either closure and no replay is authorized. PREPARED, postprepare MENU review/Start, ARMED, CPUFlip/sourcecopy/decoder/compositor PUBLISHED, quarry/currentdisplay, Escape and Quit remain unrun. The source clock must not be widened; future work needs a separately authorized fresh scope and prompt review of a fresh capture.
The adapter/context/tests remain uncommitted WIP because the affected original MENU path did not run. No new full build or commit/pull/push is claimed. Evidence retains current exact Git/binary diffs, owned new-file patch and media/module references, with result.json published last. No box was created: IDnull, NOT_CREATED, expiry null, stop receipt null; local supervision owns cleanup.
Preflight before review, 13:10–13:14 UTC
Run 20261120-mcm-preflight-before-review-quarry-progress reused the exact
four adapter/context/test hashes and prior32 qualified native controls. No
unchanged calibration campaign ran. Run-local recovery transformations moved
recursive helper syntax/import preparation, actual operator-command versus
reconstructed-handler checks, module305/import graph183/media311 verification,
exclusive fsynced output IO and actual driver offline-ready READY/CANCEL/join
before browser/original/capture. The READY/CANCEL control initialized the real
driver helpers without opening a browser or sending input; it is not a new
native physical-release control. The prior qualified physical controls remain
SHA reuse only. Startup no longer captures an image that ages during preparation;
the driver awaits an explicit new capture command for personal review.
The actual private800x600 root dimensions and whole PNG were verified before
Chrome. Registered MCM navigation began, but before startupREADY, any game
capture, review or input the independent guard measured1,485,713,408 free bytes
on /home at13:13:14.285Z and permanently closed. The last of13 endpoint requests
was13:13:13.214Z; zero requests followed closure. A separate GL measurement
at13:13:14.046Z (1,498,136,576 bytes) was reported later; no request followed
that earlier measurement either. No actual Worker module byte recovery completed.
The hardware-change OK path, PREPARED, postprepare MENU Start, ARMED/PUBLISHED,
quarry/currentdisplay, Escape and Quit remain unrun in this scope.
During cleanup an asynchronous Worker-attach event hit the closed gate and threw before sending a request. That uncaught assertion prevented individual Chrome/display/board child wait receipts and the normal cleanup.json write. The independent native supervisor actually joined/swept its process group, reported leaked=false, and all46 tracked PIDs were independently absent before owned tmpfs removal. Custodian, supervisor, watchdog and scope watcher waits completed. Missing individual child EOF receipts remain explicit. No tracked code was changed or committed; the inherited WIP remains unqualified on the affected original path. Boxnull NOT_CREATED, expiry/stopnull. No scope reopened.
Fresh prepared dialog scope, 13:23–13:25 UTC
Run 20261120-mcm-fresh-prepared-dialog-quarry-progress completed the recursive
21-helper, eight actual handler schemas, exact driver READY/CANCEL/join,
183-source/305-import/311-media and output IO preflight before browser creation.
The prior32 native controls were reused only after exact four-code/test SHA
verification. Whole private800x600 root dimensions and PNG passed before Chrome;
registered startup READY and the sole CPU Worker f5ae byte recovery completed.
A run-local event guard also prevents asynchronous Worker attach/script events
from requesting endpoints after closure; this is not original MENU qualification.
The complete current hardware dialog was personally viewed. Its ordinary click was admitted at age7810ms within unchanged direct12. The operator used normalized y0.7794, which was incorrect for the displayed OK at page211,403. The resulting whole image was blank teal; CPU EIP, wait and focus remained unchanged. This is an unverified physical effect, not successful ordinaryOK. A later whole capture had identical pixels; no further input or MENU preparation was attempted.
Before the explicit local UNKNOWN record at13:24:57.526Z, the invoker and scope watcher received SIGTERM at13:24:46.890Z, sender/cause UNKNOWN. Custodian and native supervisor joined, with all25 tracked PIDs absent and leaked=false. No endpoint request followed that first closure. Individual browser/display/board EOF receipts and cleanup.json were unavailable after group termination. Unique module evidence uses existing recovery references; owned tmpfs was removed only after checking retained evidence and all tracked processes. No actual floor was observed; watcher minimum1600794624. Scope remains permanently closed, no replay. PREPARED/Start/ARMED/PUBLISHED/quarry/currentdisplay/Escape/Quit remain unrun.
No tracked code changed or new native/full-build campaign ran; previous results are historical reuse only. Inherited WIP remains uncommitted because the affected original path is unqualified. Exact Git diffs/new-file patches and media/module references are retained, result.json last. Boxnull NOT_CREATED, expiry/stopnull.
Reviewed whole-image coordinates, 13:28–13:32 UTC
Run 20261120-mcm-reviewed-dialog-coordinate-quarry-progress fixes the ordinary
click constructor in its retained run-local driver/operator recipe. A click now
names a personally reviewed complete PNG pixel, with its retained actual viewport
and game-screen rectangle. The constructor converts PNG pixels to CDP viewport
CSS pixels using the measured image and viewport dimensions. The driver checks
the same geometry again immediately before sending the event. It refuses absent
geometry, partial images, viewport offsets/zoom, and points outside the source or
game screen. It does not infer canvas normalization from a displayed image.
Before browser creation, the recursive helper closure, eight actual operator vs handler schemas, physical helper READY/CANCEL actual join/EOF, 183 source files, 305 imports, 311 media references and output IO passed. The unchanged32 native controls were reused by exact adapter/context/test SHA. Three coordinate tests and45 focused MENU checks passed, and the full flock build exited0. The browser used the retained f5ae module rather than that newly built module. Actual private root dimensions and its whole800x600 PNG were verified before Chrome. The W3 incremental admission reader is reused run-locally, retaining measurement-time, partial-report and truncation guards; duplicate full35MB board reads are removed.
The new current hardware dialog was personally viewed at800x600. Its OK box occupied approximately185..235,396..412. Retained geometry was viewport800x600, scale1, offset0, game screen x10,y192,width460,height370. The actual constructor and command handler delivered CDP viewport210,403, with admission age4728ms. Guest input logs show DOWN/UP to HWND0x10004. The dialog retired, focus changed to0x10001, CPU execution left the modal wait, and a fresh whole image showed the name-entry menu. These observations establish ordinaryOK; no native client lParam coordinate was retained, so no exact numeric native-client mapping is claimed. This is actual registered-game progress, not solely fixture evidence.
Personal review and ordinary input then entered RIDER, accepted the name, selected Single Player, Stunt Quarry, Practice, Pine, and the rider/bike screen. The final NEXT on that screen entered Loading. PREPARED had not been established before that NEXT: this ordering does not satisfy the requested postprepare fresh MENU one-Start path. No Start capability was created or consumed, and no ARMED, CPUFlip/sourcecopy/decoder/compositor PUBLISHED proof, rendered quarry, Escape or Quit is credited. Future separately authorized work must prepare at the final rider/bike menu before its NEXT/Start transition, then capture/review fresh MENU through the existing private one-use context; do not replay this run's sources.
Both actual CPU and renderer Workers recovered f5ae exactly once. At13:32:04.369Z
the next ordinary capture encountered the real current-source guard:
current completed synchronous source publication required. The chain and
runtime permanently closed. The last of231 endpoint requests preceded that
failure; zero followed closure. No restart, retries, clock changes or guard
relaxations were made. The final NEXT was ordinary input, not qualified Start.
Native supervision joined normally with exit1 and leaked=false; all32 tracked PIDs were absent. Browser, private display, board watcher, custodian, supervisor, finite watchdog and scope watcher have actual joins; browser/display/board have both EOF receipts. Scope watcher minimum was1547214848 bytes, above the floor. No cancellation was ignored; old run SIGTERM sender/cause remains UNKNOWN. Boxnull NOT_CREATED, expiry/stopnull; delegated owned cleanup completed.
The coordinate correction is retained in ignored run-local recipes/tests; the
tracked adapter/context/tests did not change in this task and their inherited
WIP remains unqualified on the postprepare original path. Evidence contains exact
Git/binary diffs, owned new-file patches and existing media/module references,
without source snapshots or media copies. result.json is published last.