evidence · 2.2 KiB

README.md

Committed at docs/evidence/card-shell/overview-home-bleed-through/board-2026-09-30/README.md. This is what grounds the requirements that cite it.

Open raw source

Themed overview: physical closeout

On 2026-09-30 (America/Chicago), the operator used the installed panel shell with real finger input and explicitly reported: “home icons don’t appear behind the cards.” The operator then framed the camera on the settled overview. A real Foot card is selected, the actual configured wallpaper is visible, and Home grid/dock icons are absent.

Board-native overview

Physical panel photograph

The matching sanitized console probe reports active=1, appearance_enabled=1, appearance_wallpaper=1, a transparent overview canvas and home_enabled=0. Capture provenance and original hashes distinguish the native image from the optical photo. The board and serial port were reserved by the coordinator with /tmp/k230-board.lock; there was no flash, kernel change or reboot during collection.

Commands (run from the repository while holding the reservation):

python3 tools/console.py /dev/ttyACM0 --wait=3 'runuser -u shell -- env SWAYSOCK=/run/shell/sway-ipc.sock timeout 3s swaymsg -r "card_shell enter"; runuser -u shell -- env SWAYSOCK=/run/shell/sway-ipc.sock timeout 3s swaymsg -r "card_shell debug-scene"'
python3 tools/console.py /dev/ttyACM0 --wait=3 'runuser -u shell -- env XDG_RUNTIME_DIR=/run/shell WAYLAND_DISPLAY=wayland-1 grim -t jpeg -q 85 /run/shell/overview-proof.jpg'
ffmpeg -hide_banner -loglevel error -f v4l2 -i /dev/video0 -frames:v 1 -y overview-panel.jpg

The native image was obtained before the repeat enter command, while the same operator-entered overview was visible. The console state after repeat entry stayed consistent. The panel photograph has glare and an oblique angle: it records the lit physical panel and overall composition, while the native image provides readable pixels and the operator supplies the real-glass behavior check. These stills do not prove latency or every motion path. Existing headless entry/mid-drag/settlement and self-healing regressions remain their separately identified evidence class.

This closes task 5.1 of the-overview-hides-the-home-screen; it does not close other card-motion, performance, Home-placement or persistent-boot proposals.