## What happened

quickshell (Omarchy shell) segfaulted at 17:32:03 while the audio panel's
device list was being rebuilt during a full PipeWire stack restart
(`omarchy-restart-audio`, run via the terminal keybinding). Quickshell's
in-process crash handler caught it, wrote a report, and the shell
auto-restarted within about a second. No data was lost.

Sequence, from `journalctl` and the crash report:

1. `omarchy-restart-audio` stops and restarts `pipewire`, `pipewire-pulse`,
   and `wireplumber`.
2. Within the same second, quickshell's PipeWire service log shows a burst
   of new `PwNode`/`PwMetadata` objects as the whole audio stack comes back
   — not a single node add/remove, the *entire* node set torn down and
   recreated at once.
3. Quickshell segfaults immediately after.

## Stack (symbolized, from the crash report)

```
QQmlTimer::event
 -> QQmlBoundSignalExpression::evaluate -> QQmlBinding::update
 -> QQmlPropertyPrivate::write -> QQuickRepeater::setModel -> regenerate()
 -> QQmlIncubatorPrivate::incubate -> QQmlObjectCreator::createInstance
 -> QV4::Module::evaluate -> QV4::Runtime::TypeofName::call
 -> QObjectWrapper::getQmlProperty   <- SEGV
```

This is the same fault family as #11202 and #11727: a `Repeater`'s model is
reassigned while Qt is still incubating delegates from the previous model,
and the incubation step dereferences something that's no longer valid.

## Why this is a new report, not a duplicate

`plugins/panels/audio/Panel.qml` already carries the mitigation #11202
points to as the fix pattern — a debounced snapshot (`audioModelRefreshTimer`,
75ms) feeding the Repeaters instead of the live PipeWire-backed lists, with
this comment at ~line 103:

> "PipeWire can remove nodes while Quickshell is dispatching the removal
> signal; rebuilding a Repeater from that signal path has crashed in
> Quickshell's PipeWire service. The snapshot timer lets that mutation
> settle first..."

That mitigation is scoped to a single node coming and going. A full
`omarchy-restart-audio` (or any full pipewire/wireplumber restart) discards
and recreates every node at once — evidently enough to still land in the
same mid-incubation race the debounce was built to prevent. The audio
panel was open at the time (this refresh path only runs while `opened` is
true), consistent with the user having triggered the restart while watching
the panel.

## Expected

Restarting the audio stack while the audio panel is open should not crash
the shell.

## Steps to reproduce

1. Open the audio panel (bar volume applet flyout).
2. While it's open, run `omarchy-restart-audio` (or otherwise fully
   restart pipewire + wireplumber).
3. Quickshell segfaults; Omarchy's supervisor auto-restarts it.

## Related

- #11202 — same Repeater/mid-incubation race, Bluetooth panel, no
  mitigation at all yet.
- #11727 — same race, different faulting frame
  (`VDMListDelegateDataType::createMissingProperties`), triggering widget
  unidentified.

## Evidence

- `coredumpctl info 826652` (SIGSEGV, SI_TKILL via quickshell's own crash
  handler, re-raised for the core dump)
- Full symbolized report: `~/.cache/quickshell/crashes/n71fv8blt/report.txt`
- No quickshell/qt6-declarative update correlates: last package changes
  were the initial install (2026-09-07) and a qt6-base point update
  (2026-09-09); this machine last ran the trigger on 2026-09-13.

## System

- Omarchy 4.0.3, kernel 7.2.3-arch1-3
- quickshell 0.3.1-1, Qt 6.11.2 (qt6-base 6.11.2-3, qt6-declarative 6.11.2-1)
- Hyprland, Wayland session
- GPUs: Intel i915 (8086:46a6) + NVIDIA (10de:24a0)

Filed by Claude Sonnet 5 via Claude Code.