evidence · 3.6 KiB

coordinator-review.md

Committed at docs/evidence/card-shell/coordinator-review.md. This is what grounds the requirements that cite it.

Open raw source

Coordinator review of the opt-in product card adapter

Reviewed source commits 9865d65 and 3d1ae3f, plus independent touch receiver fixture 321601d. Root integration equivalents are 1284122, 523c5a4 and 952c20e. No default-image integration or product board deployment is claimed.

The root coordinator reviewed output/seat teardown, touch ownership, wrapper runtime dependencies and telemetry/commit correlation. The pinned wlroots DRM backend requires DRM_CAP_TIMESTAMP_MONOTONIC in backend/drm/drm.c; native headless presentation uses the corresponding monotonic backend path. Output teardown disarms the benchmark before releasing output-owned state.

Review found two concrete input bugs: the hidden Cards/edge regions could steal Apps touches, and a second contact on the top bar could forward a down while swallowing its up. Both were corrected before this handoff. Separately, the author found XDG popups outside the mirrored content tree and added explicit normal-mode fallback rather than leaving an unscaled/private popup above cards.

The root independently reran:

python3 tests/card_shell_runtime.py \
  --sway /nix/store/x95yrya49qds0qzrxfyqp000b01pggsp-sway-unwrapped-riscv64-unknown-linux-gnu-1.12/bin/sway \
  --client /nix/store/6nskh3ldk3ya7mwxbb258qszrxhw2i61-card-composition-probe-client-0.1/bin/card-composition-probe-client \
  --native-touch --output /mnt/MediaVolume/home/jadams/k230-root-card-product-review
python3 tests/test_card_touch_routing.py \
  --sway /nix/store/x95yrya49qds0qzrxfyqp000b01pggsp-sway-unwrapped-riscv64-unknown-linux-gnu-1.12/bin/sway \
  --output /mnt/MediaVolume/home/jadams/k230-root-touch-routing-review

PASS: seventeen product runtime assertions, seven independent Wayland receiver pairing/recovery cases, and clean compositor/client exits. Input travels through a registered test wlroots touch device, cursor, seat and protocol delivery. This is stronger host routing evidence than direct calls to card handlers; it remains injected headless execution and cannot prove the board’s touch hardware or real-finger feel. The native protocol fixture’s normalized evidence is in ../card-touch-routing-host/; adapter evidence and rejected host cost acceptance are in headless/.

The final reviewed package is /nix/store/c41z2czaiana4dh8a7a2w6pwrqkawdgj-k230-card-shell. It remains opt-in. The earlier measured headless benchmark exceeded the frame-interval budget and lacked isolated-session memory accounting; those results remain diagnostic FAIL/INCOMPLETE, never product acceptance. Default Pixman board cost, integrated controls, QEMU system-image smoke, physical gestures and optical review remain open in the product proposal. No timeout or card-count reduction substitutes for those requirements.

Native input provenance follow-up

The coordinator reviewed 935e021 (integration f514478) and independently ran:

TMPDIR=/mnt/MediaVolume/home/jadams python3 tests/test_card_touch_routing.py \
  --provenance \
  --sway /nix/store/fwyx51499i8blppf5rghziaiwm1q3i2i-sway-unwrapped-riscv64-unknown-linux-gnu-1.12/bin/sway \
  --output /mnt/MediaVolume/home/jadams/k230-root-provenance-review

PASS: seven native Wayland pairing/recovery cases and five provenance cases, including rejection of mixed or mismatched input sources. Benchmark arming does not rewrite actual device provenance. The physical-name fixture is still synthetic; a label alone never proves a real finger. Board acceptance must verify the injected device’s exact name and virtual sysfs origin. The resulting opt-in package is /nix/store/ii5g7635wldnk7jsrfp18ism634w5y70-k230-card-shell. This follow-up does not close a board or image gate.