Selected card visibility: headless regression

At 2026-09-24T05:58:49Z, the installed unwrapped Sway executable was used as a negative control against tests/card_shell_runtime.py at a2f6fc02:

python3 -u tests/card_shell_runtime.py \
  --sway /nix/store/qwja5838kgidxzvii70l0smf72snzsqs-sway-unwrapped-riscv64-unknown-linux-gnu-1.12/bin/sway \
  --client /nix/store/bxiz9zkb33m4v97gkgzx91aswfm7cf9f-card-composition-probe-client-riscv64-unknown-linux-gnu-0.1/bin/card-composition-probe-client

The new two-window assertion failed: the output pixel at (284, 700) was (32, 112, 176) both before and after opening the second card, although the IPC tree reported the second window focused. This reproduces a focused but visually obscured floating window, without using application titles or private content.

The same source built with the focus-raise fix:

nix build .#card-shell --max-jobs 1 --cores 4 --no-link --print-out-paths
# /nix/store/h3sk26nxgh4i64wbf7v0n0fx26psn3mw-k230-card-shell
python3 -u tests/card_shell_runtime.py \
  --sway /nix/store/ildcfqicn8j22yih16x42b4mmajqq6x8-sway-unwrapped-riscv64-unknown-linux-gnu-1.12/bin/sway \
  --client /nix/store/bxiz9zkb33m4v97gkgzx91aswfm7cf9f-card-composition-probe-client-riscv64-unknown-linux-gnu-0.1/bin/card-composition-probe-client

The suite passed all 17 headless scenarios. The pixel changed from the first client’s blue (32, 112, 176) to the selected second client’s purple (72, 24, 64). The source-only follow-up d455473a changes indentation; the built executable corresponds to a2f6fc02.

This is QEMU user-emulated compositor output with injected card commands. It does not prove the patched image is installed, nor real-touch or panel behavior. The device operator must install the combined image and repeat card selection on glass before closing the physical gate.