Keyboard base keys (task 3.4, partial)

Installed system

The board ran impl/keyboard-base at f8c0d5e4’s tree:

• Bundle: .#kernelMainlineDrmShellLandscapeBootFiles, system 4g5whg4y….
• Install: copied with nix copy over SSH. The four boot files were staged in /var/lib/k230/boot-staging-kbd, hash-checked against the bundle’s SHA256SUMS, and copied into /boot.
• Profile: set to the new system, then rebooted.
• Previous boot files: /var/lib/k230/boot-backup-20261013-pre-kbd/.

Board-read after boot (2026-10-13 ~21:10 UTC)

dmesg:

tca8418_keypad 1-0034: overflow occurred
input: tca8418 as /devices/platform/soc/91409000.i2c/i2c-1/1-0034/input/input2

The single overflow at probe is the chip’s event FIFO left full while LILYGO’s firmware last ran. The driver drains it.

/proc/interrupts shows gpio-k230 10 Level tca8418 (IO42).

swaymsg -t get_inputs lists 1:1:tca8418, a keyboard, with repeat 400/25 from the nix/shell.nix stanza. Sway’s log says “Translated keysyms using config for device ‘1:1:tca8418’” and reports no keymap error.

Operator

• evtest-2026-10-13T211125Z.log: evtest /dev/input/event2 for 300 s from 21:11:25 UTC. It holds 8 presses, KEY_G and KEY_BACKSPACE (DEL) 4 each, from the operator at the keys (board-read, operator pressing).
• Operator report (2026-10-13, chat): “works great” after typing on the base.

Still required for task 3.4

These are not covered by this evidence:

• evtest for every key;
• ls -la in foot, as a native grim capture and a camera frame;
• a Shift-punctuation line;
• Caps Lock with its LED, which needs task 3.3.

Shortcut recording (2026-10-14 01:12–01:17 UTC)

shortcuts-evemu-record-2026-10-14T0112Z.txt is evemu-record on the real tca8418 device while the operator pressed LILYGO+Tab five times and was asked for Fn+LILYGO; shortcuts-sway-bindings-2026-10-14T0112Z.json is swaymsg -t subscribe -m '["binding"]' over the same window (board-read).

• Presses 1–3: Tab arrived with only LILYGO held (Shift, if any, in the same report after it), and card_shell enter fired three times.
• Presses 4–5: Shift arrived 23 ms and 46 ms before Tab, so the modifiers were Super+Shift and no binding matched.
• No KEY_RIGHTALT (Fn) event appears anywhere in the recording.

Why the overview chord leaked Tab (2026-10-14 ~02:10 UTC, injected)

After the Super+Shift+Tab binding landed, the operator reported that LILYGO+Tab opened the overview but also sent Tab to the terminal. Reproduced with an evemu-device clone of the tca8418 keypad (same 1:1:tca8418 identifier), with foot focused, by rebinding Mod4+Tab at runtime with swaymsg:

So Sway ran the --input-device binding and still delivered the key. The chord bindings are now global; the operator’s earlier presses had left five ;10;9~ (Super+Shift+Tab) and one ;9;9~ in the terminal.

Fn and the fixed overview chord (2026-10-14 ~02:40 UTC)

fn-evemu-record-2026-10-14T0237Z.txt and fn-sway-bindings-2026-10-14T0237Z.json (board-read, real keypad, operator presses): Fn alone three times arrives as KEY_RIGHTALT press/release; Fn held with Up gives KEY_UP inside the KEY_RIGHTALT press; one LILYGO+Tab gives a single card_shell enter binding event with the global bindings. Whether the terminal stayed clean after this real press is not recorded; the injected check above showed nothing reached it.