## Summary

Numpad keys do not produce digits in the Quickshell **lock screen** or the Quickshell **polkit authentication dialog**, even though numlock is on. Pressing NumLock twice (off, then on) fixes it for that prompt.

The important part: **the compositor's numlock state is already correct when this happens.** This is not `numlock_by_default` failing to apply, so it is not fixable from `input.lua`.

## Affected surfaces

Both are Quickshell, and only these:

- `/usr/share/omarchy/shell/plugins/lock/` — the lock screen (`omarchy-shell lock`)
- `/usr/share/omarchy/shell/plugins/polkit/` — the escalation dialog

Ordinary application windows are unaffected. Since resuming from suspend raises the lock screen, "numlock broken after waking from sleep" and "numlock broken at the password popup" are the same bug.

## Evidence

Sampled `hyprctl devices -j` at 4 Hz across a failing polkit prompt, correlated with the journal:

```
20:19:33  leds=111  main=pixart-hp-wireless-keyboard-and-mouse:true   OFF=[]
20:19:34  leds=000  main=pixart-hp-wireless-keyboard-and-mouse:false  OFF=[pixart-hp-wireless-keyboard-and-mouse]
20:19:34  leds=111  main=pixart-hp-wireless-keyboard-and-mouse:true   OFF=[]
20:19:36  pkexec[131911]: pam_unix(polkit-1:session): session opened for user root(uid=0)
```

`leds` = the three `/sys/class/leds/input*::numlock/brightness` values. `OFF=[]` = keyboards reporting `numLock: false`.

At **20:19:33**, before touching anything: Hyprland reported `numLock: true` on all 11 keyboards, and all three hardware numlock LEDs were lit — yet numpad keys did not register in the dialog. At 20:19:34 I pressed NumLock twice (visible as the LEDs dropping to `000` and returning to `111`). The numpad then worked and the password was accepted at 20:19:36.

So the seat's numlock state is correct throughout the failure. The Quickshell surface does not honour it until a modifier-change event forces a re-read, which suggests the client is not seeding modifier state from the seat when the surface is created.

## Steps to reproduce

No suspend cycle needed — the polkit path reproduces on demand:

```bash
# in one shell, watch for keyboards reporting numlock off
while :; do
  hyprctl devices -j | jq -r '[.keyboards[]|select(.numLock==false)|.name]|join(",")'
  sleep 0.25
done &

# trigger the dialog with numlock ON and its LED lit
pkexec /bin/true
```

Type a digit from the numpad into the password field. It does not register, while the watcher shows no keyboard reporting `numLock: false`. Press NumLock twice and the numpad starts working.

## Expected

The numpad produces digits in Quickshell surfaces whenever the seat's numlock state is active, without requiring a manual toggle.

## Possibly related (separate observation)

While sampling, the `main: true` keyboard designation moved between devices as fcitx5 (started by `omarchy-fcitx5.service`) created and destroyed its Wayland virtual keyboard. Keyboard count oscillates 10↔11 and `main` lands on a different device each time:

```
main=hl-virtual-keyboard-fcitx5
main=jabra-elite-10-(avrcp)                    <- a Bluetooth headset's AVRCP control interface
main=pixart-hp-wireless-keyboard-and-mouse
```

A headset control interface becoming the main keyboard looks wrong independently of this bug. I have not established a causal link, so I'm listing it as an observation rather than a cause.

## Related

- #3224 — older report of the same user-visible symptom, filed in Nov 2025 against the pre-Quattro hyprlock lock screen. The login-screen half discussed there (adding an `input` block to `/usr/share/sddm/hyprland.conf`, which still ships without one) is a genuine and separate fix; this issue is specifically the Quickshell half, which that thread flags as an unresolved separate code path.

## System

```
Omarchy   4.0.0-1
Hyprland  0.56.2
Kernel    7.1.8-arch1-3
Hardware  ASUS Vivobook 16 (V3607VJ), Intel Core 7 240H
GPU       Intel Raptor Lake-P iGPU + NVIDIA RTX 3050 Laptop
DM        SDDM (Wayland greeter), autologin enabled
Input     fcitx5 via omarchy-fcitx5.service; kb_layout us,il
```