First compositor logo scene trial

The first opt-in Sway scene-seed image no longer reproduces the earlier approximately 1.1–1.2 second dark interval in the sampled handoff frames. It also exposes a cleanup defect: the seed is logged, but its removal is not. The diagnostic source 06cd7a8 incorrectly required a top-layer bar, whereas our normal Swaybar uses a bottom-layer surface with namespace panel. This image is a partial result, not the daily default or final acceptance.

Build provenance identifies the exact image and compositor. Flash transcript records a routine full write, reset, Linux boot and USB-host checks passing, with full readback skipped. No saved home state was restored. Runtime inspection records active shell, seatd and firewall, the successful immutable DRM-owner release, the initial scene seed, and first-enable VO/DSI preservation.

The automatic boot recording and independent sampled-frame audit place the visible logo-to-shell transition at about 72.6–72.7 seconds. Dense samples do not contain a dark frame at that transition. This does not exclude a shorter sub-frame event, and it is a warm-reset observation rather than a battery power-on test. The late physical image and native screenshot show the resulting shell.

The follow-up code uses public Wayland surface readiness and Swaybar’s actual mapped panel bottom-layer surface before matching a compositor presentation event. That corrected code requires its own build and physical recording; this video does not prove it was deployed.

Invalid-asset behavior

A temporary runtime-only service override first points Sway at a missing initial-scene asset and then at a three-byte file. The missing-file inspection and wrong-size inspection show specific diagnostics while the shell remains active. Native images after missing file and after wrong size record the resulting shell. This tests Sway’s optional asset loader, not U-Boot’s missing-logo path.

The override and temporary file are removed in invalid-restore.txt; invalid-cleanup.txt records the restored immutable environment. These runtime experiments do not change the image’s defaults or retain user configuration.

Sampled camera frames from the invalid-asset recording at approximately 20, 45 and 70 seconds also show the portrait shell after the restarts. They do not establish a seamless restart; runtime display teardown is expected here.