Inspect recorded video segments in native Review

[Jacob-Met]

Review now lets a researcher inspect retained video without leaving the package. Expand Recorded video, select an exact source/stream/package-relative segment, then explicitly play, pause or seek. Equal filenames stay distinct. The visible clock is within that media segment; it does not infer camera or checkpoint alignment.

Closes #89.

The viewer lazily uses the existing PySide6 multimedia backend. Loading a package or selecting a segment does not autoplay; no audio output is attached. Changing selections, unavailable/invalid media and failed package loads retire the old picture and active controls. Native callbacks are bound to both player identity and selection generation. Collapse or leaving Review pauses playback.

The change stays in two new modules and seven additive ReviewScreen hooks. Focused Play/Pause, Reload and disclosure buttons accept their own unmodified Space activation; other keys, modifiers and controls retain the original application shortcuts. Capture, daemon, raw files, discovery, schemas, timeline, analysis and export are unchanged by this contribution. PR83 retains the separate event-browser scope.

Current source

Head 2a30559ec98aef5ce934211a321497c15be367f5 composes the reviewed contribution with current main 9e9204f76b105426b0affaa74733175c052f27ba (Analysis prediction input and ML feature pairing included). Native equivalent 2d66a15c78b5cd2d3fe7b35b212d09a030b61c2a has the exact same complete tree d1af33caaf7154e83def1dab035920907d417e8d.

All 1,204 unrelated current-main leaves and 75 accepted contribution paths are byte/mode-exact. PROGRESS.md is the sole conflict resolution: the complete current document is followed by the exact owned append. The conflict preimage and resolution hashes are retained on LA7. All reviewed runtime, maintained test and synthetic fixture bytes remain exact. No other owner's branch was changed.

Qualification

• Current combined native source: 184 model/UI tests passed, zero failures/errors/skips, in 44.10 seconds. This includes the incoming actual Analysis chooser/application cases and the actual MainWindow video-button regression with C, Ctrl+Space and outside-viewer shortcut countercontrols.
• Real QMediaPlayer/QVideoSink cases decode known synthetic red/blue/green frames; pause, 1500/2000 ms seek, duplicate identities, invalid/missing media, reload and package retirement are exercised. No hardware, real recording or daemon is used by those video fixtures.
• Independent MSI source review and a separate six-group native oracle accept lifecycle/stale-signal behavior. The keyboard successor has a separate 90-case native event-boundary matrix: only the three intended unmodified-Space ShortcutOverride events are claimed.
• The final ordinary Windows run 37841504535 passed 673 Python cases + 6 expected daemon skips, 30 CTest cases, and 6 native daemon/recovery cases. Both jobs checked out 16858f1909182b171b5478b9b647de29c33d15dc, the exact current-main/head composition with tree d1af33caaf7154e83def1dab035920907d417e8d. The Python gate explicitly passed both actual MainWindow video-button and incoming Analysis-button shortcut regressions. Full raw logs are preserved on LA7.
• The previous corrected-head run 37838027180 remains historical (641 Python passes, 6 expected daemon skips, 30 CTest and 6 native integration passes); it is not substituted for the final gates.
• Earlier setup, receiving, style and full-application keyboard failures retain their original logs/pins. The keyboard failure was corrected before integration; the initial 151-case/native and later 24-case correction receipts remain historical.

Presentation boundary

Readable dark/light/compact control captures and actual decoded frame pixels are received separately. Offscreen widget grabs omit QVideoWidget's GPU surface; a Windows-platform HWND grab returned an entirely black image and failed its assertion. Onscreen GPU-video rendering is not qualified by those captures. The failed driver/log/image remain intact, and the standard native renderer was not changed to satisfy screenshot tooling.

Operator guide · evidence and limits · keyboard correction · independent source/lifecycle review.

[Jacob-Met]

Current-parent receiving at 2026-10-08T20:38Z: ordinary run 37838027180 receives synthetic merge ab4a9ed257d8aeb12b7e5debb3a46f66517a9783 (tree aeece9df1c6f732b6325c7e92b963c7ee030fafe), with parents current main e48825b1b37544f4c31912e3c29a3918704f5456 and corrected head e90c51fe713038f29598bb5ec420514a44283364.

The full supported Python job passed 641 tests, 6 expected daemon skips, 0 failures/errors in 298.62 seconds. test_actual_main_window_video_buttons_keep_space_activation is explicitly PASSED in the raw job log. Its JUnit SHA-256 is 393e937ba510ca5ac42d8474a9126fdf6fbc57945a03c9fc8efc726fb4913493; the runner received 1,245 source files. The unchanged 300-second bound remains in force.

Native public-Git receiving verifies all 1,169 unrelated current-main leaves exact, including the newly merged ML feature pairing and its diagnostic runner. 75 of 76 contribution paths remain byte/mode-exact; PROGRESS.md is the sole composition, preserving the whole current-main prefix and whole accepted contribution suffix. Every reviewed runtime, test and fixture byte is unchanged. The author's own isolated worktree remains clean.

This is the corrected-head Python/current-parent acceptance. The ordinary C++/native job is still running; merge remains pending that gate. Earlier initial-head and native/independent receipts keep their historical pins, and the onscreen GPU capture limitation remains unchanged.

[Jacob-Met]

Merged as 3960c0c756cd4e9facbde0d76eaaa3c31ab7c167 after the final ordinary Windows gates succeeded: 673 Python passes + 6 expected daemon skips, 30 CTest passes, and 6 native daemon/recovery passes. Both jobs used exact synthetic merge 16858f1909182b171b5478b9b647de29c33d15dc.

Actual public-Git readback of the merged commit confirms parents 9e9204f + 2a30559, tree d1af33caaf7154e83def1dab035920907d417e8d, and all 1,280 leaves equal the accepted native tree. All 1,204 unrelated current-main leaves remain exact. The native source is clean; its 184-case combined UI receipt and full hosted logs are sealed in C:\Users\minec\hamon-0378a7b6-capture-runtime\merged-native-readback.json. Issue #89 is closed as completed.

The normal post-merge main push run is running and will receive a separate actual-source readback. The documented onscreen GPU presentation limitation remains unchanged; no failed capture is represented as a successful visual qualification.

[Jacob-Met]

Final receiving is complete. The ordinary main push workflow 37844755177 passed both Windows jobs on actual merged commit 3960c0c756cd4e9facbde0d76eaaa3c31ab7c167, tree d1af33caaf7154e83def1dab035920907d417e8d:

• 673 Python cases passed, with 6 expected daemon skips; both actual MainWindow video-button and Analysis-button shortcut regressions passed.
• 30 CTest cases passed.
• 6 native daemon/recovery cases passed.

This is the post-merge run, separate from the already-successful final PR gates. Its complete raw job logs and API observation are preserved alongside the exact 1,280-leaf native readback and 184-case combined native UI receipt. All 1,204 unrelated previous-main leaves remain exact.

Final native delivery receipt: C:\Users\minec\hamon-0378a7b6-capture-runtime\delivery-final.json, SHA-256 22ed7254c8d95f10a29ae9e78f2392761c85ae9c39e4256344a4f235821a6062. The external advisory claim records delivery and released source ownership; it is not a resident lease. The documented onscreen GPU presentation limitation remains explicit. No further source work is pending in this lane.