Release 0.6.2 + AppImage packaging

Bump to 0.6.2 (Cargo + Inno .iss), promote CHANGELOG [Unreleased] -> [0.6.2].
0.6.2 rolls up the licensing (MIT + THIRD_PARTY_LICENSES) and the friends-list
liveness fixes (active offline marking, 15s refresh, manual Rescan) on the
0.6.x wire format (gossip v5, compatible with 0.6.0/0.6.1).

Adds packaging/appimage: a thin AppImage recipe (linuxdeploy) that bundles the
pixelpass screen-share helper in usr/bin so peerspeak's $PATH lookup finds it
with no code change. Assets are include_bytes!-embedded; the graphics stack and
pixelpass's gstreamer/mpv tools are left to the host. Built on Ubuntu 24.04
(glibc 2.39) for reach across Debian 13+/Fedora 40+/rolling.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

CHANGELOG.md

@@ -2,7 +2,7 @@
 
 All notable changes to PeerSpeak are documented here.
 
-## [Unreleased]
+## [0.6.2] — 2026-07-03
 
 ### Fixed
 - **Friends list now reflects status changes without a restart.** A presence probe that fails now actively marks the friend **offline**, so a friend who goes offline, leaves a room, or turns invisible no longer lingers showing a stale "online" / "in a room" status until PeerSpeak is relaunched. Previously only successful probes updated the list, so it could ratchet a friend's status up but never down. The auto-refresh interval was also shortened from 60s to **15s** so the list tracks changes more closely.
@@ -13,6 +13,8 @@ All notable changes to PeerSpeak are documented here.
 ### Licensing
 - **PeerSpeak is now released under the MIT License** (previously an unlicensed private build). Added a `LICENSE` file and a `THIRD_PARTY_LICENSES` file enumerating the full dependency manifest plus the canonical text of every referenced license, with notices for the statically bundled Opus codec and the embedded fonts (Iced-Icons, Cantarell/OFL-1.1). Both files ship in the Arch and Debian packages.
 
+[0.6.2]: https://gitbutter.xyz/mollusk/peerspeak/releases/tag/v0.6.2
+
 ## [0.6.0] — 2026-06-28
 
 ### Added

Cargo.lock

@@ -4871,7 +4871,7 @@ checksum = "35fb2e5f958ec131621fdd531e9fc186ed768cbe395337403ae56c17a74c68ec"
 
 [[package]]
 name = "peerspeak"
-version = "0.6.1"
+version = "0.6.2"
 dependencies = [
  "anyhow",
  "async-trait",

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "peerspeak"
-version = "0.6.1"
+version = "0.6.2"
 edition = "2024"
 description = "Decentralized peer-to-peer voice chat (Rust/iroh/PipeWire/Opus/iced)"
 license = "MIT"

build.rs

@@ -2,10 +2,10 @@ use std::process::Command;
 
 fn main() {
     println!("cargo:rerun-if-changed=.git/HEAD");
-    if let Ok(head) = std::fs::read_to_string(".git/HEAD") {
-        if let Some(reference) = head.strip_prefix("ref: ") {
-            println!("cargo:rerun-if-changed=.git/{}", reference.trim());
-        }
+    if let Ok(head) = std::fs::read_to_string(".git/HEAD")
+        && let Some(reference) = head.strip_prefix("ref: ")
+    {
+        println!("cargo:rerun-if-changed=.git/{}", reference.trim());
     }
 
     let short = Command::new("git")

docs/WINDOWS.md

@@ -1,19 +1,28 @@
 # PeerSpeak on Windows
 
-Current status: the Windows port cross-compiles to `x86_64-pc-windows-gnu` and the `.exe`
-launches under Wine. A real Windows/WASAPI host is still needed for the final audio-device
-checks listed below.
+Current status: PeerSpeak cross-compiles to `x86_64-pc-windows-gnu` from Linux and
+has passed an older native Windows 11 VM smoke test for launch, GUI render, call
+join, and audio flow. The build environment is **not** the Windows VM; current
+Windows binaries are built from Linux, normally inside the `peerspeak-win`
+distrobox or with the same GNU target environment.
+
+The Windows runtime still trails Linux in a few important areas. See the Claude
+handoff file `windows-parity-audit.md` for the full audit and task breakdown.
 
 ## What works today
 
 | Area | Status |
 |---|---|
-| GUI | Iced/wgpu builds and renders under Wine. |
-| Networking | Iroh QUIC transport and gossip compile on Windows. |
+| GUI | Iced/wgpu builds for Windows and rendered in the Windows 11 VM. |
+| Networking | Iroh QUIC transport and gossip compile on Windows; VM call reached two peers. |
 | Audio backend | `cpal` drives WASAPI capture/playback behind `AudioBackend`. |
+| Device selection | cpal enumerates input/output devices; see caveat below about stable IDs. |
+| Resampling/remap | WASAPI devices can run non-48 kHz formats; PeerSpeak converts at the backend boundary. |
 | Codec | Opus remains 48 kHz mono, 20 ms frames. |
-| Identity | `ring` identity generation/load is platform-neutral. |
+| Identity/config | Stored through `dirs` under the Windows profile. |
 | Chimes | Windows uses PowerShell `System.Media.SoundPlayer` for WAV playback. |
+| Game detection | Steam registry `RunningAppID` plus Toolhelp process-scan fallback compile on Windows. |
+| File dialogs | `rfd` uses the native Win32 dialog backend. |
 
 Windows paths are resolved through `dirs`:
 
@@ -23,55 +32,72 @@ Windows paths are resolved through `dirs`:
 
 ## Building
 
-### Native Windows
+### Cross-compile from Linux
 
-Install MSVC Build Tools and CMake, then build normally:
+Preferred local path:
 
-```powershell
-cargo build --release
+```sh
+distrobox enter peerspeak-win -- bash -lc '
+  cd ~/git/butter/peerspeak &&
+  RUSTC_BOOTSTRAP=1 ./win-cross-build.sh -Z build-std=std,panic_abort
+'
 ```
 
-If CMake is 4.x or newer, the vendored `opus`/`libopus` build may need:
+Equivalent direct command when the host has the GNU target, MinGW, `rust-src`, and
+CMake available:
+
+```sh
+CMAKE_POLICY_VERSION_MINIMUM=3.5 RUSTC_BOOTSTRAP=1 \
+  cargo build --release --target x86_64-pc-windows-gnu --bin peerspeak \
+  -Z build-std=std,panic_abort
+```
+
+`CMAKE_POLICY_VERSION_MINIMUM=3.5` is required with host CMake 4.x because the
+vendored Opus build used by `audiopus_sys` still declares an old minimum CMake
+version. Without that env var, the Windows build/check fails during Opus configure.
+
+### Native Windows
+
+A native MSVC build is not the active development path. If used, install MSVC Build
+Tools and CMake, then build normally:
 
 ```powershell
 $env:CMAKE_POLICY_VERSION_MINIMUM = "3.5"
 cargo build --release
 ```
 
-### Cross-compile from Linux
-
-The current dev path cross-compiles from an Arch environment to the GNU Windows target:
-
-```sh
-rustup target add x86_64-pc-windows-gnu
-sudo pacman -S mingw-w64-gcc cmake
-CMAKE_POLICY_VERSION_MINIMUM=3.5 cargo build --release --target x86_64-pc-windows-gnu --bin peerspeak
-```
-
-Wine is useful for launch/render smoke tests, but it is not a substitute for a real
-Windows audio-device pass. The deeper migration plan (phases, decisions, the opus build
-spike) lives in the maintainer's handoff docs, outside the repo.
-
 ## First run and networking
 
-Expect a Windows Firewall prompt the first time the app opens network sockets. Allow it:
-PeerSpeak uses UDP for QUIC, plus relay traffic when direct NAT traversal is not available.
+Expect a Windows Firewall prompt the first time the app opens network sockets, or
+use the Inno installer option that pre-adds a firewall allow rule. PeerSpeak uses
+UDP for QUIC plus relay traffic when direct NAT traversal is unavailable.
 
-The default network mode keeps the n0 relay available for NAT traversal without publishing
-presence to n0 DNS. Direct peer-to-peer paths may work when both networks allow them; relayed
-connections are expected and valid.
+The default network mode keeps the n0 relay available for NAT traversal without
+publishing presence to n0 DNS. Relayed connections are expected and valid.
 
 ## Known gaps
 
 | Item | Status |
 |---|---|
-| Echo cancellation | Linux-only PipeWire feature. The Windows UI shows it disabled as unavailable. |
-| Screen share | Requires a Windows `pixelpass.exe` on `PATH` or a configured override. |
-| Chimes | Now routed through Windows `SoundPlayer`; needs a real Windows host to audibly verify. |
-| Resampling/device format | Cross-compiled. cpal/WASAPI now chooses native 48 kHz when available and otherwise resamples/remaps at the device boundary; needs real Windows hardware audio verification. |
-| Device persistence | Open. WASAPI friendly names may duplicate or change across driver/profile changes. |
-| Playback pacing | Cross-compiled. The fixed playback target under WASAPI shared mode still needs real-hardware verification with `audio_probe`. |
+| Echo cancellation | Linux-only today. The Windows UI shows it disabled as unavailable. |
+| Screen share | Blocked by PixelPass, which is currently Linux-only in practice. PeerSpeak can spawn `pixelpass.exe`, but there is no Windows PixelPass host/viewer parity yet. |
+| Device persistence | Uses cpal friendly names as keys. These can duplicate or change across Windows driver/profile changes; stable WASAPI endpoint IDs are still needed. |
+| Release hygiene | Keep `.iss` and installer output in sync with `Cargo.toml`; rebuild Windows artifacts during each release. |
+| Runtime coverage | The Windows VM smoke test proved an older tester build. Current `main` needs a fresh VM smoke matrix before calling parity current. |
 
-Before calling Windows support done, verify a real Windows machine can create/join a room,
-capture mic audio, hear remote audio, select devices, restart with selections preserved, and
-play notification chimes.
+## Current smoke checklist
+
+Before calling a Windows build current, verify on the Windows VM or real Windows
+hardware:
+
+- Launch current `peerspeak.exe`; GUI renders and settings open.
+- Run `audio_probe.exe 440 30`; listen for glitches and inspect `playout-health`.
+- Create/join a Linux <-> Windows room; confirm mic and playback both directions.
+- Select input/output devices, restart, and confirm selections persist or fall back clearly.
+- Play chimes and custom chime paths.
+- Send chat, image/file attachments, and save an attachment through the native dialog.
+- Import/play/share/listen to music from Windows file paths.
+- Record mixed/stems/both and inspect the WAV output path.
+- Exercise friends/presence/recents and the clock-skew banner.
+- Test Steam and non-Steam game detection on a real Windows Steam install.
+- Install/upgrade/uninstall through the Inno installer, including firewall rule cleanup.

packaging/PKGBUILD

@@ -1,7 +1,7 @@
 # Maintainer: mollusk <jitty+lc1iz0dc@protonmail.com>
 pkgname=peerspeak-git
 _pkgname=peerspeak
-pkgver=0.6.1.r310.g660261a
+pkgver=0.6.1.r315.ga78860d
 pkgrel=1
 pkgdesc="Decentralized peer-to-peer voice chat (Rust/iroh/PipeWire/Opus/iced)"
 arch=('x86_64')

packaging/appimage/.gitignore

@@ -0,0 +1,4 @@
+.tools/
+AppDir/
+*.AppImage
+squashfs-root/

packaging/appimage/AppRun

@@ -0,0 +1,11 @@
+#!/bin/sh
+# AppRun for the PeerSpeak AppImage.
+#
+# PeerSpeak bundles the pixelpass screen-share helper in usr/bin. We prepend our
+# own usr/bin to PATH so peerspeak's $PATH lookup for `pixelpass` finds the
+# bundled copy, while the host's tools (gst-launch-1.0, pactl, mpv — which
+# pixelpass in turn shells out to) remain reachable via the appended host PATH.
+# That no-sandbox spawning is exactly why this app suits AppImage over Flatpak.
+HERE="$(dirname "$(readlink -f "$0")")"
+export PATH="$HERE/usr/bin:$PATH"
+exec "$HERE/usr/bin/peerspeak" "$@"

packaging/appimage/README.md

@@ -0,0 +1,76 @@
+# PeerSpeak AppImage
+
+A "thin" AppImage: the `peerspeak` binary, the bundled `pixelpass` screen-share
+helper, a launcher (`AppRun`), and the desktop entry + icon. Run
+`./build-appimage.sh` to produce `peerspeak-<version>-x86_64.AppImage`.
+
+## Why thin, and why pixelpass is bundled
+
+PeerSpeak owns voice; **pixelpass** owns pixels. They are never Cargo
+dependencies of each other — peerspeak shells out to the `pixelpass` binary over
+its CLI. The AppImage co-locates `pixelpass` in `usr/bin`, and `AppRun` prepends
+`usr/bin` to `PATH`, so peerspeak's normal `$PATH` lookup finds it with no code
+change. Joe gets one file, and screen-share works out of the box.
+
+Almost nothing is bundled: peerspeak's own assets (notification WAVs, avatar
+presets, window icon, fonts) are `include_bytes!`-embedded, and the graphics
+stack (`libGL`, `libvulkan`, `libwayland-*`, `libxkbcommon`, X11) is dlopen'd at
+runtime and on the AppImage excludelist because it must match the host driver.
+So the image carries just the two binaries plus a handful of small libs.
+
+## Host requirements
+
+The AppImage runs on any reasonably current glibc-based distro that has:
+
+- **A Vulkan-capable GPU + driver** (peerspeak's iced/wgpu renderer). Mesa/RADV
+  on AMD/Intel or the NVIDIA driver all work.
+- **PipeWire** (with the PulseAudio shim, for `pactl`).
+- For **screen-share only** — pixelpass shells out to these on the host `PATH`;
+  it prints the exact package names for your distro if any are missing:
+  - **GStreamer + plugins** (`gst-launch-1.0`/`gst-inspect-1.0`, base,
+    good/bad/ugly, libav, and the PipeWire plugin),
+  - **mpv** (or vlc) for the viewer side,
+  - on X11, `xwininfo` for single-window capture.
+
+On Arch/Artix that is one pacman line, e.g.:
+
+```sh
+sudo pacman -S gstreamer gst-plugins-base gst-plugins-good gst-plugins-bad \
+  gst-plugins-ugly gst-libav gst-plugin-pipewire mpv xorg-xwininfo libpulse
+```
+
+(Add `gstreamer-vaapi` for hardware H.264 encode on AMD/Intel; the software
+x264 path always works. On XLibre / X11 the capture path uses `ximagesrc` and
+needs no XDG portal — no systemd required.)
+
+## Building for broad compatibility (glibc baseline)
+
+An AppImage requires a host glibc **at least as new** as the build host's. Built
+on a rolling distro (glibc 2.43) it only runs on equally-new systems. Build
+inside **Ubuntu 24.04** (glibc 2.39, PipeWire 1.0.5) for wide reach — pixelpass's
+`pipewire` crate binds the system PipeWire headers and needs PipeWire >= 1.0, so
+the older Debian 12 `peerspeak-bookworm` box (PW 0.3.65) cannot build it. 2.39
+covers Debian 13+, Fedora 40+, and current rolling distros.
+
+```sh
+# One-time: an Ubuntu 24.04 distrobox that reuses the host rustup toolchain.
+distrobox create --yes --image ubuntu:24.04 --name peerspeak-appimage
+distrobox enter peerspeak-appimage -- sudo apt-get update
+distrobox enter peerspeak-appimage -- sudo apt-get install -y \
+  build-essential cmake clang libclang-dev pkg-config \
+  libpipewire-0.3-dev libspa-0.2-dev libasound2-dev libxcb1-dev \
+  curl ca-certificates file patchelf git
+
+# Build (the host's ~/.rustup toolchain is glibc-2.17-baseline, so it runs in the
+# box; isolated CARGO_TARGET_DIRs keep it off the host target/):
+distrobox enter peerspeak-appimage -- env \
+  PATH="$HOME/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/bin:$PATH" \
+  ./packaging/appimage/build-appimage.sh
+```
+
+## Caveats
+
+- **Hardware encode (VAAPI)** uses the host GPU driver and can't be bundled; the
+  software x264 path always works.
+- The bundled `pixelpass` is built headless (no `gui` feature) — it is only ever
+  driven by peerspeak, never launched standalone from this image.

packaging/appimage/build-appimage.sh

@@ -0,0 +1,89 @@
+#!/usr/bin/env bash
+# Build a "thin" PeerSpeak AppImage that also bundles the pixelpass screen-share
+# helper.
+#
+# PeerSpeak is an iced/wgpu GUI app; pixelpass is the separate screen-share
+# orchestrator peerspeak shells out to (never a Cargo dependency). Both link
+# almost nothing — the graphics stack (libGL, libvulkan, wayland, xkbcommon,
+# X11) is dlopen'd at runtime and is on the AppImage excludelist because it must
+# match the host driver, and pixelpass's capture/encode tools (gst-launch-1.0,
+# pactl, mpv) are expected on the host PATH. So the AppImage carries just the two
+# binaries plus their handful of non-excludelisted libs. The custom AppRun
+# prepends usr/bin to PATH so peerspeak's own $PATH lookup finds the bundled
+# pixelpass, while the host's tools stay reachable.
+#
+# All runtime assets (notification WAVs, avatar presets, window icon, fonts) are
+# include_bytes!-embedded in the peerspeak binary, so nothing else is bundled.
+#
+# Usage:   packaging/appimage/build-appimage.sh
+# Output:  packaging/appimage/peerspeak-<version>-x86_64.AppImage
+#
+# Build inside an Ubuntu 24.04 distrobox (glibc 2.39, PipeWire 1.0.5) for broad
+# reach — pixelpass's `pipewire` crate needs PipeWire >= 1.0 headers, so the
+# older peerspeak-bookworm box (PW 0.3.65) cannot build it. The 2.39 baseline
+# covers Debian 13+, Fedora 40+, and all current rolling distros. See README.md.
+set -euo pipefail
+
+here="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+repo="$(cd "$here/../.." && pwd)"
+tools="$here/.tools"
+appdir="$here/AppDir"
+mkdir -p "$tools"
+
+# linuxdeploy is itself an AppImage; run it without FUSE so this works in a
+# container / on CI without libfuse2.
+export APPIMAGE_EXTRACT_AND_RUN=1
+VERSION="$(grep -m1 '^version' "$repo/Cargo.toml" | sed -E 's/.*"(.*)".*/\1/')"
+export VERSION
+
+# Isolated target dirs so an old-glibc box build never clobbers the host target/.
+cache="${PEERSPEAK_APPIMAGE_CACHE:-$HOME/.cache/peerspeak-appimage}"
+ps_target="$cache/peerspeak-target"
+pp_target="$cache/pixelpass-target"
+
+# The pixelpass screen-share helper we bundle. Sibling checkout by default.
+pixelpass_repo="${PIXELPASS_REPO:-$repo/../pixelpass}"
+if [ ! -d "$pixelpass_repo" ]; then
+  echo "!! pixelpass repo not found at $pixelpass_repo (set PIXELPASS_REPO)" >&2
+  exit 1
+fi
+
+echo ">> building peerspeak (release)"
+( cd "$repo" && CARGO_TARGET_DIR="$ps_target" cargo build --release )
+ps_bin="$ps_target/release/peerspeak"
+
+# Headless pixelpass: peerspeak drives it via `--host`/viewer + `--output json`,
+# never its GUI, so the default (no `gui` feature) keeps the GL toolkit out.
+echo ">> building pixelpass (release, headless) from $pixelpass_repo"
+( cd "$pixelpass_repo" && CARGO_TARGET_DIR="$pp_target" cargo build --release )
+pp_bin="$pp_target/release/pixelpass"
+
+echo ">> fetching linuxdeploy"
+ld="$tools/linuxdeploy-x86_64.AppImage"
+if [ ! -x "$ld" ]; then
+  curl -fL --retry 3 -o "$ld" \
+    "https://github.com/linuxdeploy/linuxdeploy/releases/download/continuous/linuxdeploy-x86_64.AppImage"
+  chmod +x "$ld"
+fi
+
+echo ">> assembling AppDir"
+rm -rf "$appdir"
+mkdir -p "$appdir/usr/bin"
+install -m755 "$ps_bin" "$appdir/usr/bin/peerspeak"
+install -m755 "$pp_bin" "$appdir/usr/bin/pixelpass"
+
+echo ">> running linuxdeploy (bundles libs, builds the AppImage)"
+# -e (repeated): analyse both binaries for libraries to bundle; excludelisted
+#   graphics/glibc libs are skipped. -d/-i: desktop entry + icon.
+# --custom-apprun: our launcher that puts the bundled pixelpass on PATH.
+( cd "$here" && OUTPUT="peerspeak-${VERSION}-x86_64.AppImage" "$ld" \
+    --appdir "$appdir" \
+    -e "$appdir/usr/bin/peerspeak" \
+    -e "$appdir/usr/bin/pixelpass" \
+    -d "$repo/packaging/peerspeak.desktop" \
+    -i "$repo/assets/icons/peerspeak-256.png" \
+    --icon-filename peerspeak \
+    --custom-apprun "$here/AppRun" \
+    --output appimage )
+
+echo ">> done: $here/peerspeak-${VERSION}-x86_64.AppImage"

packaging/windows/INSTALL.md

@@ -8,7 +8,7 @@ it once, then you and I connect directly to each other.
 
 ## 1. Install it
 
-1. Double-click **`peerspeak-0.4.0-setup.exe`** (the file I sent you).
+1. Double-click **`peerspeak-<version>-setup.exe`** (the file I sent you).
 
 2. **Windows will probably show a blue "Windows protected your PC" warning.**
    This is normal — it shows up for any app that isn't from a big company with a

packaging/windows/README.md

@@ -11,8 +11,9 @@ runtime, so there are no extra DLLs to bundle. The installer payload is just the
 
 ## Version compatibility
 
-The installer version tracks the crate version in `Cargo.toml` (currently
-**0.4.0**) — keep `MyAppVersion` in `peerspeak.iss` in sync when it changes.
+The installer version tracks the release version in `Cargo.toml` — keep
+`MyAppVersion` in `peerspeak.iss` in sync when cutting a release. Do not reuse an
+old installer filename after a crate-version bump.
 
 Per `VERSIONING.md`, a **MINOR** bump in `0.x` is a **breaking wire change**:
 peers on different MINOR versions can't connect (they fail fast at the

packaging/windows/peerspeak.iss

@@ -12,7 +12,7 @@
 ; (x86_64-pc-windows-gnu, statically linked -- no extra DLLs needed).
 
 #define MyAppName "PeerSpeak"
-#define MyAppVersion "0.6.0"
+#define MyAppVersion "0.6.2"
 #define MyAppPublisher "mollusk"
 #define MyAppExeName "peerspeak.exe"
 

src/app/mod.rs

@@ -4,7 +4,7 @@ use crate::audio::clip_player::{
 };
 use crate::audio::eq::{EQ_GAIN_DB_MAX, EQ_GAIN_DB_MIN, EqSettings};
 use crate::audio::{AudioDevice, enumerate_audio_devices};
-use crate::config::{AppConfig, NetworkMode, RecordingMode, RoomLayout};
+use crate::config::{AppConfig, AudioProfile, NetworkMode, RecordingMode, RoomLayout};
 use crate::core::{
     CoreController,
     messages::{CoreCommand, UiEvent},
@@ -377,6 +377,7 @@ pub enum AppMessage {
     /// immediately but does not persist (saved once on release via NoiseGateChanged).
     NoiseGateDragging(f32),
     NetworkModeSelected(NetworkMode),
+    AudioProfileSelected(AudioProfile),
     RecordingModeSelected(RecordingMode),
     /// Choose the friends presence posture (W7): invisible / normal / discoverable.
     PresenceModeSelected(PresenceMode),
@@ -919,6 +920,7 @@ impl Default for AppState {
         let _ = controller.send(CoreCommand::SetInputVolume(config.input_volume));
         let _ = controller.send(CoreCommand::SetOutputVolume(config.output_volume));
         let _ = controller.send(CoreCommand::SetNetworkMode(config.network_mode));
+        let _ = controller.send(CoreCommand::SetAudioProfile(config.audio_profile));
         let _ = controller.send(CoreCommand::SetRecordingMode(config.recording_mode));
         let _ = controller.send(CoreCommand::SetPixelpassPath(config.pixelpass_path.clone()));
         let _ = controller.send(CoreCommand::SetPresenceMode(config.presence_mode));
@@ -2050,6 +2052,12 @@ fn update(state: &mut AppState, message: AppMessage) -> Task<AppMessage> {
             // Applied on the next join, since the endpoint is rebuilt then.
             let _ = state.controller.send(CoreCommand::SetNetworkMode(mode));
         }
+        AppMessage::AudioProfileSelected(profile) => {
+            state.config.audio_profile = profile;
+            state.config.save();
+            // Applies live to the running encoder, and to the next call.
+            let _ = state.controller.send(CoreCommand::SetAudioProfile(profile));
+        }
         AppMessage::RecordingModeSelected(mode) => {
             state.config.recording_mode = mode;
             state.config.save();
@@ -3123,6 +3131,19 @@ fn network_mode_hint(mode: NetworkMode) -> &'static str {
     }
 }
 
+/// One-line explanation of an audio/network profile for the settings picker (W12).
+fn audio_profile_hint(profile: AudioProfile) -> &'static str {
+    match profile {
+        AudioProfile::LowLatency => {
+            "Lowest delay, no loss recovery. Best on a clean LAN or wired link."
+        }
+        AudioProfile::Balanced => "Default: voice quality with light loss recovery.",
+        AudioProfile::BadNetwork => {
+            "Most resilient on a lossy/congested link: heavier loss recovery, lower bitrate."
+        }
+    }
+}
+
 /// One-line explanation of a recording mode for the settings picker.
 fn recording_mode_hint(mode: RecordingMode) -> &'static str {
     match mode {
@@ -4981,6 +5002,17 @@ fn view(state: &AppState) -> Element<'_, AppMessage> {
                         control
                     },
                 ].spacing(8).width(iced::Length::Fill),
+                vertical_space(section_gap),
+                section_header("Connection quality"),
+                column![
+                    pick_list(
+                        &AudioProfile::ALL[..],
+                        Some(state.config.audio_profile),
+                        AppMessage::AudioProfileSelected,
+                    ).width(iced::Length::Fill),
+                    text(audio_profile_hint(state.config.audio_profile)).size(11).color(color_subtext),
+                    text("Applies immediately, even mid-call.").size(11).color(color_subtext),
+                ].spacing(4).width(iced::Length::Fill),
             ]
             .spacing(10)
             .width(iced::Length::Fill)

src/codec/mod.rs

@@ -20,6 +20,9 @@ pub trait AudioDecoder: Send {
     /// If `compressed` is `None` (or `Some(&[])`), it indicates packet loss,
     /// enabling the decoder to perform packet loss concealment (PLC).
     fn decode(&mut self, compressed: Option<&[u8]>) -> Result<Vec<i16>, CodecError>;
+
+    /// Reconstructs the previous lost frame from the next packet's in-band FEC.
+    fn decode_fec(&mut self, next_payload: &[u8]) -> Result<Vec<i16>, CodecError>;
 }
 
 pub mod opus_impl;

src/codec/opus_impl.rs

@@ -1,5 +1,49 @@
 use crate::codec::{AudioDecoder, AudioEncoder, CodecError};
-use opus::{Application, Channels, Decoder, Encoder};
+use crate::config::AudioProfile;
+use opus::{Application, Bitrate, Channels, Decoder, Encoder};
+
+/// Concrete libopus encoder settings derived from an [`AudioProfile`]. Plain
+/// data, so the profile→params mapping ([`opus_params`]) stays a pure,
+/// unit-testable function (W12).
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct OpusParams {
+    /// Target bitrate in bits/sec.
+    pub bitrate: i32,
+    /// Enable in-band forward error correction (loss redundancy in the bitstream).
+    pub inband_fec: bool,
+    /// Expected packet-loss percentage (0..=100); tunes how much FEC libopus adds.
+    pub packet_loss_perc: i32,
+    /// Discontinuous transmission: stop sending during silence to save bandwidth.
+    pub dtx: bool,
+}
+
+/// Map a named profile to concrete Opus parameters. Pure — the W12 testable seam.
+///
+/// `BadNetwork` deliberately runs a *lower* bitrate than `Balanced`: in-band FEC
+/// redundancy is carried inside the same bitstream, so trimming the base bitrate
+/// leaves headroom for the redundancy on a congested link.
+pub fn opus_params(profile: AudioProfile) -> OpusParams {
+    match profile {
+        AudioProfile::LowLatency => OpusParams {
+            bitrate: 24_000,
+            inband_fec: false,
+            packet_loss_perc: 0,
+            dtx: false,
+        },
+        AudioProfile::Balanced => OpusParams {
+            bitrate: 32_000,
+            inband_fec: true,
+            packet_loss_perc: 10,
+            dtx: false,
+        },
+        AudioProfile::BadNetwork => OpusParams {
+            bitrate: 20_000,
+            inband_fec: true,
+            packet_loss_perc: 25,
+            dtx: false,
+        },
+    }
+}
 
 pub struct OpusEncoder {
     encoder: Encoder,
@@ -17,6 +61,29 @@ impl OpusEncoder {
             .map_err(|e| CodecError::Init(format!("Failed to create Opus encoder: {}", e)))?;
         Ok(Self { encoder })
     }
+
+    /// Apply concrete codec parameters to the live encoder. Safe to call between
+    /// frames, so the user can switch profile mid-call.
+    pub fn apply_params(&mut self, params: &OpusParams) -> Result<(), CodecError> {
+        self.encoder
+            .set_bitrate(Bitrate::Bits(params.bitrate))
+            .map_err(|e| CodecError::Init(format!("set_bitrate: {}", e)))?;
+        self.encoder
+            .set_inband_fec(params.inband_fec)
+            .map_err(|e| CodecError::Init(format!("set_inband_fec: {}", e)))?;
+        self.encoder
+            .set_packet_loss_perc(params.packet_loss_perc)
+            .map_err(|e| CodecError::Init(format!("set_packet_loss_perc: {}", e)))?;
+        self.encoder
+            .set_dtx(params.dtx)
+            .map_err(|e| CodecError::Init(format!("set_dtx: {}", e)))?;
+        Ok(())
+    }
+
+    /// Apply a named [`AudioProfile`] (shorthand for `apply_params(&opus_params(p))`).
+    pub fn apply_profile(&mut self, profile: AudioProfile) -> Result<(), CodecError> {
+        self.apply_params(&opus_params(profile))
+    }
 }
 
 impl AudioEncoder for OpusEncoder {
@@ -95,12 +162,64 @@ impl AudioDecoder for OpusDecoder {
         pcm.truncate(decoded_per_channel * channels_count);
         Ok(pcm)
     }
+
+    fn decode_fec(&mut self, next_payload: &[u8]) -> Result<Vec<i16>, CodecError> {
+        let channels_count = self.channels_count();
+        let mut pcm = vec![0i16; self.frame_samples * channels_count];
+
+        let decoded_per_channel = self
+            .decoder
+            .decode(next_payload, &mut pcm, true)
+            .map_err(|e| CodecError::Decode(format!("Opus FEC decoding failed: {}", e)))?;
+
+        pcm.truncate(decoded_per_channel * channels_count);
+        Ok(pcm)
+    }
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
 
+    #[test]
+    fn test_opus_params_mapping() {
+        let low = opus_params(AudioProfile::LowLatency);
+        let bal = opus_params(AudioProfile::Balanced);
+        let bad = opus_params(AudioProfile::BadNetwork);
+
+        // LowLatency has no loss redundancy; the other two do.
+        assert!(!low.inband_fec);
+        assert_eq!(low.packet_loss_perc, 0);
+        assert!(bal.inband_fec);
+        assert!(bad.inband_fec);
+
+        // Capture-side gating suppresses silence; no profile adds Opus DTX.
+        assert!(!low.dtx && !bal.dtx && !bad.dtx);
+        assert!(bad.packet_loss_perc > bal.packet_loss_perc);
+
+        // BadNetwork trims base bitrate to make room for FEC redundancy.
+        assert!(bad.bitrate < bal.bitrate);
+
+        // All bitrates are sane positive voice rates.
+        for p in [low, bal, bad] {
+            assert!(p.bitrate > 0 && p.bitrate <= 64_000);
+            assert!((0..=100).contains(&p.packet_loss_perc));
+        }
+    }
+
+    #[test]
+    fn test_apply_profile_sets_bitrate() {
+        let mut encoder = OpusEncoder::new(48000, Channels::Mono, Application::Voip).unwrap();
+        // Every profile applies cleanly to a real encoder...
+        for profile in AudioProfile::ALL {
+            encoder.apply_profile(profile).unwrap();
+        }
+        // ...and the last-applied bitrate is reflected by the encoder.
+        encoder.apply_profile(AudioProfile::Balanced).unwrap();
+        let want = opus_params(AudioProfile::Balanced).bitrate;
+        assert_eq!(encoder.encoder.get_bitrate().unwrap(), Bitrate::Bits(want));
+    }
+
     #[test]
     fn test_round_trip() {
         let mut encoder = OpusEncoder::new(48000, Channels::Mono, Application::Voip).unwrap();

src/config.rs

@@ -93,6 +93,60 @@ impl std::fmt::Display for RecordingMode {
     }
 }
 
+/// Named Opus encoder / network-resilience policy (W12). The user picks a
+/// profile instead of raw codec knobs; the concrete libopus parameters live in
+/// `codec::opus_impl::opus_params`. Applies live to the running encoder.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Default)]
+pub enum AudioProfile {
+    /// Lowest mouth-to-ear delay: modest bitrate, no FEC redundancy. Best on a
+    /// clean LAN / low-loss link where added latency matters more than loss.
+    LowLatency,
+    /// Sensible default: voice bitrate with in-band FEC for light packet loss.
+    #[default]
+    Balanced,
+    /// Maximum resilience on a lossy/congested link: in-band FEC tuned for heavy
+    /// loss, at a lower bitrate to leave headroom for the redundancy.
+    BadNetwork,
+}
+
+impl AudioProfile {
+    /// All variants, in picker display order.
+    pub const ALL: [AudioProfile; 3] = [
+        AudioProfile::LowLatency,
+        AudioProfile::Balanced,
+        AudioProfile::BadNetwork,
+    ];
+
+    /// Compact discriminant for handing the profile to the capture thread via an
+    /// atomic. Pairs with [`AudioProfile::from_u8`].
+    pub fn as_u8(self) -> u8 {
+        match self {
+            AudioProfile::LowLatency => 0,
+            AudioProfile::Balanced => 1,
+            AudioProfile::BadNetwork => 2,
+        }
+    }
+
+    /// Inverse of [`AudioProfile::as_u8`]; unknown values fall back to the default.
+    pub fn from_u8(v: u8) -> AudioProfile {
+        match v {
+            0 => AudioProfile::LowLatency,
+            2 => AudioProfile::BadNetwork,
+            _ => AudioProfile::Balanced,
+        }
+    }
+}
+
+impl std::fmt::Display for AudioProfile {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.write_str(match self {
+            AudioProfile::LowLatency => "Low latency",
+            AudioProfile::Balanced => "Balanced",
+            AudioProfile::BadNetwork => "Bad network",
+        })
+    }
+}
+
 impl std::fmt::Display for RoomLayout {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         f.write_str(match self {
@@ -199,6 +253,10 @@ pub struct AppConfig {
     pub clip_volume_universal: bool,
     #[serde(default)]
     pub network_mode: NetworkMode,
+    /// Opus encoder / network-resilience profile (W12). Applies live to the
+    /// running encoder; default `Balanced`.
+    #[serde(default)]
+    pub audio_profile: AudioProfile,
     /// Presence posture for the friends idle listener (W7): invisible / normal /
     /// discoverable. Default `Normal` = answer friends only, no DNS beacon.
     #[serde(default)]
@@ -380,6 +438,7 @@ impl Default for AppConfig {
             show_player_bar: true,
             clip_volume_universal: true,
             network_mode: NetworkMode::default(),
+            audio_profile: AudioProfile::default(),
             presence_mode: crate::presence::PresenceMode::default(),
             echo_cancellation_enabled: false,
             notifications_enabled: true,
@@ -1019,6 +1078,38 @@ mod tests {
         assert_ne!(display_0, display_2);
     }
 
+    #[test]
+    fn test_audio_profile() {
+        // Default is Balanced.
+        assert_eq!(AudioProfile::default(), AudioProfile::Balanced);
+
+        // ALL holds the three variants.
+        assert_eq!(AudioProfile::ALL.len(), 3);
+        assert!(AudioProfile::ALL.contains(&AudioProfile::LowLatency));
+        assert!(AudioProfile::ALL.contains(&AudioProfile::Balanced));
+        assert!(AudioProfile::ALL.contains(&AudioProfile::BadNetwork));
+
+        // as_u8 / from_u8 round-trip every variant, and unknown bytes fall back
+        // to the default rather than panicking.
+        for p in AudioProfile::ALL {
+            assert_eq!(AudioProfile::from_u8(p.as_u8()), p);
+        }
+        assert_eq!(AudioProfile::from_u8(99), AudioProfile::Balanced);
+
+        // serde round-trips, and Display strings are non-empty + distinct.
+        let mut labels = Vec::new();
+        for p in AudioProfile::ALL {
+            let s = serde_json::to_string(&p).unwrap();
+            assert_eq!(serde_json::from_str::<AudioProfile>(&s).unwrap(), p);
+            let label = p.to_string();
+            assert!(!label.is_empty());
+            labels.push(label);
+        }
+        labels.sort();
+        labels.dedup();
+        assert_eq!(labels.len(), 3);
+    }
+
     #[test]
     fn test_unknown_field_tolerance() {
         // Unknown/extra field tolerance: a config JSON containing an extra unrecognized key should still deserialize.

src/core/jitter.rs

@@ -202,11 +202,15 @@ impl JitterBuffer {
                     None
                 } else {
                     // Gap with later packets already buffered: a packet was lost
-                    // or reordered out of window. Conceal this frame via Opus PLC
-                    // and grow the cushion — the jitter beat our current delay.
+                    // or reordered out of window. First try Opus in-band FEC from
+                    // the next packet; if unavailable, fall back to plain PLC.
                     self.next_seq = Some(next.wrapping_add(1));
                     self.note_disruption();
-                    self.decoder.decode(None).ok()
+                    let next_payload = self.packets.values().next().expect("non-empty");
+                    self.decoder
+                        .decode_fec(next_payload)
+                        .or_else(|_| self.decoder.decode(None))
+                        .ok()
                 }
             }
         }
@@ -221,8 +225,8 @@ impl JitterBuffer {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::codec::AudioEncoder;
-    use crate::codec::opus_impl::OpusEncoder;
+    use crate::codec::opus_impl::{OpusDecoder, OpusEncoder, OpusParams};
+    use crate::codec::{AudioDecoder, AudioEncoder};
     use opus::{Application, Channels};
 
     /// A real, decodable Opus packet for one 20ms mono frame at amplitude `amp`.
@@ -233,6 +237,32 @@ mod tests {
         enc.encode(&pcm).unwrap()
     }
 
+    fn tone_frame(enc: &mut OpusEncoder, amp: i16, frame_index: usize) -> Vec<u8> {
+        let pcm: Vec<i16> = (0..FRAME_SAMPLES)
+            .map(|i| {
+                let sample_index = frame_index * FRAME_SAMPLES + i;
+                let t = sample_index as f32 / 48_000.0;
+                let fundamental = (t * 220.0 * 2.0 * std::f32::consts::PI).sin();
+                let harmonic = (t * 440.0 * 2.0 * std::f32::consts::PI).sin();
+                ((fundamental * 0.7 + harmonic * 0.3) * amp as f32) as i16
+            })
+            .collect();
+        enc.encode(&pcm).unwrap()
+    }
+
+    fn rms_error(a: &[i16], b: &[i16]) -> f64 {
+        assert_eq!(a.len(), b.len());
+        let sum_sq: f64 = a
+            .iter()
+            .zip(b)
+            .map(|(&left, &right)| {
+                let diff = left as f64 - right as f64;
+                diff * diff
+            })
+            .sum();
+        (sum_sq / a.len() as f64).sqrt()
+    }
+
     #[test]
     fn buffers_then_plays_in_order() {
         let mut enc = OpusEncoder::new(48000, Channels::Mono, Application::Voip).unwrap();
@@ -289,6 +319,66 @@ mod tests {
         assert!(jb.pop_frame().is_none());
     }
 
+    #[test]
+    fn uses_in_band_fec_from_next_packet_for_gap() {
+        let mut enc = OpusEncoder::new(48000, Channels::Mono, Application::Voip).unwrap();
+        enc.apply_params(&OpusParams {
+            bitrate: 20_000,
+            inband_fec: true,
+            packet_loss_perc: 60,
+            dtx: false,
+        })
+        .unwrap();
+
+        let dropped_seq = 5usize;
+        let amps = [1800, 1800, 1800, 1800, 1800, 12_000, 12_000, 12_000];
+        let packets: Vec<Vec<u8>> = amps
+            .into_iter()
+            .enumerate()
+            .map(|(seq, amp)| tone_frame(&mut enc, amp, seq))
+            .collect();
+
+        let mut expected_decoder = OpusDecoder::new(48000, Channels::Mono, FRAME_SAMPLES).unwrap();
+        for packet in packets.iter().take(dropped_seq) {
+            expected_decoder.decode(Some(packet)).unwrap();
+        }
+        let expected_lost = expected_decoder
+            .decode(Some(&packets[dropped_seq]))
+            .unwrap();
+
+        let mut plc_decoder = OpusDecoder::new(48000, Channels::Mono, FRAME_SAMPLES).unwrap();
+        for packet in packets.iter().take(dropped_seq) {
+            plc_decoder.decode(Some(packet)).unwrap();
+        }
+        let pure_plc = plc_decoder.decode(None).unwrap();
+
+        let mut jb = JitterBuffer::new().unwrap();
+        for (seq, packet) in packets.iter().enumerate() {
+            if seq != dropped_seq {
+                jb.insert(seq as u32, packet.clone());
+            }
+        }
+
+        for _ in 0..dropped_seq {
+            assert_eq!(jb.pop_frame().map(|frame| frame.len()), Some(FRAME_SAMPLES));
+        }
+
+        let recovered = jb.pop_frame().expect("gap should be reconstructed");
+        assert_eq!(recovered.len(), FRAME_SAMPLES);
+        assert!(
+            jb.packets.contains_key(&(dropped_seq as u32 + 1)),
+            "FEC source packet must remain buffered for normal decode"
+        );
+        assert_eq!(jb.pop_frame().map(|frame| frame.len()), Some(FRAME_SAMPLES));
+
+        let fec_error = rms_error(&recovered, &expected_lost);
+        let plc_error = rms_error(&pure_plc, &expected_lost);
+        assert!(
+            fec_error < plc_error * 0.75,
+            "FEC reconstruction should be materially closer than PLC (fec_error={fec_error}, plc_error={plc_error})"
+        );
+    }
+
     #[test]
     fn drops_packets_already_played() {
         let mut enc = OpusEncoder::new(48000, Channels::Mono, Application::Voip).unwrap();

src/core/messages.rs

@@ -1,4 +1,4 @@
-use crate::config::{NetworkMode, RecordingMode};
+use crate::config::{AudioProfile, NetworkMode, RecordingMode};
 use crate::friends::Friend;
 use crate::network::PeerState;
 use crate::presence::{FriendPresence, PresenceMode};
@@ -56,6 +56,10 @@ pub enum CoreCommand {
     /// Set the relay/discovery posture. Takes effect on the next room join,
     /// since the endpoint is (re)built then.
     SetNetworkMode(NetworkMode),
+    /// Set the Opus encoder / network-resilience profile (W12). Applies live to
+    /// the running capture encoder, and to the next call's encoder. Sent at
+    /// startup from config and whenever the user changes it.
+    SetAudioProfile(AudioProfile),
     /// Start/stop recording the call to a local WAV (your mic + the incoming
     /// mix). No-op start if already recording / not in a call.
     SetRecording(bool),
@@ -212,6 +216,7 @@ pub fn delivery_class(cmd: &CoreCommand) -> DeliveryClass {
             input_device: _,
         }
         | CoreCommand::SetNetworkMode(_)
+        | CoreCommand::SetAudioProfile(_)
         | CoreCommand::SetRecording(_)
         | CoreCommand::SetRecordingMode(_)
         | CoreCommand::SendChat(_)
@@ -292,6 +297,7 @@ pub fn coalesce_key(cmd: &CoreCommand) -> Option<CoalesceKey> {
             input_device: _,
         }
         | CoreCommand::SetNetworkMode(_)
+        | CoreCommand::SetAudioProfile(_)
         | CoreCommand::SetRecording(_)
         | CoreCommand::SetRecordingMode(_)
         | CoreCommand::SendChat(_)
@@ -574,6 +580,7 @@ mod tests {
             },
             CoreCommand::SetPeerMuted(peer, true),
             CoreCommand::SetPresenceMode(PresenceMode::Normal),
+            CoreCommand::SetAudioProfile(crate::config::AudioProfile::BadNetwork),
             CoreCommand::SendChat("hello".to_string()),
         ];
 

src/core/mod.rs

@@ -17,7 +17,7 @@ use crate::network::{
 };
 
 use crate::audio::multitrack::MultitrackRecorder;
-use crate::config::{NetworkMode, RecordingMode};
+use crate::config::{AudioProfile, NetworkMode, RecordingMode};
 use crate::presence::PresenceMode;
 use iroh::{
     Endpoint, EndpointAddr, EndpointId, RelayMode, SecretKey, endpoint::presets, protocol::Router,
@@ -31,6 +31,12 @@ use tokio::sync::{Mutex, mpsc};
 
 type CoalesceStore = Arc<StdMutex<HashMap<CoalesceKey, CoreCommand>>>;
 
+// Mixer -> playback-worker handoff. The playback ring itself targets three
+// 20ms frames; allow at most two more in flight so worker lag applies
+// backpressure before the ring can overshoot to its 200ms cap (A6).
+const PLAYBACK_HANDOFF_QUEUE_FRAMES: usize = 2;
+const PLAYBACK_HANDOFF_RETRY: Duration = Duration::from_millis(1);
+
 pub struct CoreController {
     reliable_tx: mpsc::UnboundedSender<CoreCommand>,
     coalesce: CoalesceStore,
@@ -156,6 +162,22 @@ fn audio_datagram_len_ok(len: usize) -> bool {
     (4..=4 + MAX_OPUS_PAYLOAD).contains(&len)
 }
 
+async fn send_playback_frame(
+    tx: &std::sync::mpsc::SyncSender<Vec<i16>>,
+    mut frame: Vec<i16>,
+) -> bool {
+    loop {
+        match tx.try_send(frame) {
+            Ok(()) => return true,
+            Err(std::sync::mpsc::TrySendError::Full(returned)) => {
+                frame = returned;
+                tokio::time::sleep(PLAYBACK_HANDOFF_RETRY).await;
+            }
+            Err(std::sync::mpsc::TrySendError::Disconnected(_)) => return false,
+        }
+    }
+}
+
 /// The presence label to broadcast for a detected game: its display name,
 /// sanitized + length-capped, or `None` when there's no game or no broadcastable
 /// name (a Steam appid without a manifest name, or a label that sanitizes empty).
@@ -1162,6 +1184,12 @@ async fn run_core_loop(
     // App-internal capture/playback gains (f32 bits), live-read by the audio loops.
     let input_gain = Arc::new(std::sync::atomic::AtomicU32::new(1.0f32.to_bits()));
     let output_gain = Arc::new(std::sync::atomic::AtomicU32::new(1.0f32.to_bits()));
+    // Opus encoder profile (W12) as a discriminant, live-read by the capture
+    // thread so a mid-call profile switch re-tunes the running encoder. Set from
+    // config via the GUI's startup `SetAudioProfile`; defaults to Balanced.
+    let audio_profile = Arc::new(std::sync::atomic::AtomicU8::new(
+        AudioProfile::default().as_u8(),
+    ));
     // Call recording: an optional live recorder (mic FIFO + WAV writer), shared
     // by the capture thread (pushes mic) and the mixer task (writes mix frames).
     // `is_recording` is a fast-path gate so the audio loops only take the lock
@@ -1648,7 +1676,8 @@ async fn run_core_loop(
 
                 // Setup raw audio channels
                 let (capture_tx, capture_rx) = std::sync::mpsc::channel();
-                let (playback_tx, playback_rx) = std::sync::mpsc::channel();
+                let (playback_tx, playback_rx) =
+                    std::sync::mpsc::sync_channel(PLAYBACK_HANDOFF_QUEUE_FRAMES);
 
                 // Echo cancellation: if enabled, load PipeWire's echo-cancel module
                 // bound to the chosen real devices and route capture/playback
@@ -1740,6 +1769,7 @@ async fn run_core_loop(
                 let is_recording_capture = is_recording.clone();
                 let multitrack_capture = multitrack.clone();
                 let is_multitrack_capture = is_multitrack.clone();
+                let audio_profile_capture = audio_profile.clone();
 
                 let capture_thread = std::thread::spawn(move || {
                     use opus::{Application, Channels};
@@ -1751,6 +1781,13 @@ async fn run_core_loop(
                                 return;
                             }
                         };
+                    // Tune the encoder to the configured profile (W12), then track
+                    // the live discriminant so a mid-call switch re-applies it.
+                    let mut current_profile =
+                        AudioProfile::from_u8(audio_profile_capture.load(Ordering::Relaxed));
+                    if let Err(e) = encoder.apply_profile(current_profile) {
+                        crate::log_msg(&format!("Opus profile apply failed: {:?}", e));
+                    }
                     // Per-sender packet sequence number, prepended to every frame so
                     // receivers can reorder and conceal loss. Wraps after ~years.
                     let mut seq: u32 = 0;
@@ -1763,6 +1800,14 @@ async fn run_core_loop(
                     let mut mic_meter = MicLevelMeter::new();
 
                     while let Ok(mut pcm) = capture_rx.recv() {
+                        // Re-tune the encoder if the user switched profile mid-call.
+                        // Cheap atomic load per frame; only reconfigures on change.
+                        let want =
+                            AudioProfile::from_u8(audio_profile_capture.load(Ordering::Relaxed));
+                        if want != current_profile && encoder.apply_profile(want).is_ok() {
+                            current_profile = want;
+                        }
+
                         // Apply the input gain first so the meter, gate, and what we
                         // transmit all reflect the same (gained) signal.
                         apply_volume(
@@ -2092,7 +2137,7 @@ async fn run_core_loop(
                             mixed
                         };
 
-                        if playback_tx.send(frame_to_send).is_err() {
+                        if !send_playback_frame(&playback_tx, frame_to_send).await {
                             break;
                         }
 
@@ -2682,6 +2727,13 @@ async fn run_core_loop(
                 }
             }
 
+            CoreCommand::SetAudioProfile(profile) => {
+                // Publish the new profile to the capture thread (W12). It picks up
+                // the change on its next frame and re-tunes the live encoder; a
+                // call that starts later reads the same atomic at encoder creation.
+                audio_profile.store(profile.as_u8(), Ordering::Relaxed);
+            }
+
             CoreCommand::RegenerateIdentity => {
                 // Mint + persist a fresh identity, discarding the old one. The
                 // persistent endpoint is rebuilt with the new key (now if idle, else
@@ -3193,12 +3245,15 @@ async fn run_core_loop(
 mod tests {
     use super::{
         KnownPeers, MAX_OPUS_PAYLOAD, MAX_RETAINED_PEERS, MIC_LEVEL_REPORT_SAMPLES, MicLevelMeter,
-        PeerSpeakTicket, admit_retained, apply_peer_volume, apply_volume, audio_datagram_len_ok,
-        coalesce_insert, coalesce_pop, frame_level, mix_frames, mix_stereo_frames,
-        next_game_change, should_auto_fetch, stereo_to_mono,
+        PLAYBACK_HANDOFF_QUEUE_FRAMES, PeerSpeakTicket, admit_retained, apply_peer_volume,
+        apply_volume, audio_datagram_len_ok, coalesce_insert, coalesce_pop, frame_level,
+        mix_frames, mix_stereo_frames, next_game_change, send_playback_frame, should_auto_fetch,
+        stereo_to_mono,
     };
     use crate::core::messages::{CoalesceKey, CoreCommand, coalesce_key};
     use std::collections::{HashMap, HashSet};
+    use std::sync::mpsc::sync_channel;
+    use std::time::Duration;
 
     fn endpoint_id() -> iroh::EndpointId {
         iroh::SecretKey::generate().public()
@@ -3432,6 +3487,33 @@ mod tests {
         assert!(!audio_datagram_len_ok(5 + MAX_OPUS_PAYLOAD));
     }
 
+    #[tokio::test]
+    async fn playback_handoff_waits_for_bounded_queue_space() {
+        let (tx, rx) = sync_channel::<Vec<i16>>(PLAYBACK_HANDOFF_QUEUE_FRAMES);
+        for n in 0..PLAYBACK_HANDOFF_QUEUE_FRAMES {
+            tx.try_send(vec![n as i16]).unwrap();
+        }
+
+        let worker = std::thread::spawn(move || {
+            std::thread::sleep(Duration::from_millis(20));
+            for n in 0..PLAYBACK_HANDOFF_QUEUE_FRAMES {
+                assert_eq!(rx.recv().unwrap(), vec![n as i16]);
+            }
+            assert_eq!(rx.recv().unwrap(), vec![99, 100]);
+        });
+
+        let sent = tokio::time::timeout(
+            Duration::from_secs(1),
+            send_playback_frame(&tx, vec![99, 100]),
+        )
+        .await
+        .expect("bounded handoff should unblock after the worker drains a frame");
+
+        assert!(sent);
+        drop(tx);
+        worker.join().unwrap();
+    }
+
     #[test]
     fn mic_meter_holds_the_peak_across_the_window() {
         let mut m = MicLevelMeter::new();

src/game/scan.rs

@@ -97,14 +97,12 @@ fn windows_toolhelp_executables() -> Vec<String> {
 
 #[cfg(test)]
 mod tests {
-    use super::*;
-
     #[cfg(target_os = "linux")]
     #[test]
     fn enumerates_at_least_this_process() {
         // The test runner itself is a process, so /proc enumeration must be
         // non-empty and include something that normalizes to our own exe basename.
-        let exes = running_executables();
+        let exes = super::running_executables();
         assert!(
             !exes.is_empty(),
             "expected to see running processes via /proc"