Current-display freshness repair and bounded GLTron attempt
The admission helper confused publication age with observation freshness. Commit f6962ba5b1cf2829a66f5cc4d80b293c30719b37 removes only the displayedAt age cutoff. A newly requested readback of unchanged, legitimately displayed pixels is now accepted; old requests still expire at the original 30,000 ms limit. No timestamp, renderer publication, runtime source, or game media was changed.
Evidence is scratch/runs/20261118-static-display-fresh-capture-gltron-gameplay/result.json, written last. This task repaired and tested the helper, but did not demonstrate GLTron turns, a round outcome, Controls, or ordinary Quit. Its single original scope closed before input on an identified run-local filename collision. There was no original retry.
Semantics and causal evidence
Win98Renderer.requestCurrentDisplayCapture records a new request time and reads the complete retained ordinary publication, or waits for ordinary publication if required. _validateCurrentDisplayCapture checks the current record, full source/output, visibility, geometry, window/process/instance and canvas continuity. _readImmutableCapture validates again after reading pixels. displayedAt remains the time of that ordinary publication. It need not advance while a static native window remains displayed. Completed tickets retain historical immutable bytes; their existence does not grant fresh observation admission.
verifyReadback validates receipt fields and timing, not an independently live document. Callers must retain exact document/process/window/canvas pins and validate around capture. The change preserves that division. An old request cannot pass by completing recently. Replaced publication/identity during readback remains a renderer failure; no maxAge increase or forced repaint was introduced.
Three bounded authored native GDI browser scopes used the retained Caesar control PE, running through the production WAT interpreter/Worker/renderer. calibration-static reproduced the original rejection at display age 31,888 ms and request age 82 ms. calibration-fixed retained further negatives; its candidate validation attempted outside the browser run was already stale and is not a positive. calibration-final performed the old/new comparison immediately on the same actual receipt: display age 32,761 ms, request age 43 ms, old failure and new pass, unchanged publication and original display timestamp. The genuinely old startup ticket was rejected for request freshness.
Actual browser negatives refused hidden/clipped windows, equal-looking window replacement, changed process/instance, replaced logical canvas and hidden output. Ordinary trusted Enter, with a separately prearmed release and observed physical UP, changed the authored window to “MENU: Escape Quit” and advanced the native publication. Selected whole screenshots were personally reviewed. A new actual document-navigation negative was not completed in this task; document-scoped object tokens and the existing identity checks remain required. No such new test is claimed.
Focused admission tests and all 33 existing native current-display/capture tests passed, including identity change during readback, pending owner replacement, cancellation, writer/publication continuity and producer retirement. flock /tmp/wine-build.lock bash tools/build.sh passed on the shared working tree; foreign WIP was excluded from the browser's immutable Git runtime. The code commit was immediately pulled and pushed with the required strict SSH settings. ReVolt explicitly confirmed its in-flight helper bodies were frozen to 30e35805f and excluded this edit. LTris and foreign WIP were untouched.
One original, closed before input
The original registered GLTron used published graph f6962ba5b, with all 92 original media hashes checked before and after. Its runtime source was unchanged from the compiled calibration graph 81150aeb2; the retained 2,418,710-byte module is SHA-256 a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8. Compilation/served bytes and the source graph are known. Independent executing child bytecode and role classification were omitted and remain unknown.
The prepared operator fixed startup 180 s, original runtime 1,200 s, Quit reserve 180 s, cleanup 20 s, capture/transport 3 s, review 30 s and independent release lease 20 s. The copied policy's authoredScopes:1 field was an inaccurate template value: three authored scopes actually ran, each with its own fixed 300 s runtime. No deadline was extended. No original input was sent.
The initialized-publication observation reached native publication 7. The operator saved host-initial.json for the host observation, then called the whole-display helper with host-initial; that helper also creates host-initial.json with exclusive creation. EEXIST permanently closed the shared scope before ready/input. The helper had already acquired host-initial.xwd; offline PNG encoding recovered that whole host image, which shows a black transitional app area and fullscreen-consent strip. It is not gameplay or a newly qualified native scene. This is an orchestration defect, not evidence of a GLTron runtime defect or a remaining static-age failure. The attempted post-closure calls were rejected locally, with zero later wire issues. No endpoint was reopened.
The authored scopes exited 0; the original driver and supervisor exited 1. Browser, Xvfb, Openbox, release child, actual driver, supervisor and watchdog waits record stdout/stderr EOF. Supervisor receipts report no leaks. The outer tool supplied exit status but no separate EOF flags, which remain unknown. No box was created: ID null, state NOT_CREATED, expiry null, stop receipt null. Disk sampling stayed above 1.5 GB; no RAM gate was used.
Recover runtime from the recorded Git graph with gitFiles(commit) and compile(files), then compare the retained module hash. Media recovery uses the recorded existing paths/hashes, with no media copies. Relevant binary diffs, the owned commit patch, run-local new-file patches, unique inputs, selected screenshots, waits and offline checks are retained in the run. The original scope remains closed; this task claims only the tested freshness repair and the bounded original operator failure.