feat: direct calling UI for desktop Tauri app + merge android branch

Tauri backend:
- register_signal: persistent _signal connection, presence registration
- place_call: send DirectCallOffer by fingerprint
- answer_call: accept/reject incoming calls
- get_signal_status: poll signal state

Frontend:
- Mode toggle: "Room" vs "Direct Call"
- Register button → registers on relay signal channel
- Incoming call panel with Accept/Reject
- Fingerprint input + Call button
- Auto-connect to media room on CallSetup event

Also merges feat/android-voip-client into desktop branch:
- Federation fixes, time-based dedup, FEC stale blocks
- Direct calling protocol types
- ACL + SAS verification

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/wzp-android/Cargo.toml

@@ -17,7 +17,7 @@ wzp-crypto = { workspace = true }
 wzp-transport = { workspace = true }
 tokio = { workspace = true }
 tracing = { workspace = true }
-tracing-subscriber = { workspace = true }
+tracing-subscriber = { workspace = true, features = ["env-filter"] }
 bytes = { workspace = true }
 serde = { workspace = true }
 serde_json = "1"

crates/wzp-android/src/audio_ring.rs

@@ -1,91 +1,128 @@
-//! Lock-free SPSC ring buffers for audio PCM transfer between
-//! Kotlin AudioRecord/AudioTrack threads and the Rust engine.
+//! Lock-free SPSC ring buffer — "Reader-Detects-Lap" architecture.
 //!
-//! These use a simple spin-free design: the producer writes and advances
-//! a write cursor, the consumer reads and advances a read cursor.
-//! Both cursors are atomic so no mutex is needed.
+//! SPSC invariant: the producer ONLY writes `write_pos`, the consumer
+//! ONLY writes `read_pos`.  Neither thread touches the other's cursor.
+//!
+//! On overflow (writer laps the reader), the writer simply overwrites
+//! old buffer data.  The reader detects the lap via `available() >
+//! RING_CAPACITY` and snaps its own `read_pos` forward.
+//!
+//! Capacity is a power of 2 for bitmask indexing (no modulo).
 
-use std::sync::atomic::{AtomicUsize, Ordering};
+use std::sync::atomic::{AtomicU64, AtomicUsize, Ordering};
 
-/// Ring buffer capacity in i16 samples.
-/// 960 samples * 10 frames = ~200ms of audio at 48kHz mono.
-const RING_CAPACITY: usize = 960 * 10;
+/// Ring buffer capacity — power of 2 for bitmask indexing.
+/// 16384 samples = 341.3ms at 48kHz mono.  70% more headroom
+/// than the previous 9600 (200ms) for surviving Android GC pauses.
+const RING_CAPACITY: usize = 16384; // 2^14
+const RING_MASK: usize = RING_CAPACITY - 1;
 
 /// Lock-free single-producer single-consumer ring buffer for i16 PCM samples.
 pub struct AudioRing {
-    buf: Box<[i16; RING_CAPACITY]>,
+    buf: Box<[i16]>,
+    /// Monotonically increasing write cursor. ONLY written by producer.
     write_pos: AtomicUsize,
+    /// Monotonically increasing read cursor. ONLY written by consumer.
     read_pos: AtomicUsize,
+    /// Incremented by reader when it detects it was lapped (overflow).
+    overflow_count: AtomicU64,
+    /// Incremented by reader when ring is empty (underrun).
+    underrun_count: AtomicU64,
 }
 
-// SAFETY: AudioRing is designed for SPSC — one thread writes, one reads.
-// The atomics ensure visibility. The buffer itself is never accessed
-// from the same index by both threads simultaneously because the
-// producer only writes to positions between write_pos and read_pos,
-// and the consumer only reads from positions between read_pos and write_pos.
+// SAFETY: AudioRing is SPSC — one thread writes (producer), one reads (consumer).
+// The producer only writes write_pos. The consumer only writes read_pos.
+// Neither thread writes the other's cursor. Buffer indices are derived from
+// the owning thread's cursor, ensuring no concurrent access to the same index.
 unsafe impl Send for AudioRing {}
 unsafe impl Sync for AudioRing {}
 
 impl AudioRing {
     pub fn new() -> Self {
+        debug_assert!(RING_CAPACITY.is_power_of_two());
         Self {
-            buf: Box::new([0i16; RING_CAPACITY]),
+            buf: vec![0i16; RING_CAPACITY].into_boxed_slice(),
             write_pos: AtomicUsize::new(0),
             read_pos: AtomicUsize::new(0),
+            overflow_count: AtomicU64::new(0),
+            underrun_count: AtomicU64::new(0),
         }
     }
 
-    /// Number of samples available to read.
+    /// Number of samples available to read (clamped to capacity).
     pub fn available(&self) -> usize {
         let w = self.write_pos.load(Ordering::Acquire);
-        let r = self.read_pos.load(Ordering::Acquire);
-        w.wrapping_sub(r)
+        let r = self.read_pos.load(Ordering::Relaxed);
+        w.wrapping_sub(r).min(RING_CAPACITY)
     }
 
-    /// Number of samples that can be written without overwriting.
+    /// Number of samples that can be written without overwriting unread data.
     pub fn free_space(&self) -> usize {
-        RING_CAPACITY - self.available()
+        RING_CAPACITY.saturating_sub(self.available())
     }
 
     /// Write samples into the ring. Returns number of samples written.
-    /// Drops oldest samples if the ring is full.
+    ///
+    /// If the ring is full, old data is silently overwritten.  The reader
+    /// will detect the lap and self-correct.  The writer NEVER touches
+    /// `read_pos` — this is the key invariant that prevents cursor desync.
     pub fn write(&self, samples: &[i16]) -> usize {
-        let w = self.write_pos.load(Ordering::Relaxed);
         let count = samples.len().min(RING_CAPACITY);
+        let w = self.write_pos.load(Ordering::Relaxed);
 
         for i in 0..count {
-            let idx = (w + i) % RING_CAPACITY;
-            // SAFETY: We're the only writer, and the reader won't read
-            // past read_pos which we haven't advanced past yet.
             unsafe {
                 let ptr = self.buf.as_ptr() as *mut i16;
-                *ptr.add(idx) = samples[i];
+                *ptr.add((w + i) & RING_MASK) = samples[i];
             }
         }
 
         self.write_pos.store(w.wrapping_add(count), Ordering::Release);
-
-        // If we overwrote unread data, advance read_pos
-        if self.available() > RING_CAPACITY {
-            let new_read = self.write_pos.load(Ordering::Relaxed).wrapping_sub(RING_CAPACITY);
-            self.read_pos.store(new_read, Ordering::Release);
-        }
-
         count
     }
 
     /// Read samples from the ring into `out`. Returns number of samples read.
+    ///
+    /// If the writer has lapped the reader (overflow), `read_pos` is snapped
+    /// forward to the oldest valid data.  This is safe because only the
+    /// reader thread writes `read_pos`.
     pub fn read(&self, out: &mut [i16]) -> usize {
-        let avail = self.available();
-        let count = out.len().min(avail);
+        let w = self.write_pos.load(Ordering::Acquire);
+        let mut r = self.read_pos.load(Ordering::Relaxed);
+
+        let mut avail = w.wrapping_sub(r);
+
+        // Lap detection: writer has overwritten our unread data.
+        // Snap read_pos forward to oldest valid data in the buffer.
+        if avail > RING_CAPACITY {
+            r = w.wrapping_sub(RING_CAPACITY);
+            avail = RING_CAPACITY;
+            self.overflow_count.fetch_add(1, Ordering::Relaxed);
+        }
+
+        let count = out.len().min(avail);
+        if count == 0 {
+            if w == r {
+                self.underrun_count.fetch_add(1, Ordering::Relaxed);
+            }
+            return 0;
+        }
 
-        let r = self.read_pos.load(Ordering::Relaxed);
         for i in 0..count {
-            let idx = (r + i) % RING_CAPACITY;
-            out[i] = unsafe { *self.buf.as_ptr().add(idx) };
+            out[i] = unsafe { *self.buf.as_ptr().add((r + i) & RING_MASK) };
         }
 
         self.read_pos.store(r.wrapping_add(count), Ordering::Release);
         count
     }
+
+    /// Number of overflow events (reader was lapped by writer).
+    pub fn overflow_count(&self) -> u64 {
+        self.overflow_count.load(Ordering::Relaxed)
+    }
+
+    /// Number of underrun events (reader found empty buffer).
+    pub fn underrun_count(&self) -> u64 {
+        self.underrun_count.load(Ordering::Relaxed)
+    }
 }

crates/wzp-android/src/commands.rs

@@ -12,4 +12,13 @@ pub enum EngineCommand {
     ForceProfile(QualityProfile),
     /// Stop the call and shut down the engine.
     Stop,
+    /// Place a direct call to a fingerprint (requires signal connection).
+    PlaceCall { target_fingerprint: String },
+    /// Answer an incoming direct call.
+    AnswerCall {
+        call_id: String,
+        accept_mode: wzp_proto::CallAcceptMode,
+    },
+    /// Reject an incoming direct call.
+    RejectCall { call_id: String },
 }

crates/wzp-android/src/engine.rs

@@ -9,32 +9,58 @@
 //! and AudioTrack. PCM samples are transferred through lock-free ring buffers.
 
 use std::net::SocketAddr;
-use std::sync::atomic::{AtomicBool, AtomicU16, AtomicU32, Ordering};
+use std::sync::atomic::{AtomicBool, AtomicU8, AtomicU16, AtomicU32, Ordering};
 use std::sync::{Arc, Mutex};
 use std::time::Instant;
 
 use bytes::Bytes;
 use tracing::{error, info, warn};
 use wzp_codec::agc::AutoGainControl;
-use wzp_codec::opus_dec::OpusDecoder;
-use wzp_codec::opus_enc::OpusEncoder;
 use wzp_crypto::{KeyExchange, WarzoneKeyExchange};
 use wzp_fec::{RaptorQFecDecoder, RaptorQFecEncoder};
 use wzp_proto::{
-    AudioDecoder, AudioEncoder, CodecId, FecDecoder, FecEncoder,
-    MediaHeader, MediaPacket, MediaTransport, QualityProfile, SignalMessage,
+    AdaptiveQualityController, AudioDecoder, AudioEncoder, CodecId, FecDecoder, FecEncoder,
+    MediaHeader, MediaPacket, MediaTransport, QualityController, QualityProfile, SignalMessage,
 };
 
 use crate::audio_ring::AudioRing;
 use crate::commands::EngineCommand;
 use crate::stats::{CallState, CallStats};
 
-/// Opus frame size at 48kHz mono, 20ms = 960 samples.
-const FRAME_SAMPLES: usize = 960;
+/// Max frame size at 48kHz mono (40ms = 1920 samples, for Codec2/Opus6k).
+const MAX_FRAME_SAMPLES: usize = 1920;
+
+/// Sentinel value: no profile change pending.
+const PROFILE_NO_CHANGE: u8 = 0xFF;
+
+/// All quality profiles in index order, for AtomicU8-based signaling.
+const PROFILES: [QualityProfile; 6] = [
+    QualityProfile::STUDIO_64K,   // 0
+    QualityProfile::STUDIO_48K,   // 1
+    QualityProfile::STUDIO_32K,   // 2
+    QualityProfile::GOOD,         // 3
+    QualityProfile::DEGRADED,     // 4
+    QualityProfile::CATASTROPHIC, // 5
+];
+
+fn profile_to_index(p: &QualityProfile) -> u8 {
+    PROFILES.iter().position(|pp| pp.codec == p.codec).map(|i| i as u8).unwrap_or(3)
+}
+
+fn index_to_profile(idx: u8) -> Option<QualityProfile> {
+    PROFILES.get(idx as usize).copied()
+}
+
+/// Compute frame samples at 48kHz for a given profile.
+fn frame_samples_for(profile: &QualityProfile) -> usize {
+    (profile.frame_duration_ms as usize) * 48 // 48000 / 1000
+}
 
 /// Configuration to start a call.
 pub struct CallStartConfig {
     pub profile: QualityProfile,
+    /// When true, use the relay's chosen_profile from CallAnswer instead of local profile.
+    pub auto_profile: bool,
     pub relay_addr: String,
     pub room: String,
     pub auth_token: Vec<u8>,
@@ -46,6 +72,7 @@ impl Default for CallStartConfig {
     fn default() -> Self {
         Self {
             profile: QualityProfile::GOOD,
+            auto_profile: false,
             relay_addr: String::new(),
             room: String::new(),
             auth_token: Vec::new(),
@@ -123,6 +150,7 @@ impl WzpEngine {
         let room = config.room.clone();
         let identity_seed = config.identity_seed;
         let profile = config.profile;
+        let auto_profile = config.auto_profile;
         let alias = config.alias.clone();
         let state = self.state.clone();
 
@@ -131,7 +159,7 @@ impl WzpEngine {
 
         let state_clone = state.clone();
         runtime.block_on(async move {
-            if let Err(e) = run_call(relay_addr, &room, &identity_seed, profile, alias.as_deref(), state_clone).await
+            if let Err(e) = run_call(relay_addr, &room, &identity_seed, profile, auto_profile, alias.as_deref(), state_clone).await
             {
                 error!("call failed: {e}");
             }
@@ -169,6 +197,203 @@ impl WzpEngine {
         info!("stop_call: done");
     }
 
+    /// Ping a relay — same pattern as start_call (creates runtime on calling thread).
+    /// Returns JSON `{"rtt_ms":N,"server_fingerprint":"hex"}` or error.
+    pub fn ping_relay(&self, address: &str) -> Result<String, anyhow::Error> {
+        let addr: SocketAddr = address.parse()?;
+        let _ = rustls::crypto::ring::default_provider().install_default();
+
+        let rt = tokio::runtime::Builder::new_current_thread()
+            .enable_all()
+            .build()?;
+
+        let result = rt.block_on(async {
+            let bind: SocketAddr = "0.0.0.0:0".parse().unwrap();
+            let endpoint = wzp_transport::create_endpoint(bind, None)?;
+            let client_cfg = wzp_transport::client_config();
+            let start = Instant::now();
+
+            let conn_result = tokio::time::timeout(
+                std::time::Duration::from_secs(3),
+                wzp_transport::connect(&endpoint, addr, "ping", client_cfg),
+            )
+            .await;
+
+            // Always close endpoint to prevent resource leaks
+            endpoint.close(0u32.into(), b"done");
+
+            let conn = conn_result.map_err(|_| anyhow::anyhow!("timeout"))??;
+            let rtt_ms = start.elapsed().as_millis() as u64;
+            let server_fp = conn
+                .peer_identity()
+                .and_then(|id| id.downcast::<Vec<rustls::pki_types::CertificateDer>>().ok())
+                .and_then(|certs| certs.first().map(|c| {
+                    use std::hash::{Hash, Hasher};
+                    let mut h = std::collections::hash_map::DefaultHasher::new();
+                    c.as_ref().hash(&mut h);
+                    format!("{:016x}", h.finish())
+                }))
+                .unwrap_or_default();
+            conn.close(0u32.into(), b"ping");
+
+            Ok::<_, anyhow::Error>(format!(r#"{{"rtt_ms":{},"server_fingerprint":"{}"}}"#, rtt_ms, server_fp))
+        });
+
+        // Shutdown runtime cleanly with timeout
+        rt.shutdown_timeout(std::time::Duration::from_millis(500));
+        result
+    }
+
+    /// Start persistent signaling connection for direct calls.
+    /// Spawns a background task that maintains the `_signal` connection.
+    pub fn start_signaling(
+        &mut self,
+        relay_addr: &str,
+        seed_hex: &str,
+        token: Option<&str>,
+        alias: Option<&str>,
+    ) -> Result<(), anyhow::Error> {
+        use wzp_proto::{MediaTransport, SignalMessage};
+
+        let addr: SocketAddr = relay_addr.parse()?;
+        let seed = if seed_hex.is_empty() {
+            wzp_crypto::Seed::generate()
+        } else {
+            wzp_crypto::Seed::from_hex(seed_hex).map_err(|e| anyhow::anyhow!(e))?
+        };
+        let identity = seed.derive_identity();
+        let pub_id = identity.public_identity();
+        let identity_pub = *pub_id.signing.as_bytes();
+        let fp = pub_id.fingerprint.to_string();
+        let token = token.map(|s| s.to_string());
+        let alias = alias.map(|s| s.to_string());
+        let state = self.state.clone();
+        let seed_bytes = seed.0;
+
+        info!(fingerprint = %fp, relay = %addr, "starting signaling");
+
+        // Create runtime for signaling (separate from call runtime)
+        let rt = tokio::runtime::Builder::new_multi_thread()
+            .worker_threads(1)
+            .enable_all()
+            .build()?;
+
+        let signal_state = state.clone();
+        rt.spawn(async move {
+            let _ = rustls::crypto::ring::default_provider().install_default();
+            let bind: SocketAddr = "0.0.0.0:0".parse().unwrap();
+            let endpoint = match wzp_transport::create_endpoint(bind, None) {
+                Ok(e) => e,
+                Err(e) => { error!("signal endpoint: {e}"); return; }
+            };
+            let client_cfg = wzp_transport::client_config();
+            let conn = match wzp_transport::connect(&endpoint, addr, "_signal", client_cfg).await {
+                Ok(c) => c,
+                Err(e) => { error!("signal connect: {e}"); return; }
+            };
+            let transport = std::sync::Arc::new(wzp_transport::QuinnTransport::new(conn));
+
+            // Auth if token provided
+            if let Some(ref tok) = token {
+                let _ = transport.send_signal(&SignalMessage::AuthToken { token: tok.clone() }).await;
+            }
+
+            // Register presence
+            let _ = transport.send_signal(&SignalMessage::RegisterPresence {
+                identity_pub,
+                signature: vec![],
+                alias: alias.clone(),
+            }).await;
+
+            // Wait for ack
+            match transport.recv_signal().await {
+                Ok(Some(SignalMessage::RegisterPresenceAck { success: true, .. })) => {
+                    info!(fingerprint = %fp, "signal: registered");
+                    let mut stats = signal_state.stats.lock().unwrap();
+                    stats.state = crate::stats::CallState::Registered;
+                }
+                other => {
+                    error!("signal registration failed: {other:?}");
+                    return;
+                }
+            }
+
+            // Signal recv loop
+            loop {
+                if !signal_state.running.load(Ordering::Relaxed) {
+                    break;
+                }
+                match transport.recv_signal().await {
+                    Ok(Some(SignalMessage::CallRinging { call_id })) => {
+                        info!(call_id = %call_id, "signal: ringing");
+                        let mut stats = signal_state.stats.lock().unwrap();
+                        stats.state = crate::stats::CallState::Ringing;
+                    }
+                    Ok(Some(SignalMessage::DirectCallOffer { caller_fingerprint, caller_alias, call_id, .. })) => {
+                        info!(from = %caller_fingerprint, call_id = %call_id, "signal: incoming call");
+                        let mut stats = signal_state.stats.lock().unwrap();
+                        stats.state = crate::stats::CallState::IncomingCall;
+                        stats.incoming_call_id = Some(call_id);
+                        stats.incoming_caller_fp = Some(caller_fingerprint);
+                        stats.incoming_caller_alias = caller_alias;
+                    }
+                    Ok(Some(SignalMessage::DirectCallAnswer { call_id, accept_mode, .. })) => {
+                        info!(call_id = %call_id, mode = ?accept_mode, "signal: call answered");
+                    }
+                    Ok(Some(SignalMessage::CallSetup { call_id, room, relay_addr })) => {
+                        info!(call_id = %call_id, room = %room, relay = %relay_addr, "signal: call setup");
+                        // Connect to media room via the existing start_call mechanism
+                        // Store the room info so Kotlin can call startCall with it
+                        let mut stats = signal_state.stats.lock().unwrap();
+                        stats.state = crate::stats::CallState::Connecting;
+                        // Store call setup info for Kotlin to pick up
+                        stats.incoming_call_id = Some(format!("{relay_addr}|{room}"));
+                    }
+                    Ok(Some(SignalMessage::Hangup { reason })) => {
+                        info!(reason = ?reason, "signal: call ended by remote");
+                        let mut stats = signal_state.stats.lock().unwrap();
+                        stats.state = crate::stats::CallState::Closed;
+                        stats.incoming_call_id = None;
+                        stats.incoming_caller_fp = None;
+                        stats.incoming_caller_alias = None;
+                    }
+                    Ok(Some(_)) => {}
+                    Ok(None) => {
+                        info!("signal: connection closed");
+                        break;
+                    }
+                    Err(e) => {
+                        error!("signal recv error: {e}");
+                        break;
+                    }
+                }
+            }
+
+            let mut stats = signal_state.stats.lock().unwrap();
+            stats.state = crate::stats::CallState::Closed;
+        });
+
+        self.tokio_runtime = Some(rt);
+        Ok(())
+    }
+
+    /// Place a direct call to a target fingerprint via the signal connection.
+    pub fn place_call(&self, target_fingerprint: &str) -> Result<(), anyhow::Error> {
+        let _ = self.state.command_tx.send(EngineCommand::PlaceCall {
+            target_fingerprint: target_fingerprint.to_string(),
+        });
+        Ok(())
+    }
+
+    /// Answer an incoming direct call.
+    pub fn answer_call(&self, call_id: &str, mode: wzp_proto::CallAcceptMode) -> Result<(), anyhow::Error> {
+        let _ = self.state.command_tx.send(EngineCommand::AnswerCall {
+            call_id: call_id.to_string(),
+            accept_mode: mode,
+        });
+        Ok(())
+    }
+
     pub fn set_mute(&self, muted: bool) {
         self.state.muted.store(muted, Ordering::Relaxed);
     }
@@ -183,6 +408,9 @@ impl WzpEngine {
             stats.duration_secs = start.elapsed().as_secs_f64();
         }
         stats.audio_level = self.state.audio_level_rms.load(Ordering::Relaxed);
+        stats.playout_overflows = self.state.playout_ring.overflow_count();
+        stats.playout_underruns = self.state.playout_ring.underrun_count();
+        stats.capture_overflows = self.state.capture_ring.overflow_count();
         stats
     }
 
@@ -224,6 +452,7 @@ async fn run_call(
     room: &str,
     identity_seed: &[u8; 32],
     profile: QualityProfile,
+    auto_profile: bool,
     alias: Option<&str>,
     state: Arc<EngineState>,
 ) -> Result<(), anyhow::Error> {
@@ -258,6 +487,9 @@ async fn run_call(
         ephemeral_pub,
         signature,
         supported_profiles: vec![
+            QualityProfile::STUDIO_64K,
+            QualityProfile::STUDIO_48K,
+            QualityProfile::STUDIO_32K,
             QualityProfile::GOOD,
             QualityProfile::DEGRADED,
             QualityProfile::CATASTROPHIC,
@@ -272,8 +504,8 @@ async fn run_call(
         .await?
         .ok_or_else(|| anyhow::anyhow!("connection closed before CallAnswer"))?;
 
-    let relay_ephemeral_pub = match answer {
-        SignalMessage::CallAnswer { ephemeral_pub, .. } => ephemeral_pub,
+    let (relay_ephemeral_pub, chosen_profile) = match answer {
+        SignalMessage::CallAnswer { ephemeral_pub, chosen_profile, .. } => (ephemeral_pub, chosen_profile),
         other => {
             return Err(anyhow::anyhow!(
                 "expected CallAnswer, got {:?}",
@@ -282,19 +514,25 @@ async fn run_call(
         }
     };
 
+    // Auto mode: use the relay's chosen profile instead of the local preference
+    let profile = if auto_profile {
+        info!(chosen = ?chosen_profile.codec, "auto mode: using relay's chosen profile");
+        chosen_profile
+    } else {
+        profile
+    };
+
     let _session = kx.derive_session(&relay_ephemeral_pub)?;
-    info!("handshake complete, call active");
+    info!(codec = ?profile.codec, "handshake complete, call active");
 
     {
         let mut stats = state.stats.lock().unwrap();
         stats.state = CallState::Active;
     }
 
-    // Initialize Opus codec
-    let mut encoder =
-        OpusEncoder::new(profile).map_err(|e| anyhow::anyhow!("opus encoder init: {e}"))?;
-    let mut decoder =
-        OpusDecoder::new(profile).map_err(|e| anyhow::anyhow!("opus decoder init: {e}"))?;
+    // Initialize codec (Opus or Codec2 based on profile)
+    let mut encoder = wzp_codec::create_encoder(profile);
+    let mut decoder = wzp_codec::create_decoder(profile);
 
     // Initialize FEC encoder/decoder
     let mut fec_enc = wzp_fec::create_encoder(&profile);
@@ -304,21 +542,37 @@ async fn run_call(
     let mut capture_agc = AutoGainControl::new();
     let mut playout_agc = AutoGainControl::new();
 
+    let mut frame_samples = frame_samples_for(&profile);
     info!(
+        codec = ?profile.codec,
         fec_ratio = profile.fec_ratio,
         frames_per_block = profile.frames_per_block,
-        "codec + FEC + AGC initialized (48kHz mono, 20ms frames)"
+        frame_ms = profile.frame_duration_ms,
+        frame_samples,
+        "codec + FEC + AGC initialized"
     );
 
+    {
+        let mut stats = state.stats.lock().unwrap();
+        stats.current_codec = format!("{:?}", profile.codec);
+        stats.auto_mode = auto_profile;
+    }
+
     let seq = AtomicU16::new(0);
     let ts = AtomicU32::new(0);
     let transport_recv = transport.clone();
 
-    // Pre-allocate buffers
-    let mut capture_buf = vec![0i16; FRAME_SAMPLES];
+    // Adaptive quality: shared AtomicU8 between recv task (writer) and send task (reader).
+    // 0xFF = no change pending, 0-5 = index into PROFILES array.
+    let pending_profile = Arc::new(AtomicU8::new(PROFILE_NO_CHANGE));
+    let pending_profile_recv = pending_profile.clone();
+
+    // Pre-allocate buffers (sized for current profile)
+    let mut capture_buf = vec![0i16; frame_samples];
     let mut encode_buf = vec![0u8; encoder.max_frame_bytes()];
     let mut frame_in_block: u8 = 0;
     let mut block_id: u8 = 0;
+    let mut current_profile = profile;
 
     // Send task: capture ring → Opus encode → FEC → MediaPackets
     //
@@ -333,19 +587,58 @@ async fn run_call(
         let mut last_stats_log = Instant::now();
         let mut frames_sent: u64 = 0;
         let mut frames_dropped: u64 = 0;
+        // Per-step timing accumulators (reset every stats log)
+        let mut t_agc_us: u64 = 0;
+        let mut t_opus_us: u64 = 0;
+        let mut t_fec_us: u64 = 0;
+        let mut t_send_us: u64 = 0;
+        let mut t_frames: u64 = 0;
         loop {
             if !state.running.load(Ordering::Relaxed) {
                 break;
             }
 
+            // Check for adaptive profile switch from recv task
+            if auto_profile {
+                let p = pending_profile.swap(PROFILE_NO_CHANGE, Ordering::Acquire);
+                if p != PROFILE_NO_CHANGE {
+                    if let Some(new_profile) = index_to_profile(p) {
+                        info!(
+                            from = ?current_profile.codec,
+                            to = ?new_profile.codec,
+                            "auto: switching encoder profile"
+                        );
+                        if let Err(e) = encoder.set_profile(new_profile) {
+                            warn!("encoder set_profile failed: {e}");
+                        } else {
+                            fec_enc = wzp_fec::create_encoder(&new_profile);
+                            current_profile = new_profile;
+                            let new_frame_samples = frame_samples_for(&new_profile);
+                            if new_frame_samples != frame_samples {
+                                frame_samples = new_frame_samples;
+                                capture_buf.resize(frame_samples, 0);
+                            }
+                            encode_buf.resize(encoder.max_frame_bytes(), 0);
+                            // Reset FEC block state for clean switch
+                            frame_in_block = 0;
+                            block_id = block_id.wrapping_add(1);
+                            // Update stats with new codec
+                            if let Ok(mut stats) = state.stats.lock() {
+                                stats.current_codec = format!("{:?}", new_profile.codec);
+                            }
+                        }
+                    }
+                }
+            }
+
             let avail = state.capture_ring.available();
-            if avail < FRAME_SAMPLES {
+            if avail < frame_samples {
                 tokio::time::sleep(std::time::Duration::from_millis(5)).await;
                 continue;
             }
 
             let read = state.capture_ring.read(&mut capture_buf);
-            if read < FRAME_SAMPLES {
+            if read < frame_samples {
                 continue;
             }
 
@@ -356,9 +649,12 @@ async fn run_call(
             }
 
             // AGC: normalize capture volume before encoding
+            let t0 = Instant::now();
             capture_agc.process_frame(&mut capture_buf);
+            t_agc_us += t0.elapsed().as_micros() as u64;
 
             // Opus encode
+            let t0 = Instant::now();
             let encoded_len = match encoder.encode(&capture_buf, &mut encode_buf) {
                 Ok(n) => n,
                 Err(e) => {
@@ -366,19 +662,20 @@ async fn run_call(
                     continue;
                 }
             };
+            t_opus_us += t0.elapsed().as_micros() as u64;
             let encoded = &encode_buf[..encoded_len];
 
             // Build source packet
             let s = seq.fetch_add(1, Ordering::Relaxed);
-            let t = ts.fetch_add(FRAME_SAMPLES as u32, Ordering::Relaxed);
+            let t = ts.fetch_add(frame_samples as u32, Ordering::Relaxed);
 
             let source_pkt = MediaPacket {
                 header: MediaHeader {
                     version: 0,
                     is_repair: false,
-                    codec_id: profile.codec,
+                    codec_id: current_profile.codec,
                     has_quality_report: false,
-                    fec_ratio_encoded: MediaHeader::encode_fec_ratio(profile.fec_ratio),
+                    fec_ratio_encoded: MediaHeader::encode_fec_ratio(current_profile.fec_ratio),
                     seq: s,
                     timestamp: t,
                     fec_block: block_id,
@@ -391,6 +688,7 @@ async fn run_call(
             };
 
             // Send source packet — drop on error, never break
+            let t0 = Instant::now();
             if let Err(e) = transport.send_media(&source_pkt).await {
                 send_errors += 1;
                 frames_dropped += 1;
@@ -405,19 +703,22 @@ async fn run_call(
                     last_send_error_log = Instant::now();
                 }
                 // Don't feed to FEC either — the source is lost
+                t_send_us += t0.elapsed().as_micros() as u64;
                 continue;
             }
+            t_send_us += t0.elapsed().as_micros() as u64;
             frames_sent += 1;
 
             // Feed encoded frame to FEC encoder
+            let t0 = Instant::now();
             if let Err(e) = fec_enc.add_source_symbol(encoded) {
                 warn!("fec add_source error: {e}");
             }
             frame_in_block += 1;
 
             // When block is full, generate repair packets
-            if frame_in_block >= profile.frames_per_block {
-                match fec_enc.generate_repair(profile.fec_ratio) {
+            if frame_in_block >= current_profile.frames_per_block {
+                match fec_enc.generate_repair(current_profile.fec_ratio) {
                     Ok(repairs) => {
                         let repair_count = repairs.len();
                         for (sym_idx, repair_data) in repairs {
@@ -426,10 +727,10 @@ async fn run_call(
                                 header: MediaHeader {
                                     version: 0,
                                     is_repair: true,
-                                    codec_id: profile.codec,
+                                    codec_id: current_profile.codec,
                                     has_quality_report: false,
                                     fec_ratio_encoded: MediaHeader::encode_fec_ratio(
-                                        profile.fec_ratio,
+                                        current_profile.fec_ratio,
                                     ),
                                     seq: rs,
                                     timestamp: t,
@@ -452,7 +753,7 @@ async fn run_call(
                             info!(
                                 block_id,
                                 repair_count,
-                                fec_ratio = profile.fec_ratio,
+                                fec_ratio = current_profile.fec_ratio,
                                 "FEC block complete"
                             );
                         }
@@ -466,9 +767,12 @@ async fn run_call(
                 block_id = block_id.wrapping_add(1);
                 frame_in_block = 0;
             }
+            t_fec_us += t0.elapsed().as_micros() as u64;
+            t_frames += 1;
 
             // Periodic stats every 5 seconds
             if last_stats_log.elapsed().as_secs() >= 5 {
+                let avg = |total: u64| if t_frames > 0 { total / t_frames } else { 0 };
                 info!(
                     seq = s,
                     block_id,
@@ -476,16 +780,23 @@ async fn run_call(
                     frames_dropped,
                     send_errors,
                     ring_avail = state.capture_ring.available(),
+                    capture_overflows = state.capture_ring.overflow_count(),
+                    avg_agc_us = avg(t_agc_us),
+                    avg_opus_us = avg(t_opus_us),
+                    avg_fec_us = avg(t_fec_us),
+                    avg_send_us = avg(t_send_us),
+                    avg_total_us = avg(t_agc_us + t_opus_us + t_fec_us + t_send_us),
                     "send stats"
                 );
+                t_agc_us = 0; t_opus_us = 0; t_fec_us = 0; t_send_us = 0; t_frames = 0;
                 last_stats_log = Instant::now();
             }
         }
         info!(frames_sent, frames_dropped, send_errors, "send task ended");
     };
 
-    // Pre-allocate decode buffer
-    let mut decode_buf = vec![0i16; FRAME_SAMPLES];
+    // Pre-allocate decode buffer (max size to handle any incoming codec)
+    let mut decode_buf = vec![0i16; MAX_FRAME_SAMPLES];
 
     // Recv task: MediaPackets → FEC decode → Opus decode → playout ring
     let recv_task = async {
@@ -495,6 +806,8 @@ async fn run_call(
         let mut last_recv_instant = Instant::now();
         let mut max_recv_gap_ms: u64 = 0;
         let mut last_stats_log = Instant::now();
+        let mut quality_ctrl = AdaptiveQualityController::new();
+        let mut last_peer_codec: Option<CodecId> = None;
         info!("recv task started (Opus + RaptorQ FEC)");
         loop {
             if !state.running.load(Ordering::Relaxed) {
@@ -517,6 +830,23 @@ async fn run_call(
                         );
                     }
 
+                    // Adaptive quality: ingest quality reports from relay
+                    if auto_profile {
+                        if let Some(ref qr) = pkt.quality_report {
+                            if let Some(new_profile) = quality_ctrl.observe(qr) {
+                                let idx = profile_to_index(&new_profile);
+                                info!(
+                                    loss = qr.loss_percent(),
+                                    rtt = qr.rtt_ms(),
+                                    tier = ?quality_ctrl.tier(),
+                                    to = ?new_profile.codec,
+                                    "auto: quality adapter recommends switch"
+                                );
+                                pending_profile_recv.store(idx, Ordering::Release);
+                            }
+                        }
+                    }
+
                     let is_repair = pkt.header.is_repair;
                     let pkt_block = pkt.header.fec_block;
                     let pkt_symbol = pkt.header.fec_symbol;
@@ -530,7 +860,34 @@ async fn run_call(
                     );
 
                     // Source packets: decode directly
-                    if !is_repair {
+                    if !is_repair && pkt.header.codec_id != CodecId::ComfortNoise {
+                        // Switch decoder to match incoming codec if different
+                        if pkt.header.codec_id != decoder.codec_id() {
+                            let switch_profile = match pkt.header.codec_id {
+                                CodecId::Opus24k => QualityProfile::GOOD,
+                                CodecId::Opus6k => QualityProfile::DEGRADED,
+                                CodecId::Opus32k => QualityProfile::STUDIO_32K,
+                                CodecId::Opus48k => QualityProfile::STUDIO_48K,
+                                CodecId::Opus64k => QualityProfile::STUDIO_64K,
+                                CodecId::Codec2_1200 => QualityProfile::CATASTROPHIC,
+                                CodecId::Codec2_3200 => QualityProfile {
+                                    codec: CodecId::Codec2_3200,
+                                    fec_ratio: 0.5,
+                                    frame_duration_ms: 20,
+                                    frames_per_block: 5,
+                                },
+                                other => QualityProfile { codec: other, ..QualityProfile::GOOD },
+                            };
+                            info!(from = ?decoder.codec_id(), to = ?pkt.header.codec_id, "recv: switching decoder");
+                            let _ = decoder.set_profile(switch_profile);
+                        }
+                        // Track peer codec for UI display
+                        if last_peer_codec != Some(pkt.header.codec_id) {
+                            last_peer_codec = Some(pkt.header.codec_id);
+                            if let Ok(mut stats) = state.stats.lock() {
+                                stats.peer_codec = format!("{:?}", pkt.header.codec_id);
+                            }
+                        }
                         match decoder.decode(&pkt.payload, &mut decode_buf) {
                             Ok(samples) => {
                                 playout_agc.process_frame(&mut decode_buf[..samples]);
@@ -578,6 +935,8 @@ async fn run_call(
                             recv_errors,
                             max_recv_gap_ms,
                             playout_avail = state.playout_ring.available(),
+                            playout_overflows = state.playout_ring.overflow_count(),
+                            playout_underruns = state.playout_ring.underrun_count(),
                             "recv stats"
                         );
                         max_recv_gap_ms = 0;
@@ -643,6 +1002,7 @@ async fn run_call(
                         .map(|p| crate::stats::RoomMember {
                             fingerprint: p.fingerprint.clone(),
                             alias: p.alias.clone(),
+                            relay_label: p.relay_label.clone(),
                         })
                         .collect();
                     let mut stats = state_signal.stats.lock().unwrap();

crates/wzp-android/src/jni_bridge.rs

@@ -21,11 +21,24 @@ unsafe fn handle_ref(handle: jlong) -> &'static mut EngineHandle {
     unsafe { &mut *(handle as *mut EngineHandle) }
 }
 
+/// 7 = auto (use relay's chosen profile)
+const PROFILE_AUTO: jint = 7;
+
 fn profile_from_int(value: jint) -> QualityProfile {
     match value {
-        1 => QualityProfile::DEGRADED,
-        2 => QualityProfile::CATASTROPHIC,
-        _ => QualityProfile::GOOD,
+        0 => QualityProfile::GOOD,            // Opus 24k
+        1 => QualityProfile::DEGRADED,        // Opus 6k
+        2 => QualityProfile::CATASTROPHIC,    // Codec2 1.2k
+        3 => QualityProfile {                 // Codec2 3.2k
+            codec: wzp_proto::CodecId::Codec2_3200,
+            fec_ratio: 0.5,
+            frame_duration_ms: 20,
+            frames_per_block: 5,
+        },
+        4 => QualityProfile::STUDIO_32K,      // Opus 32k
+        5 => QualityProfile::STUDIO_48K,      // Opus 48k
+        6 => QualityProfile::STUDIO_64K,      // Opus 64k
+        _ => QualityProfile::GOOD,            // auto falls back to GOOD
     }
 }
 
@@ -35,11 +48,25 @@ static INIT_LOGGING: Once = Once::new();
 /// Safe to call multiple times — only the first call takes effect.
 fn init_logging() {
     INIT_LOGGING.call_once(|| {
-        use tracing_subscriber::layer::SubscriberExt;
-        use tracing_subscriber::util::SubscriberInitExt;
-        if let Ok(layer) = tracing_android::layer("wzp_android") {
-            let _ = tracing_subscriber::registry().with(layer).try_init();
-        }
+        // Wrap in catch_unwind — sharded_slab allocation inside
+        // tracing_subscriber::registry() can crash on some Android
+        // devices if scudo malloc fails during early initialization.
+        let _ = std::panic::catch_unwind(|| {
+            use tracing_subscriber::layer::SubscriberExt;
+            use tracing_subscriber::util::SubscriberInitExt;
+            use tracing_subscriber::EnvFilter;
+            if let Ok(layer) = tracing_android::layer("wzp_android") {
+                // Filter: INFO for our crates, WARN for everything else.
+                // The jni crate emits VERBOSE logs for every method lookup
+                // (~10 lines per JNI call, 100+ calls/sec) which floods logcat
+                // and causes the system to kill the app.
+                let filter = EnvFilter::new("warn,wzp_android=info,wzp_proto=info,wzp_transport=info,wzp_codec=info,wzp_fec=info,wzp_crypto=info");
+                let _ = tracing_subscriber::registry()
+                    .with(layer)
+                    .with(filter)
+                    .try_init();
+            }
+        });
     });
 }
 
@@ -71,6 +98,7 @@ pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeStartCall(
     seed_hex_j: JString,
     token_j: JString,
     alias_j: JString,
+    profile_j: jint,
 ) -> jint {
     let result = panic::catch_unwind(panic::AssertUnwindSafe(|| {
         let relay_addr: String = env.get_string(&relay_addr_j).map(|s| s.into()).unwrap_or_default();
@@ -96,7 +124,8 @@ pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeStartCall(
         }
 
         let config = CallStartConfig {
-            profile: QualityProfile::GOOD,
+            profile: profile_from_int(profile_j),
+            auto_profile: profile_j == PROFILE_AUTO,
             relay_addr,
             room,
             auth_token: if token.is_empty() { Vec::new() } else { token.into_bytes() },
@@ -209,7 +238,6 @@ pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeWriteAudio(
             return 0;
         }
         let mut buf = vec![0i16; len];
-        // GetShortArrayRegion copies Java array into our buffer
         if env.get_short_array_region(&pcm, 0, &mut buf).is_err() {
             return 0;
         }
@@ -243,6 +271,56 @@ pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeReadAudio(
     result.unwrap_or(0)
 }
 
+/// Write captured PCM from a DirectByteBuffer — zero JNI array copies.
+/// The ByteBuffer must contain little-endian i16 samples.
+/// Called from the AudioRecord capture thread.
+#[unsafe(no_mangle)]
+pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeWriteAudioDirect(
+    env: JNIEnv,
+    _class: JClass,
+    handle: jlong,
+    buffer: jni::objects::JByteBuffer,
+    sample_count: jint,
+) -> jint {
+    let result = panic::catch_unwind(panic::AssertUnwindSafe(|| {
+        let h = unsafe { handle_ref(handle) };
+        let ptr = env.get_direct_buffer_address(&buffer).unwrap_or(std::ptr::null_mut());
+        if ptr.is_null() || sample_count <= 0 {
+            return 0;
+        }
+        let samples = unsafe {
+            std::slice::from_raw_parts(ptr as *const i16, sample_count as usize)
+        };
+        h.engine.write_audio(samples) as jint
+    }));
+    result.unwrap_or(0)
+}
+
+/// Read decoded PCM into a DirectByteBuffer — zero JNI array copies.
+/// The ByteBuffer will be filled with little-endian i16 samples.
+/// Called from the AudioTrack playout thread.
+#[unsafe(no_mangle)]
+pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeReadAudioDirect(
+    env: JNIEnv,
+    _class: JClass,
+    handle: jlong,
+    buffer: jni::objects::JByteBuffer,
+    max_samples: jint,
+) -> jint {
+    let result = panic::catch_unwind(panic::AssertUnwindSafe(|| {
+        let h = unsafe { handle_ref(handle) };
+        let ptr = env.get_direct_buffer_address(&buffer).unwrap_or(std::ptr::null_mut());
+        if ptr.is_null() || max_samples <= 0 {
+            return 0;
+        }
+        let samples = unsafe {
+            std::slice::from_raw_parts_mut(ptr as *mut i16, max_samples as usize)
+        };
+        h.engine.read_audio(samples) as jint
+    }));
+    result.unwrap_or(0)
+}
+
 #[unsafe(no_mangle)]
 pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeDestroy(
     _env: JNIEnv,
@@ -254,3 +332,116 @@ pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeDestroy(
         drop(h);
     }));
 }
+
+/// Ping a relay server — instance method, requires engine handle.
+/// Returns JSON `{"rtt_ms":N,"server_fingerprint":"hex"}` or null on failure.
+#[unsafe(no_mangle)]
+pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativePingRelay<'a>(
+    mut env: JNIEnv<'a>,
+    _class: JClass,
+    handle: jlong,
+    relay_j: JString,
+) -> jstring {
+    let result = panic::catch_unwind(panic::AssertUnwindSafe(|| {
+        let h = unsafe { handle_ref(handle) };
+        let relay: String = env.get_string(&relay_j).map(|s| s.into()).unwrap_or_default();
+        match h.engine.ping_relay(&relay) {
+            Ok(json) => Some(json),
+            Err(_) => None,
+        }
+    }));
+
+    let json = match result {
+        Ok(Some(s)) => s,
+        _ => return JObject::null().into_raw(),
+    };
+    env.new_string(&json)
+        .map(|s| s.into_raw())
+        .unwrap_or(JObject::null().into_raw())
+}
+
+// ── Direct calling JNI functions ──
+
+/// Start persistent signaling connection to relay for direct calls.
+/// Returns 0 on success, -1 on error.
+#[unsafe(no_mangle)]
+pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeStartSignaling<'a>(
+    mut env: JNIEnv<'a>,
+    _class: JClass,
+    handle: jlong,
+    relay_addr_j: JString,
+    seed_hex_j: JString,
+    token_j: JString,
+    alias_j: JString,
+) -> jint {
+    let result = panic::catch_unwind(panic::AssertUnwindSafe(|| {
+        let h = unsafe { handle_ref(handle) };
+        let relay_addr: String = env.get_string(&relay_addr_j).map(|s| s.into()).unwrap_or_default();
+        let seed_hex: String = env.get_string(&seed_hex_j).map(|s| s.into()).unwrap_or_default();
+        let token: String = env.get_string(&token_j).map(|s| s.into()).unwrap_or_default();
+        let alias: String = env.get_string(&alias_j).map(|s| s.into()).unwrap_or_default();
+
+        h.engine.start_signaling(
+            &relay_addr,
+            &seed_hex,
+            if token.is_empty() { None } else { Some(&token) },
+            if alias.is_empty() { None } else { Some(&alias) },
+        )
+    }));
+
+    match result {
+        Ok(Ok(())) => 0,
+        Ok(Err(e)) => { error!("start_signaling failed: {e}"); -1 }
+        Err(_) => { error!("start_signaling panicked"); -1 }
+    }
+}
+
+/// Place a direct call to a target fingerprint.
+/// Returns 0 on success, -1 on error.
+#[unsafe(no_mangle)]
+pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativePlaceCall<'a>(
+    mut env: JNIEnv<'a>,
+    _class: JClass,
+    handle: jlong,
+    target_fp_j: JString,
+) -> jint {
+    let result = panic::catch_unwind(panic::AssertUnwindSafe(|| {
+        let h = unsafe { handle_ref(handle) };
+        let target: String = env.get_string(&target_fp_j).map(|s| s.into()).unwrap_or_default();
+        h.engine.place_call(&target)
+    }));
+
+    match result {
+        Ok(Ok(())) => 0,
+        Ok(Err(e)) => { error!("place_call failed: {e}"); -1 }
+        Err(_) => { error!("place_call panicked"); -1 }
+    }
+}
+
+/// Answer an incoming direct call.
+/// mode: 0=Reject, 1=AcceptTrusted, 2=AcceptGeneric
+#[unsafe(no_mangle)]
+pub unsafe extern "system" fn Java_com_wzp_engine_WzpEngine_nativeAnswerCall<'a>(
+    mut env: JNIEnv<'a>,
+    _class: JClass,
+    handle: jlong,
+    call_id_j: JString,
+    mode: jint,
+) -> jint {
+    let result = panic::catch_unwind(panic::AssertUnwindSafe(|| {
+        let h = unsafe { handle_ref(handle) };
+        let call_id: String = env.get_string(&call_id_j).map(|s| s.into()).unwrap_or_default();
+        let accept_mode = match mode {
+            0 => wzp_proto::CallAcceptMode::Reject,
+            1 => wzp_proto::CallAcceptMode::AcceptTrusted,
+            _ => wzp_proto::CallAcceptMode::AcceptGeneric,
+        };
+        h.engine.answer_call(&call_id, accept_mode)
+    }));
+
+    match result {
+        Ok(Ok(())) => 0,
+        Ok(Err(e)) => { error!("answer_call failed: {e}"); -1 }
+        Err(_) => { error!("answer_call panicked"); -1 }
+    }
+}

crates/wzp-android/src/stats.rs

@@ -11,6 +11,12 @@ pub enum CallState {
     Active,
     Reconnecting,
     Closed,
+    /// Connected to relay signal channel, registered for direct calls.
+    Registered,
+    /// Outgoing call ringing on callee's side.
+    Ringing,
+    /// Incoming call received, waiting for user to accept/reject.
+    IncomingCall,
 }
 
 impl serde::Serialize for CallState {
@@ -21,6 +27,9 @@ impl serde::Serialize for CallState {
             CallState::Active => 2,
             CallState::Reconnecting => 3,
             CallState::Closed => 4,
+            CallState::Registered => 5,
+            CallState::Ringing => 6,
+            CallState::IncomingCall => 7,
         };
         serializer.serialize_u8(n)
     }
@@ -51,12 +60,36 @@ pub struct CallStats {
     pub underruns: u64,
     /// Frames recovered by FEC.
     pub fec_recovered: u64,
+    /// Playout ring overflow count (reader was lapped by writer).
+    pub playout_overflows: u64,
+    /// Playout ring underrun count (reader found empty buffer).
+    pub playout_underruns: u64,
+    /// Capture ring overflow count.
+    pub capture_overflows: u64,
     /// Current mic audio level (RMS of i16 samples, 0-32767).
     pub audio_level: u32,
+    /// Our current outgoing codec name (e.g. "Opus24k", "Codec2_1200").
+    pub current_codec: String,
+    /// Last seen incoming codec from other participants.
+    pub peer_codec: String,
+    /// Whether auto quality mode is active.
+    pub auto_mode: bool,
     /// Number of participants in the room (from last RoomUpdate).
     pub room_participant_count: u32,
     /// Participant list (fingerprint + optional alias) serialized as JSON array.
     pub room_participants: Vec<RoomMember>,
+    /// SAS code for verbal verification (None if not in a call).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub sas_code: Option<u32>,
+    /// Incoming call info (present when state == IncomingCall).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub incoming_call_id: Option<String>,
+    /// Fingerprint of the caller (present when state == IncomingCall).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub incoming_caller_fp: Option<String>,
+    /// Alias of the caller (present when state == IncomingCall).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub incoming_caller_alias: Option<String>,
 }
 
 /// A room member entry, serialized into the stats JSON.
@@ -64,4 +97,5 @@ pub struct CallStats {
 pub struct RoomMember {
     pub fingerprint: String,
     pub alias: Option<String>,
+    pub relay_label: Option<String>,
 }

crates/wzp-client/src/featherchat.rs

@@ -110,7 +110,15 @@ pub fn signal_to_call_type(signal: &SignalMessage) -> CallSignalType {
         SignalMessage::SessionForward { .. } => CallSignalType::Offer, // reuse
         SignalMessage::SessionForwardAck { .. } => CallSignalType::Offer, // reuse
         SignalMessage::RoomUpdate { .. } => CallSignalType::Offer, // reuse
-        SignalMessage::SetAlias { .. } => CallSignalType::Offer, // reuse
+        SignalMessage::FederationHello { .. }
+        | SignalMessage::GlobalRoomActive { .. }
+        | SignalMessage::GlobalRoomInactive { .. } => CallSignalType::Offer, // relay-only
+        SignalMessage::DirectCallOffer { .. } => CallSignalType::Offer,
+        SignalMessage::DirectCallAnswer { .. } => CallSignalType::Answer,
+        SignalMessage::CallSetup { .. } => CallSignalType::Offer, // relay-only
+        SignalMessage::CallRinging { .. } => CallSignalType::Ringing,
+        SignalMessage::RegisterPresence { .. }
+        | SignalMessage::RegisterPresenceAck { .. } => CallSignalType::Offer, // relay-only
     }
 }
 

crates/wzp-crypto/src/handshake.rs

@@ -110,7 +110,18 @@ impl KeyExchange for WarzoneKeyExchange {
         hk.expand(b"warzone-session-key", &mut session_key)
             .expect("HKDF expand for session key should not fail");
 
-        Ok(Box::new(ChaChaSession::new(session_key)))
+        // Derive SAS (Short Authentication String) from shared secret only.
+        // The shared secret is identical on both sides (X25519 DH property).
+        // A MITM would produce a different shared secret → different SAS.
+        // We use a dedicated HKDF label so SAS is independent of the session key.
+        let mut sas_key = [0u8; 4];
+        hk.expand(b"warzone-sas-code", &mut sas_key)
+            .expect("HKDF expand for SAS should not fail");
+        let sas_code = u32::from_be_bytes(sas_key) % 10000;
+
+        let mut session = ChaChaSession::new(session_key);
+        session.set_sas(sas_code);
+        Ok(Box::new(session))
     }
 }
 
@@ -211,4 +222,47 @@ mod tests {
 
         assert_eq!(&decrypted, plaintext);
     }
+
+    #[test]
+    fn sas_codes_match_between_peers() {
+        let mut alice = WarzoneKeyExchange::from_identity_seed(&[0xAA; 32]);
+        let mut bob = WarzoneKeyExchange::from_identity_seed(&[0xBB; 32]);
+
+        let alice_eph_pub = alice.generate_ephemeral();
+        let bob_eph_pub = bob.generate_ephemeral();
+
+        let alice_session = alice.derive_session(&bob_eph_pub).unwrap();
+        let bob_session = bob.derive_session(&alice_eph_pub).unwrap();
+
+        let alice_sas = alice_session.sas_code();
+        let bob_sas = bob_session.sas_code();
+
+        assert!(alice_sas.is_some(), "Alice should have SAS");
+        assert!(bob_sas.is_some(), "Bob should have SAS");
+        assert_eq!(alice_sas, bob_sas, "SAS codes must match between peers");
+        assert!(alice_sas.unwrap() < 10000, "SAS should be 4 digits");
+    }
+
+    #[test]
+    fn sas_differs_for_different_peers() {
+        let mut alice = WarzoneKeyExchange::from_identity_seed(&[0xAA; 32]);
+        let mut bob = WarzoneKeyExchange::from_identity_seed(&[0xBB; 32]);
+        let mut eve = WarzoneKeyExchange::from_identity_seed(&[0xEE; 32]);
+
+        let alice_eph = alice.generate_ephemeral();
+        let bob_eph = bob.generate_ephemeral();
+        let eve_eph = eve.generate_ephemeral();
+
+        let alice_bob_session = alice.derive_session(&bob_eph).unwrap();
+
+        // Eve does separate handshake with Bob (MITM scenario)
+        let eve_bob_session = eve.derive_session(&bob_eph).unwrap();
+
+        // SAS codes should differ — Eve's session has different shared secret
+        assert_ne!(
+            alice_bob_session.sas_code(),
+            eve_bob_session.sas_code(),
+            "MITM session should produce different SAS"
+        );
+    }
 }

crates/wzp-crypto/src/session.rs

@@ -26,6 +26,8 @@ pub struct ChaChaSession {
     rekey_mgr: RekeyManager,
     /// Pending ephemeral secret for rekey (stored until peer responds).
     pending_rekey_secret: Option<StaticSecret>,
+    /// Short Authentication String (4-digit code for verbal verification).
+    sas_code: Option<u32>,
 }
 
 impl ChaChaSession {
@@ -46,9 +48,15 @@ impl ChaChaSession {
             recv_seq: 0,
             rekey_mgr: RekeyManager::new(shared_secret),
             pending_rekey_secret: None,
+            sas_code: None,
         }
     }
 
+    /// Set the SAS code (called by key exchange after derivation).
+    pub fn set_sas(&mut self, code: u32) {
+        self.sas_code = Some(code);
+    }
+
     /// Install a new key (after rekeying).
     fn install_key(&mut self, new_key: [u8; 32]) {
         use sha2::Digest;
@@ -136,6 +144,10 @@ impl CryptoSession for ChaChaSession {
 
         Ok(())
     }
+
+    fn sas_code(&self) -> Option<u32> {
+        self.sas_code
+    }
 }
 
 #[cfg(test)]

crates/wzp-fec/src/decoder.rs

@@ -1,6 +1,7 @@
 //! RaptorQ FEC decoder — reassembles source blocks from received source and repair symbols.
 
 use std::collections::HashMap;
+use std::time::Instant;
 
 use raptorq::{EncodingPacket, ObjectTransmissionInformation, PayloadId, SourceBlockDecoder};
 use wzp_proto::error::FecError;
@@ -9,6 +10,9 @@ use wzp_proto::FecDecoder;
 /// Length prefix size (u16 little-endian), must match encoder.
 const LEN_PREFIX: usize = 2;
 
+/// Decoded blocks older than this are eligible for reuse by a new sender.
+const BLOCK_STALE_SECS: u64 = 2;
+
 /// State for one in-flight block being decoded.
 struct BlockState {
     /// Number of source symbols expected.
@@ -21,6 +25,8 @@ struct BlockState {
     decoded: bool,
     /// Cached decoded result.
     result: Option<Vec<Vec<u8>>>,
+    /// When this block was last decoded (for staleness check).
+    decoded_at: Option<Instant>,
 }
 
 /// RaptorQ-based FEC decoder that handles multiple concurrent blocks.
@@ -58,6 +64,7 @@ impl RaptorQFecDecoder {
             symbol_size: self.symbol_size,
             decoded: false,
             result: None,
+            decoded_at: None,
         })
     }
 }
@@ -74,8 +81,20 @@ impl FecDecoder for RaptorQFecDecoder {
         let block = self.get_or_create_block(block_id);
 
         if block.decoded {
-            // Already decoded, ignore additional symbols.
-            return Ok(());
+            // If the block was decoded recently, skip (normal duplicate).
+            // If it's stale (>2s), a new sender is reusing this block_id — reset it.
+            if let Some(at) = block.decoded_at {
+                if at.elapsed().as_secs() >= BLOCK_STALE_SECS {
+                    block.decoded = false;
+                    block.result = None;
+                    block.decoded_at = None;
+                    block.packets.clear();
+                } else {
+                    return Ok(());
+                }
+            } else {
+                return Ok(());
+            }
         }
 
         // Data should already be at symbol_size (length-prefixed and padded by the encoder).
@@ -132,6 +151,7 @@ impl FecDecoder for RaptorQFecDecoder {
 
                 let block = self.blocks.get_mut(&block_id).unwrap();
                 block.decoded = true;
+                block.decoded_at = Some(Instant::now());
                 block.result = Some(frames.clone());
                 Ok(Some(frames))
             }

crates/wzp-proto/src/jitter.rs

@@ -273,10 +273,21 @@ impl JitterBuffer {
             return;
         }
 
-        // Check if packet is too old (already played out)
+        // Check if packet is too old (already played out).
+        // A backward jump of >100 seq (~2s at 50fps) indicates a new sender in a
+        // federation room — reset instead of dropping.
         if self.stats.packets_played > 0 && seq_before(seq, self.next_playout_seq) {
-            self.stats.packets_late += 1;
-            return;
+            let backward_distance = self.next_playout_seq.wrapping_sub(seq);
+            tracing::warn!(seq, next = self.next_playout_seq, backward_distance, "jitter: backward seq detected");
+            if backward_distance > 100 {
+                tracing::info!(seq, next = self.next_playout_seq, "jitter: RESET — new sender detected");
+                self.buffer.clear();
+                self.next_playout_seq = seq;
+                self.stats.packets_late = 0;
+            } else {
+                self.stats.packets_late += 1;
+                return;
+            }
         }
 
         // If we haven't started playout yet, adjust next_playout_seq to earliest known
@@ -412,10 +423,21 @@ impl JitterBuffer {
             return;
         }
 
-        // Check if packet is too old (already played out)
+        // Check if packet is too old (already played out).
+        // A backward jump of >100 seq (~2s at 50fps) indicates a new sender in a
+        // federation room — reset instead of dropping.
         if self.stats.packets_played > 0 && seq_before(seq, self.next_playout_seq) {
-            self.stats.packets_late += 1;
-            return;
+            let backward_distance = self.next_playout_seq.wrapping_sub(seq);
+            tracing::warn!(seq, next = self.next_playout_seq, backward_distance, "jitter: backward seq detected");
+            if backward_distance > 100 {
+                tracing::info!(seq, next = self.next_playout_seq, "jitter: RESET — new sender detected");
+                self.buffer.clear();
+                self.next_playout_seq = seq;
+                self.stats.packets_late = 0;
+            } else {
+                self.stats.packets_late += 1;
+                return;
+            }
         }
 
         // If we haven't started playout yet, adjust next_playout_seq to earliest known

crates/wzp-proto/src/lib.rs

@@ -25,8 +25,9 @@ pub mod traits;
 pub use codec_id::{CodecId, QualityProfile};
 pub use error::*;
 pub use packet::{
-    HangupReason, MediaHeader, MediaPacket, MiniFrameContext, MiniHeader, QualityReport,
-    RoomParticipant, SignalMessage, TrunkEntry, TrunkFrame, FRAME_TYPE_FULL, FRAME_TYPE_MINI,
+    CallAcceptMode, HangupReason, MediaHeader, MediaPacket, MiniFrameContext, MiniHeader,
+    QualityReport, RoomParticipant, SignalMessage, TrunkEntry, TrunkFrame, FRAME_TYPE_FULL,
+    FRAME_TYPE_MINI,
 };
 pub use bandwidth::{BandwidthEstimator, CongestionState};
 pub use quality::{AdaptiveQualityController, NetworkContext, Tier};

crates/wzp-proto/src/packet.rs

@@ -657,10 +657,111 @@ pub enum SignalMessage {
         participants: Vec<RoomParticipant>,
     },
 
-    /// Set or update the client's display name.
-    /// Sent by client after joining; relay updates the participant entry and
-    /// re-broadcasts a RoomUpdate to all participants.
-    SetAlias { alias: String },
+    // ── Federation signals (relay-to-relay) ──
+
+    /// Federation: initial handshake — the connecting relay identifies itself.
+    FederationHello {
+        /// TLS certificate fingerprint of the connecting relay.
+        tls_fingerprint: String,
+    },
+
+    /// Federation: this relay now has local participants in a global room.
+    GlobalRoomActive {
+        room: String,
+        /// Participants on the announcing relay (for federated presence).
+        #[serde(default)]
+        participants: Vec<RoomParticipant>,
+    },
+
+    /// Federation: this relay's last local participant left a global room.
+    GlobalRoomInactive {
+        room: String,
+    },
+
+    // ── Direct calling signals (client ↔ relay signaling) ──
+
+    /// Register on relay for direct calls. Sent on `_signal` connections
+    /// after optional AuthToken.
+    RegisterPresence {
+        /// Client's Ed25519 identity public key.
+        identity_pub: [u8; 32],
+        /// Signature over ("register-presence" || identity_pub).
+        signature: Vec<u8>,
+        /// Optional display name.
+        alias: Option<String>,
+    },
+
+    /// Relay confirms presence registration.
+    RegisterPresenceAck {
+        success: bool,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        error: Option<String>,
+    },
+
+    /// Direct call offer routed through the relay to a specific peer.
+    DirectCallOffer {
+        /// Caller's fingerprint.
+        caller_fingerprint: String,
+        /// Caller's display name.
+        caller_alias: Option<String>,
+        /// Target's fingerprint.
+        target_fingerprint: String,
+        /// Unique call session ID (UUID).
+        call_id: String,
+        /// Caller's Ed25519 identity pub.
+        identity_pub: [u8; 32],
+        /// Caller's ephemeral X25519 pub (for key exchange on media connect).
+        ephemeral_pub: [u8; 32],
+        /// Signature over (ephemeral_pub || target_fingerprint || call_id).
+        signature: Vec<u8>,
+        /// Supported quality profiles.
+        supported_profiles: Vec<crate::QualityProfile>,
+    },
+
+    /// Callee's response to a direct call.
+    DirectCallAnswer {
+        call_id: String,
+        /// How the callee accepts (or rejects).
+        accept_mode: CallAcceptMode,
+        /// Callee's identity pub (present when accepting).
+        #[serde(skip_serializing_if = "Option::is_none")]
+        identity_pub: Option<[u8; 32]>,
+        /// Callee's ephemeral pub (present when accepting).
+        #[serde(skip_serializing_if = "Option::is_none")]
+        ephemeral_pub: Option<[u8; 32]>,
+        /// Signature (present when accepting).
+        #[serde(skip_serializing_if = "Option::is_none")]
+        signature: Option<Vec<u8>>,
+        /// Chosen quality profile (present when accepting).
+        #[serde(skip_serializing_if = "Option::is_none")]
+        chosen_profile: Option<crate::QualityProfile>,
+    },
+
+    /// Relay tells both parties: media room is ready.
+    CallSetup {
+        call_id: String,
+        /// Room name on the relay for the media session (e.g., "_call:a1b2c3d4").
+        room: String,
+        /// Relay address for the QUIC media connection.
+        relay_addr: String,
+    },
+
+    /// Ringing notification (relay → caller, callee received the offer).
+    CallRinging {
+        call_id: String,
+    },
+}
+
+/// How the callee responds to a direct call.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
+pub enum CallAcceptMode {
+    /// Reject the call.
+    Reject,
+    /// Accept with trust — in Phase 2, this enables P2P (reveals IP).
+    /// In Phase 1, behaves the same as AcceptGeneric.
+    AcceptTrusted,
+    /// Accept with privacy — relay always mediates media.
+    AcceptGeneric,
 }
 
 /// A participant entry in a RoomUpdate message.

crates/wzp-proto/src/traits.rs

@@ -132,6 +132,14 @@ pub trait CryptoSession: Send + Sync {
     fn overhead(&self) -> usize {
         16 // ChaCha20-Poly1305 tag
     }
+
+    /// Short Authentication String (SAS) — 4-digit code for verbal verification.
+    /// Both peers derive the same code from the shared secret + identity keys.
+    /// If a MITM relay is intercepting, the codes will differ.
+    /// Returns None if SAS was not computed (e.g., relay-side sessions).
+    fn sas_code(&self) -> Option<u32> {
+        None
+    }
 }
 
 /// Key exchange using the Warzone identity model.

crates/wzp-relay/Cargo.toml

@@ -29,6 +29,8 @@ axum = { version = "0.7", default-features = false, features = ["tokio", "http1"
 tower-http = { version = "0.6", features = ["fs"] }
 futures-util = "0.3"
 dirs = "6"
+sha2 = { workspace = true }
+chrono = "0.4"
 
 [[bin]]
 name = "wzp-relay"

crates/wzp-relay/build.rs

@@ -0,0 +1,18 @@
+use std::process::Command;
+
+fn main() {
+    // Get git hash at build time
+    let output = Command::new("git")
+        .args(["rev-parse", "--short", "HEAD"])
+        .output();
+
+    let hash = match output {
+        Ok(o) if o.status.success() => {
+            String::from_utf8_lossy(&o.stdout).trim().to_string()
+        }
+        _ => "unknown".to_string(),
+    };
+
+    println!("cargo:rustc-env=WZP_BUILD_HASH={hash}");
+    println!("cargo:rerun-if-changed=.git/HEAD");
+}

crates/wzp-relay/src/call_registry.rs

@@ -0,0 +1,199 @@
+//! Direct call state tracking.
+//!
+//! Manages the lifecycle of 1:1 direct calls placed via the `_signal` channel.
+//! Each call goes through: Pending → Ringing → Active → Ended.
+
+use std::collections::HashMap;
+use std::time::{Duration, Instant};
+
+/// State of a direct call.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum DirectCallState {
+    /// Offer sent to callee, waiting for response.
+    Pending,
+    /// Callee acknowledged, ringing.
+    Ringing,
+    /// Call accepted, media room active.
+    Active,
+    /// Call ended (hangup, reject, timeout, or error).
+    Ended,
+}
+
+/// A tracked direct call between two users.
+pub struct DirectCall {
+    pub call_id: String,
+    pub caller_fingerprint: String,
+    pub callee_fingerprint: String,
+    pub state: DirectCallState,
+    pub accept_mode: Option<wzp_proto::CallAcceptMode>,
+    /// Private room name (set when accepted).
+    pub room_name: Option<String>,
+    pub created_at: Instant,
+    pub answered_at: Option<Instant>,
+    pub ended_at: Option<Instant>,
+}
+
+/// Registry of active direct calls.
+pub struct CallRegistry {
+    calls: HashMap<String, DirectCall>,
+}
+
+impl CallRegistry {
+    pub fn new() -> Self {
+        Self {
+            calls: HashMap::new(),
+        }
+    }
+
+    /// Create a new pending call. Returns the call_id.
+    pub fn create_call(&mut self, call_id: String, caller_fp: String, callee_fp: String) -> &DirectCall {
+        let call = DirectCall {
+            call_id: call_id.clone(),
+            caller_fingerprint: caller_fp,
+            callee_fingerprint: callee_fp,
+            state: DirectCallState::Pending,
+            accept_mode: None,
+            room_name: None,
+            created_at: Instant::now(),
+            answered_at: None,
+            ended_at: None,
+        };
+        self.calls.insert(call_id.clone(), call);
+        self.calls.get(&call_id).unwrap()
+    }
+
+    /// Get a call by ID.
+    pub fn get(&self, call_id: &str) -> Option<&DirectCall> {
+        self.calls.get(call_id)
+    }
+
+    /// Get a mutable call by ID.
+    pub fn get_mut(&mut self, call_id: &str) -> Option<&mut DirectCall> {
+        self.calls.get_mut(call_id)
+    }
+
+    /// Transition to Ringing state.
+    pub fn set_ringing(&mut self, call_id: &str) -> bool {
+        if let Some(call) = self.calls.get_mut(call_id) {
+            if call.state == DirectCallState::Pending {
+                call.state = DirectCallState::Ringing;
+                return true;
+            }
+        }
+        false
+    }
+
+    /// Transition to Active state.
+    pub fn set_active(&mut self, call_id: &str, mode: wzp_proto::CallAcceptMode, room: String) -> bool {
+        if let Some(call) = self.calls.get_mut(call_id) {
+            if call.state == DirectCallState::Pending || call.state == DirectCallState::Ringing {
+                call.state = DirectCallState::Active;
+                call.accept_mode = Some(mode);
+                call.room_name = Some(room);
+                call.answered_at = Some(Instant::now());
+                return true;
+            }
+        }
+        false
+    }
+
+    /// End a call.
+    pub fn end_call(&mut self, call_id: &str) -> Option<DirectCall> {
+        if let Some(call) = self.calls.get_mut(call_id) {
+            call.state = DirectCallState::Ended;
+            call.ended_at = Some(Instant::now());
+        }
+        self.calls.remove(call_id)
+    }
+
+    /// Find active/pending calls involving a fingerprint.
+    pub fn calls_for_fingerprint(&self, fp: &str) -> Vec<&DirectCall> {
+        self.calls.values()
+            .filter(|c| {
+                c.state != DirectCallState::Ended
+                    && (c.caller_fingerprint == fp || c.callee_fingerprint == fp)
+            })
+            .collect()
+    }
+
+    /// Find the peer's fingerprint in a call.
+    pub fn peer_fingerprint(&self, call_id: &str, my_fp: &str) -> Option<&str> {
+        self.calls.get(call_id).map(|c| {
+            if c.caller_fingerprint == my_fp {
+                c.callee_fingerprint.as_str()
+            } else {
+                c.caller_fingerprint.as_str()
+            }
+        })
+    }
+
+    /// Remove calls that have been pending longer than the timeout.
+    /// Returns call IDs of expired calls.
+    pub fn expire_stale(&mut self, timeout: Duration) -> Vec<DirectCall> {
+        let now = Instant::now();
+        let expired: Vec<String> = self.calls.iter()
+            .filter(|(_, c)| {
+                c.state == DirectCallState::Pending
+                    && now.duration_since(c.created_at) > timeout
+            })
+            .map(|(id, _)| id.clone())
+            .collect();
+
+        expired.into_iter()
+            .filter_map(|id| self.calls.remove(&id))
+            .collect()
+    }
+
+    /// Number of active (non-ended) calls.
+    pub fn active_count(&self) -> usize {
+        self.calls.values()
+            .filter(|c| c.state != DirectCallState::Ended)
+            .count()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn call_lifecycle() {
+        let mut reg = CallRegistry::new();
+        reg.create_call("c1".into(), "alice".into(), "bob".into());
+
+        assert_eq!(reg.get("c1").unwrap().state, DirectCallState::Pending);
+        assert!(reg.set_ringing("c1"));
+        assert_eq!(reg.get("c1").unwrap().state, DirectCallState::Ringing);
+
+        assert!(reg.set_active("c1", wzp_proto::CallAcceptMode::AcceptGeneric, "_call:c1".into()));
+        assert_eq!(reg.get("c1").unwrap().state, DirectCallState::Active);
+        assert_eq!(reg.get("c1").unwrap().room_name.as_deref(), Some("_call:c1"));
+
+        let ended = reg.end_call("c1").unwrap();
+        assert_eq!(ended.state, DirectCallState::Ended);
+        assert_eq!(reg.active_count(), 0);
+    }
+
+    #[test]
+    fn expire_stale_calls() {
+        let mut reg = CallRegistry::new();
+        reg.create_call("c1".into(), "alice".into(), "bob".into());
+
+        // Not expired yet
+        let expired = reg.expire_stale(Duration::from_secs(30));
+        assert!(expired.is_empty());
+
+        // Force expiry with 0 timeout
+        let expired = reg.expire_stale(Duration::from_secs(0));
+        assert_eq!(expired.len(), 1);
+        assert_eq!(expired[0].call_id, "c1");
+    }
+
+    #[test]
+    fn peer_lookup() {
+        let mut reg = CallRegistry::new();
+        reg.create_call("c1".into(), "alice".into(), "bob".into());
+        assert_eq!(reg.peer_fingerprint("c1", "alice"), Some("bob"));
+        assert_eq!(reg.peer_fingerprint("c1", "bob"), Some("alice"));
+    }
+}

crates/wzp-relay/src/config.rs

@@ -3,8 +3,41 @@
 use serde::{Deserialize, Serialize};
 use std::net::SocketAddr;
 
-/// Configuration for the relay daemon.
+/// A federated peer relay.
 #[derive(Clone, Debug, Serialize, Deserialize)]
+pub struct PeerConfig {
+    /// Address of the peer relay (e.g., "193.180.213.68:4433").
+    pub url: String,
+    /// Expected TLS certificate fingerprint (hex, with colons).
+    pub fingerprint: String,
+    /// Optional human-readable label.
+    #[serde(default)]
+    pub label: Option<String>,
+}
+
+/// A trusted relay — accepts inbound federation without needing the peer's address.
+#[derive(Clone, Debug, Serialize, Deserialize)]
+pub struct TrustedConfig {
+    /// Expected TLS certificate fingerprint (hex, with colons).
+    pub fingerprint: String,
+    /// Optional human-readable label.
+    #[serde(default)]
+    pub label: Option<String>,
+}
+
+/// A room declared global — bridged across all federated peers.
+#[derive(Clone, Debug, Serialize, Deserialize)]
+pub struct GlobalRoomConfig {
+    /// Room name to bridge (e.g., "android").
+    pub name: String,
+}
+
+/// Configuration for the relay daemon.
+///
+/// All fields have defaults, so a minimal TOML file only needs the
+/// fields you want to override (e.g., just `[[peers]]`).
+#[derive(Clone, Debug, Serialize, Deserialize)]
+#[serde(default)]
 pub struct RelayConfig {
     /// Address to listen on for incoming connections (client-facing).
     pub listen_addr: SocketAddr,
@@ -44,6 +77,22 @@ pub struct RelayConfig {
     pub ws_port: Option<u16>,
     /// Directory to serve static files from (HTML/JS/WASM for web clients).
     pub static_dir: Option<String>,
+    /// Federation peer relays.
+    #[serde(default)]
+    pub peers: Vec<PeerConfig>,
+    /// Global rooms bridged across federation.
+    #[serde(default)]
+    pub global_rooms: Vec<GlobalRoomConfig>,
+    /// Trusted relay fingerprints — accept inbound federation from these relays.
+    /// Unlike [[peers]], no url is needed — the peer connects to us.
+    #[serde(default)]
+    pub trusted: Vec<TrustedConfig>,
+    /// Debug tap: log packet headers for matching rooms ("*" = all rooms).
+    /// Activated via --debug-tap <room> or debug_tap = "room" in TOML.
+    pub debug_tap: Option<String>,
+    /// JSONL event log path for protocol analysis (--event-log).
+    #[serde(skip)]
+    pub event_log: Option<String>,
 }
 
 impl Default for RelayConfig {
@@ -62,6 +111,100 @@ impl Default for RelayConfig {
             trunking_enabled: false,
             ws_port: None,
             static_dir: None,
+            peers: Vec::new(),
+            global_rooms: Vec::new(),
+            trusted: Vec::new(),
+            debug_tap: None,
+            event_log: None,
         }
     }
 }
+
+/// Load relay configuration from a TOML file.
+pub fn load_config(path: &str) -> Result<RelayConfig, anyhow::Error> {
+    let content = std::fs::read_to_string(path)?;
+    let config: RelayConfig = toml::from_str(&content)?;
+    Ok(config)
+}
+
+/// Info about this relay instance, used to generate personalized example configs.
+pub struct RelayInfo {
+    pub listen_addr: String,
+    pub tls_fingerprint: String,
+    pub public_ip: Option<String>,
+}
+
+/// Load config from path, or create a personalized example config if it doesn't exist.
+pub fn load_or_create_config(path: &str, info: Option<&RelayInfo>) -> Result<RelayConfig, anyhow::Error> {
+    let p = std::path::Path::new(path);
+    if p.exists() {
+        return load_config(path);
+    }
+    // Create parent directory if needed
+    if let Some(parent) = p.parent() {
+        std::fs::create_dir_all(parent)?;
+    }
+    // Generate personalized example config
+    let example = generate_example_config(info);
+    std::fs::write(p, &example)?;
+    eprintln!("Created example config at {path} — edit it and restart.");
+    let config: RelayConfig = toml::from_str(&example)?;
+    Ok(config)
+}
+
+/// Generate an example TOML config, personalized with this relay's info if available.
+fn generate_example_config(info: Option<&RelayInfo>) -> String {
+    let listen = info.map(|i| i.listen_addr.as_str()).unwrap_or("0.0.0.0:4433");
+    let peer_example = if let Some(i) = info {
+        let ip = i.public_ip.as_deref().unwrap_or("this-relay-ip");
+        format!(
+            r#"# Other relays can peer with this relay using:
+# [[peers]]
+# url = "{ip}:{port}"
+# fingerprint = "{fp}"
+# label = "This Relay""#,
+            port = listen.rsplit(':').next().unwrap_or("4433"),
+            fp = i.tls_fingerprint,
+        )
+    } else {
+        "# To peer with another relay, add its url + fingerprint:".to_string()
+    };
+
+    format!(
+        r#"# WarzonePhone Relay Configuration
+# See docs/ADMINISTRATION.md for full reference.
+
+# Listen address for client connections
+listen_addr = "{listen}"
+
+# Maximum concurrent sessions
+# max_sessions = 100
+
+# Prometheus metrics endpoint (uncomment to enable)
+# metrics_port = 9090
+
+# featherChat auth endpoint (uncomment to enable)
+# auth_url = "https://chat.example.com/v1/auth/validate"
+
+{peer_example}
+
+# Federation: peer relays we connect to (outbound)
+# [[peers]]
+# url = "other-relay.example.com:4433"
+# fingerprint = "aa:bb:cc:dd:..."
+# label = "Relay B"
+
+# Federation: relays we trust inbound connections from
+# [[trusted]]
+# fingerprint = "ee:ff:00:11:..."
+# label = "Relay X"
+
+# Global rooms bridged across all federated peers
+# [[global_rooms]]
+# name = "general"
+
+# Debug: log packet headers for a room ("*" for all)
+# debug_tap = "*"
+"#
+    )
+}

crates/wzp-relay/src/event_log.rs

@@ -0,0 +1,201 @@
+//! JSONL event log for protocol analysis.
+//!
+//! When `--event-log <path>` is set, every media packet emits a structured
+//! event at each decision point (recv, forward, drop, deliver).
+//! Use `wzp-analyzer` to correlate events across multiple relays.
+
+use std::path::PathBuf;
+use std::sync::Arc;
+
+use serde::Serialize;
+use tokio::sync::mpsc;
+use tracing::{error, info};
+
+/// A single protocol event for JSONL output.
+#[derive(Debug, Serialize)]
+pub struct Event {
+    /// ISO 8601 timestamp with microseconds.
+    pub ts: String,
+    /// Event type.
+    pub event: &'static str,
+    /// Room name.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub room: Option<String>,
+    /// Source address or peer label.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub src: Option<String>,
+    /// Packet sequence number.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub seq: Option<u16>,
+    /// Codec identifier.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub codec: Option<String>,
+    /// FEC block ID.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub fec_block: Option<u8>,
+    /// FEC symbol index.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub fec_sym: Option<u8>,
+    /// Is FEC repair packet.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub repair: Option<bool>,
+    /// Payload length in bytes.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub len: Option<usize>,
+    /// Number of recipients.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub to_count: Option<usize>,
+    /// Peer label (for federation events).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub peer: Option<String>,
+    /// Drop/error reason.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub reason: Option<String>,
+    /// Presence action (active/inactive).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub action: Option<String>,
+    /// Participant count (presence events).
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub participants: Option<usize>,
+}
+
+impl Event {
+    fn now() -> String {
+        chrono::Utc::now().format("%Y-%m-%dT%H:%M:%S%.6fZ").to_string()
+    }
+
+    /// Create a minimal event with just type and timestamp.
+    pub fn new(event: &'static str) -> Self {
+        Self {
+            ts: Self::now(),
+            event,
+            room: None,
+            src: None,
+            seq: None,
+            codec: None,
+            fec_block: None,
+            fec_sym: None,
+            repair: None,
+            len: None,
+            to_count: None,
+            peer: None,
+            reason: None,
+            action: None,
+            participants: None,
+        }
+    }
+
+    /// Set room.
+    pub fn room(mut self, room: &str) -> Self { self.room = Some(room.to_string()); self }
+    /// Set source.
+    pub fn src(mut self, src: &str) -> Self { self.src = Some(src.to_string()); self }
+    /// Set packet header fields from a MediaPacket.
+    pub fn packet(mut self, pkt: &wzp_proto::MediaPacket) -> Self {
+        self.seq = Some(pkt.header.seq);
+        self.codec = Some(format!("{:?}", pkt.header.codec_id));
+        self.fec_block = Some(pkt.header.fec_block);
+        self.fec_sym = Some(pkt.header.fec_symbol);
+        self.repair = Some(pkt.header.is_repair);
+        self.len = Some(pkt.payload.len());
+        self
+    }
+    /// Set seq only (when full packet not available).
+    pub fn seq(mut self, seq: u16) -> Self { self.seq = Some(seq); self }
+    /// Set payload length.
+    pub fn len(mut self, len: usize) -> Self { self.len = Some(len); self }
+    /// Set recipient count.
+    pub fn to_count(mut self, n: usize) -> Self { self.to_count = Some(n); self }
+    /// Set peer label.
+    pub fn peer(mut self, peer: &str) -> Self { self.peer = Some(peer.to_string()); self }
+    /// Set drop reason.
+    pub fn reason(mut self, reason: &str) -> Self { self.reason = Some(reason.to_string()); self }
+    /// Set presence action.
+    pub fn action(mut self, action: &str) -> Self { self.action = Some(action.to_string()); self }
+    /// Set participant count.
+    pub fn participants(mut self, n: usize) -> Self { self.participants = Some(n); self }
+}
+
+/// Handle for emitting events. Cheap to clone.
+#[derive(Clone)]
+pub struct EventLog {
+    tx: mpsc::UnboundedSender<Event>,
+}
+
+impl EventLog {
+    /// Emit an event (non-blocking, drops if channel is full).
+    pub fn emit(&self, event: Event) {
+        let _ = self.tx.send(event);
+    }
+}
+
+/// No-op event log for when `--event-log` is not set.
+/// All methods are no-ops that compile to nothing.
+#[derive(Clone)]
+pub struct NoopEventLog;
+
+/// Unified event log handle — either real or no-op.
+#[derive(Clone)]
+pub enum EventLogger {
+    Active(EventLog),
+    Noop,
+}
+
+impl EventLogger {
+    pub fn emit(&self, event: Event) {
+        if let EventLogger::Active(log) = self {
+            log.emit(event);
+        }
+    }
+
+    pub fn is_active(&self) -> bool {
+        matches!(self, EventLogger::Active(_))
+    }
+}
+
+/// Start the event log writer. Returns an `EventLogger` handle.
+pub fn start_event_log(path: Option<PathBuf>) -> EventLogger {
+    match path {
+        Some(path) => {
+            let (tx, rx) = mpsc::unbounded_channel();
+            tokio::spawn(writer_task(path, rx));
+            info!("event log enabled");
+            EventLogger::Active(EventLog { tx })
+        }
+        None => EventLogger::Noop,
+    }
+}
+
+/// Background task that writes events to a JSONL file.
+async fn writer_task(path: PathBuf, mut rx: mpsc::UnboundedReceiver<Event>) {
+    use tokio::io::AsyncWriteExt;
+
+    let file = match tokio::fs::File::create(&path).await {
+        Ok(f) => f,
+        Err(e) => {
+            error!("failed to create event log {}: {e}", path.display());
+            return;
+        }
+    };
+    let mut writer = tokio::io::BufWriter::new(file);
+    let mut count: u64 = 0;
+
+    while let Some(event) = rx.recv().await {
+        match serde_json::to_string(&event) {
+            Ok(json) => {
+                if writer.write_all(json.as_bytes()).await.is_err() { break; }
+                if writer.write_all(b"\n").await.is_err() { break; }
+                count += 1;
+                // Flush every 100 events
+                if count % 100 == 0 {
+                    let _ = writer.flush().await;
+                }
+            }
+            Err(e) => {
+                error!("event log serialize error: {e}");
+            }
+        }
+    }
+
+    let _ = writer.flush().await;
+    info!(events = count, "event log closed");
+}

crates/wzp-relay/src/federation.rs

@@ -0,0 +1,966 @@
+//! Relay federation — global room routing between peer relays.
+//!
+//! Each relay maintains a forwarding table per global room. When a local participant
+//! sends media in a global room, it's forwarded to all peer relays that have the room
+//! active. Incoming federated media is delivered to local participants and optionally
+//! forwarded to other active peers (multi-hop).
+
+use std::collections::{HashMap, HashSet};
+use std::net::SocketAddr;
+use std::sync::Arc;
+use std::time::{Duration, Instant};
+
+use bytes::Bytes;
+use sha2::{Sha256, Digest};
+use tokio::sync::Mutex;
+use tracing::{error, info, warn};
+
+use wzp_proto::{MediaTransport, SignalMessage};
+use wzp_transport::QuinnTransport;
+
+use crate::config::{PeerConfig, TrustedConfig};
+use crate::event_log::{Event, EventLogger};
+use crate::room::{self, FederationMediaOut, RoomEvent, RoomManager};
+
+/// Compute 8-byte room hash for federation datagram tagging.
+pub fn room_hash(room_name: &str) -> [u8; 8] {
+    let h = Sha256::digest(room_name.as_bytes());
+    let mut out = [0u8; 8];
+    out.copy_from_slice(&h[..8]);
+    out
+}
+
+/// Normalize a fingerprint string (remove colons, lowercase).
+fn normalize_fp(fp: &str) -> String {
+    fp.replace(':', "").to_lowercase()
+}
+
+/// Time-based dedup filter for federation datagrams.
+/// Tracks recently seen packets and expires entries older than 2 seconds.
+/// This prevents duplicate delivery when the same packet arrives via
+/// multiple federation paths, while allowing new senders that happen to
+/// reuse the same seq numbers.
+struct Deduplicator {
+    /// Recently seen packet keys with insertion time.
+    entries: HashMap<u64, Instant>,
+    /// Expiry duration.
+    ttl: Duration,
+}
+
+impl Deduplicator {
+    fn new(_capacity: usize) -> Self {
+        Self {
+            entries: HashMap::with_capacity(512),
+            ttl: Duration::from_secs(2),
+        }
+    }
+
+    /// Returns true if this packet is a duplicate (already seen within TTL).
+    fn is_dup(&mut self, room_hash: &[u8; 8], seq: u16, extra: u64) -> bool {
+        let key = u64::from_be_bytes(*room_hash) ^ (seq as u64) ^ extra;
+        let now = Instant::now();
+
+        // Periodic cleanup (every ~256 packets)
+        if self.entries.len() > 256 {
+            self.entries.retain(|_, ts| now.duration_since(*ts) < self.ttl);
+        }
+
+        if let Some(ts) = self.entries.get(&key) {
+            if now.duration_since(*ts) < self.ttl {
+                return true; // seen recently — duplicate
+            }
+        }
+        self.entries.insert(key, now);
+        false
+    }
+}
+
+/// Per-room token bucket rate limiter for federation forwarding.
+struct RateLimiter {
+    /// Max packets per second per room.
+    max_pps: u32,
+    /// Tokens remaining in current window.
+    tokens: u32,
+    /// When the current window started.
+    window_start: Instant,
+}
+
+impl RateLimiter {
+    fn new(max_pps: u32) -> Self {
+        Self {
+            max_pps,
+            tokens: max_pps,
+            window_start: Instant::now(),
+        }
+    }
+
+    /// Returns true if the packet should be allowed through.
+    fn allow(&mut self) -> bool {
+        let elapsed = self.window_start.elapsed();
+        if elapsed >= Duration::from_secs(1) {
+            self.tokens = self.max_pps;
+            self.window_start = Instant::now();
+        }
+        if self.tokens > 0 {
+            self.tokens -= 1;
+            true
+        } else {
+            false
+        }
+    }
+}
+
+/// Active link to a peer relay.
+struct PeerLink {
+    transport: Arc<QuinnTransport>,
+    label: String,
+    /// Global rooms that this peer has reported as active.
+    active_rooms: HashSet<String>,
+    /// Remote participants per room (for federated presence in RoomUpdate).
+    remote_participants: HashMap<String, Vec<wzp_proto::packet::RoomParticipant>>,
+    /// Last time we received any data (signal or media) from this peer.
+    last_seen: Instant,
+}
+
+/// Max federation packets per second per room (0 = unlimited).
+const FEDERATION_RATE_LIMIT_PPS: u32 = 500;
+/// Dedup window size (number of recent packets to remember).
+const DEDUP_WINDOW_SIZE: usize = 4096;
+/// Remote participants are considered stale after this duration with no updates.
+const REMOTE_PARTICIPANT_STALE_SECS: u64 = 15;
+
+/// Manages federation connections and global room forwarding.
+pub struct FederationManager {
+    peers: Vec<PeerConfig>,
+    trusted: Vec<TrustedConfig>,
+    global_rooms: HashSet<String>,
+    room_mgr: Arc<Mutex<RoomManager>>,
+    endpoint: quinn::Endpoint,
+    local_tls_fp: String,
+    metrics: Arc<crate::metrics::RelayMetrics>,
+    /// Active peer connections, keyed by normalized fingerprint.
+    peer_links: Arc<Mutex<HashMap<String, PeerLink>>>,
+    /// Dedup filter for incoming federation datagrams.
+    dedup: Mutex<Deduplicator>,
+    /// Per-room seq counter for federation media delivered to local clients.
+    /// Ensures clients see monotonically increasing seq regardless of federation sender.
+    local_delivery_seq: std::sync::atomic::AtomicU16,
+    /// JSONL event log for protocol analysis.
+    event_log: EventLogger,
+    /// Per-room rate limiters for inbound federation media.
+    rate_limiters: Mutex<HashMap<String, RateLimiter>>,
+}
+
+impl FederationManager {
+    pub fn new(
+        peers: Vec<PeerConfig>,
+        trusted: Vec<TrustedConfig>,
+        global_rooms: HashSet<String>,
+        room_mgr: Arc<Mutex<RoomManager>>,
+        endpoint: quinn::Endpoint,
+        local_tls_fp: String,
+        metrics: Arc<crate::metrics::RelayMetrics>,
+        event_log: EventLogger,
+    ) -> Self {
+        Self {
+            peers,
+            trusted,
+            global_rooms,
+            room_mgr,
+            endpoint,
+            local_tls_fp,
+            metrics,
+            peer_links: Arc::new(Mutex::new(HashMap::new())),
+            dedup: Mutex::new(Deduplicator::new(DEDUP_WINDOW_SIZE)),
+            local_delivery_seq: std::sync::atomic::AtomicU16::new(0),
+            event_log,
+            rate_limiters: Mutex::new(HashMap::new()),
+        }
+    }
+
+    /// Check if a room name (which may be hashed) is a global room.
+    pub fn is_global_room(&self, room: &str) -> bool {
+        self.resolve_global_room(room).is_some()
+    }
+
+    /// Resolve a room name (raw or hashed) to the canonical global room name.
+    /// Returns the configured global room name if it matches.
+    pub fn resolve_global_room(&self, room: &str) -> Option<&str> {
+        // Direct match (raw room name, e.g. Android clients)
+        if self.global_rooms.contains(room) {
+            return Some(self.global_rooms.iter().find(|n| n.as_str() == room).unwrap());
+        }
+        // Hashed match (desktop clients hash room names for SNI privacy)
+        self.global_rooms.iter().find(|name| {
+            wzp_crypto::hash_room_name(name) == room
+        }).map(|s| s.as_str())
+    }
+
+    /// Get the canonical federation room hash for a room.
+    /// Always uses the configured global room name, not the client-provided name.
+    pub fn global_room_hash(&self, room: &str) -> [u8; 8] {
+        if let Some(canonical) = self.resolve_global_room(room) {
+            room_hash(canonical)
+        } else {
+            room_hash(room)
+        }
+    }
+
+    /// Start federation — spawns connection loops + event dispatcher.
+    pub async fn run(self: Arc<Self>) {
+        if self.peers.is_empty() && self.global_rooms.is_empty() {
+            return;
+        }
+        info!(
+            peers = self.peers.len(),
+            global_rooms = self.global_rooms.len(),
+            "federation starting"
+        );
+
+        let mut handles = Vec::new();
+
+        // Per-peer outbound connection loops
+        for peer in &self.peers {
+            let this = self.clone();
+            let peer = peer.clone();
+            handles.push(tokio::spawn(async move {
+                run_peer_loop(this, peer).await;
+            }));
+        }
+
+        // Room event dispatcher
+        let room_events = {
+            let mgr = self.room_mgr.lock().await;
+            mgr.subscribe_events()
+        };
+        let this = self.clone();
+        handles.push(tokio::spawn(async move {
+            run_room_event_dispatcher(this, room_events).await;
+        }));
+
+        // Stale presence sweeper — purges remote participants from dead peers
+        let this = self.clone();
+        handles.push(tokio::spawn(async move {
+            run_stale_presence_sweeper(this).await;
+        }));
+
+        for h in handles {
+            let _ = h.await;
+        }
+    }
+
+    /// Handle an inbound federation connection from a recognized peer.
+    pub async fn handle_inbound(
+        self: &Arc<Self>,
+        transport: Arc<QuinnTransport>,
+        peer_config: PeerConfig,
+    ) {
+        let peer_fp = normalize_fp(&peer_config.fingerprint);
+        let label = peer_config.label.unwrap_or_else(|| peer_config.url.clone());
+        info!(peer = %label, "inbound federation link active");
+        if let Err(e) = run_federation_link(self.clone(), transport, peer_fp, label.clone()).await {
+            warn!(peer = %label, "inbound federation link ended: {e}");
+        }
+    }
+
+    /// Get all remote participants for a room from all peer links.
+    /// Deduplicates by fingerprint (same participant may appear via multiple links).
+    pub async fn get_remote_participants(&self, room: &str) -> Vec<wzp_proto::packet::RoomParticipant> {
+        let canonical = self.resolve_global_room(room);
+        let links = self.peer_links.lock().await;
+        let mut result = Vec::new();
+        for link in links.values() {
+            // Check canonical name
+            if let Some(c) = canonical {
+                if let Some(remote) = link.remote_participants.get(c) {
+                    result.extend(remote.iter().cloned());
+                }
+                // Also check raw room name, but only if different from canonical
+                if c != room {
+                    if let Some(remote) = link.remote_participants.get(room) {
+                        result.extend(remote.iter().cloned());
+                    }
+                }
+            } else {
+                if let Some(remote) = link.remote_participants.get(room) {
+                    result.extend(remote.iter().cloned());
+                }
+            }
+        }
+        // Deduplicate by fingerprint
+        let mut seen = HashSet::new();
+        result.retain(|p| seen.insert(p.fingerprint.clone()));
+        result
+    }
+
+    /// Forward locally-generated media to all connected peers.
+    /// For locally-originated media, we send to ALL peers (they decide whether to deliver).
+    /// For forwarded media (multi-hop), handle_datagram filters by active_rooms.
+    pub async fn forward_to_peers(&self, room_name: &str, room_hash: &[u8; 8], media_data: &Bytes) {
+        let links = self.peer_links.lock().await;
+        if links.is_empty() {
+            return;
+        }
+        for (_fp, link) in links.iter() {
+            let mut tagged = Vec::with_capacity(8 + media_data.len());
+            tagged.extend_from_slice(room_hash);
+            tagged.extend_from_slice(media_data);
+            match link.transport.send_raw_datagram(&tagged) {
+                Ok(()) => {
+                    self.metrics.federation_packets_forwarded
+                        .with_label_values(&[&link.label, "out"]).inc();
+                }
+                Err(e) => warn!(peer = %link.label, "federation send error: {e}"),
+            }
+        }
+    }
+
+    // ── Trust verification (kept from previous implementation) ──
+
+    pub fn find_peer_by_fingerprint(&self, fp: &str) -> Option<&PeerConfig> {
+        self.peers.iter().find(|p| normalize_fp(&p.fingerprint) == normalize_fp(fp))
+    }
+
+    pub fn find_peer_by_addr(&self, addr: SocketAddr) -> Option<&PeerConfig> {
+        let addr_ip = addr.ip();
+        self.peers.iter().find(|p| {
+            p.url.parse::<SocketAddr>()
+                .map(|sa| sa.ip() == addr_ip)
+                .unwrap_or(false)
+        })
+    }
+
+    pub fn find_trusted_by_fingerprint(&self, fp: &str) -> Option<&TrustedConfig> {
+        self.trusted.iter().find(|t| normalize_fp(&t.fingerprint) == normalize_fp(fp))
+    }
+
+    pub fn check_inbound_trust(&self, addr: SocketAddr, hello_fp: &str) -> Option<String> {
+        if let Some(peer) = self.find_peer_by_addr(addr) {
+            return Some(peer.label.clone().unwrap_or_else(|| peer.url.clone()));
+        }
+        if let Some(trusted) = self.find_trusted_by_fingerprint(hello_fp) {
+            return Some(trusted.label.clone().unwrap_or_else(|| hello_fp[..16].to_string()));
+        }
+        None
+    }
+}
+
+// ── Outbound media egress task ──
+
+/// Drains the federation media channel and forwards to active peers.
+pub async fn run_federation_media_egress(
+    fm: Arc<FederationManager>,
+    mut rx: tokio::sync::mpsc::Receiver<FederationMediaOut>,
+) {
+    let mut count: u64 = 0;
+    while let Some(out) = rx.recv().await {
+        count += 1;
+        if count == 1 || count % 250 == 0 {
+            info!(room = %out.room_name, count, "federation egress: forwarding media");
+        }
+        fm.forward_to_peers(&out.room_name, &out.room_hash, &out.data).await;
+    }
+    info!(total = count, "federation egress task ended");
+}
+
+// ── Room event dispatcher ──
+
+/// Watches RoomManager events and sends GlobalRoomActive/Inactive to peers.
+async fn run_room_event_dispatcher(
+    fm: Arc<FederationManager>,
+    mut events: tokio::sync::broadcast::Receiver<RoomEvent>,
+) {
+    loop {
+        match events.recv().await {
+            Ok(RoomEvent::LocalJoin { room }) => {
+                if fm.is_global_room(&room) {
+                    let participants = {
+                        let mgr = fm.room_mgr.lock().await;
+                        mgr.local_participant_list(&room)
+                    };
+                    info!(room = %room, count = participants.len(), "global room now active, announcing to peers");
+                    let msg = SignalMessage::GlobalRoomActive { room, participants };
+                    let links = fm.peer_links.lock().await;
+                    for link in links.values() {
+                        let _ = link.transport.send_signal(&msg).await;
+                    }
+                }
+            }
+            Ok(RoomEvent::LocalLeave { room }) => {
+                if fm.is_global_room(&room) {
+                    info!(room = %room, "global room now inactive, announcing to peers");
+                    let msg = SignalMessage::GlobalRoomInactive { room };
+                    let links = fm.peer_links.lock().await;
+                    for link in links.values() {
+                        let _ = link.transport.send_signal(&msg).await;
+                    }
+                }
+            }
+            Err(tokio::sync::broadcast::error::RecvError::Lagged(n)) => {
+                warn!(missed = n, "room event receiver lagged");
+            }
+            Err(tokio::sync::broadcast::error::RecvError::Closed) => break,
+        }
+    }
+}
+
+// ── Stale presence sweeper ──
+
+/// Periodically checks for stale remote participants and purges them.
+/// This handles the case where a peer link dies without sending GlobalRoomInactive
+/// (e.g., QUIC timeout, network partition, crash).
+async fn run_stale_presence_sweeper(fm: Arc<FederationManager>) {
+    let mut interval = tokio::time::interval(Duration::from_secs(5));
+    loop {
+        interval.tick().await;
+        let stale_threshold = Duration::from_secs(REMOTE_PARTICIPANT_STALE_SECS);
+
+        // Find peers with stale remote_participants whose link is also gone or idle
+        let stale_rooms: Vec<(String, String)> = {
+            let links = fm.peer_links.lock().await;
+            let mut stale = Vec::new();
+            for (fp, link) in links.iter() {
+                if link.last_seen.elapsed() > stale_threshold && !link.remote_participants.is_empty() {
+                    for room in link.remote_participants.keys() {
+                        stale.push((fp.clone(), room.clone()));
+                    }
+                }
+            }
+            stale
+        };
+
+        if stale_rooms.is_empty() {
+            continue;
+        }
+
+        // Purge stale entries and collect affected rooms
+        let mut affected_rooms = HashSet::new();
+        {
+            let mut links = fm.peer_links.lock().await;
+            for (fp, room) in &stale_rooms {
+                if let Some(link) = links.get_mut(fp.as_str()) {
+                    if link.last_seen.elapsed() > stale_threshold {
+                        info!(peer = %link.label, room = %room, "purging stale remote participants (no data for {}s)", link.last_seen.elapsed().as_secs());
+                        link.remote_participants.remove(room);
+                        link.active_rooms.remove(room);
+                        affected_rooms.insert(room.clone());
+                    }
+                }
+            }
+        }
+
+        // Broadcast updated RoomUpdate for affected rooms
+        for room in &affected_rooms {
+            let mgr = fm.room_mgr.lock().await;
+            for local_room in mgr.active_rooms() {
+                if fm.resolve_global_room(&local_room) == fm.resolve_global_room(room) {
+                    let mut all_participants = mgr.local_participant_list(&local_room);
+                    let remote = fm.get_remote_participants(&local_room).await;
+                    all_participants.extend(remote);
+                    let mut seen = HashSet::new();
+                    all_participants.retain(|p| seen.insert(p.fingerprint.clone()));
+                    let update = SignalMessage::RoomUpdate {
+                        count: all_participants.len() as u32,
+                        participants: all_participants,
+                    };
+                    let senders = mgr.local_senders(&local_room);
+                    drop(mgr);
+                    room::broadcast_signal(&senders, &update).await;
+                    info!(room = %room, "swept stale presence — broadcast updated RoomUpdate");
+                    break;
+                }
+            }
+        }
+    }
+}
+
+// ── Peer connection management ──
+
+/// Persistent connection loop for one peer — reconnects with backoff.
+async fn run_peer_loop(fm: Arc<FederationManager>, peer: PeerConfig) {
+    let mut backoff = Duration::from_secs(5);
+    loop {
+        info!(peer_url = %peer.url, label = ?peer.label, "federation: connecting to peer...");
+        match connect_to_peer(&fm, &peer).await {
+            Ok(transport) => {
+                backoff = Duration::from_secs(5);
+                let peer_fp = normalize_fp(&peer.fingerprint);
+                let label = peer.label.clone().unwrap_or_else(|| peer.url.clone());
+                if let Err(e) = run_federation_link(fm.clone(), transport, peer_fp, label).await {
+                    warn!(peer_url = %peer.url, "federation link ended: {e}");
+                }
+            }
+            Err(e) => {
+                warn!(peer_url = %peer.url, backoff_s = backoff.as_secs(), "federation connect failed: {e}");
+            }
+        }
+        tokio::time::sleep(backoff).await;
+        backoff = (backoff * 2).min(Duration::from_secs(300));
+    }
+}
+
+/// Connect to a peer relay and send hello.
+async fn connect_to_peer(fm: &FederationManager, peer: &PeerConfig) -> Result<Arc<QuinnTransport>, anyhow::Error> {
+    let addr: SocketAddr = peer.url.parse()?;
+    let client_cfg = wzp_transport::client_config();
+    let conn = wzp_transport::connect(&fm.endpoint, addr, "_federation", client_cfg).await?;
+    let transport = Arc::new(QuinnTransport::new(conn));
+
+    // Send hello with our TLS fingerprint
+    let hello = SignalMessage::FederationHello {
+        tls_fingerprint: fm.local_tls_fp.clone(),
+    };
+    transport.send_signal(&hello).await
+        .map_err(|e| anyhow::anyhow!("federation hello send failed: {e}"))?;
+
+    info!(peer_url = %peer.url, label = ?peer.label, "federation: connected (hello sent)");
+    Ok(transport)
+}
+
+// ── Federation link (runs on a single QUIC connection) ──
+
+/// Run the federation link: exchange global room state and forward media.
+async fn run_federation_link(
+    fm: Arc<FederationManager>,
+    transport: Arc<QuinnTransport>,
+    peer_fp: String,
+    peer_label: String,
+) -> Result<(), anyhow::Error> {
+    // Register peer link + metrics
+    fm.metrics.federation_peer_status.with_label_values(&[&peer_label]).set(1);
+    {
+        let mut links = fm.peer_links.lock().await;
+        links.insert(peer_fp.clone(), PeerLink {
+            transport: transport.clone(),
+            label: peer_label.clone(),
+            active_rooms: HashSet::new(),
+            remote_participants: HashMap::new(),
+            last_seen: Instant::now(),
+        });
+    }
+
+    // Announce our currently active global rooms to this new peer
+    // Collect all announcements first, then send (avoid holding locks across await)
+    let announcements = {
+        let mgr = fm.room_mgr.lock().await;
+        let active = mgr.active_rooms();
+        let mut msgs = Vec::new();
+
+        // Local rooms
+        for room_name in &active {
+            if fm.is_global_room(room_name) {
+                let participants = mgr.local_participant_list(room_name);
+                info!(peer = %peer_label, room = %room_name, participants = participants.len(), "announcing local global room to new peer");
+                msgs.push(SignalMessage::GlobalRoomActive { room: room_name.clone(), participants });
+            }
+        }
+
+        // Remote rooms from OTHER peers (for multi-hop propagation)
+        let links = fm.peer_links.lock().await;
+        for (fp, link) in links.iter() {
+            if fp != &peer_fp {
+                for (room, participants) in &link.remote_participants {
+                    if fm.is_global_room(room) {
+                        info!(peer = %peer_label, room = %room, via = %link.label, "propagating remote room to new peer");
+                        msgs.push(SignalMessage::GlobalRoomActive {
+                            room: room.clone(),
+                            participants: participants.clone(),
+                        });
+                    }
+                }
+            }
+        }
+        msgs
+    };
+    for msg in &announcements {
+        let _ = transport.send_signal(msg).await;
+    }
+
+    // Three concurrent tasks: signal recv + media recv + RTT monitor
+    let signal_transport = transport.clone();
+    let media_transport = transport.clone();
+    let rtt_transport = transport.clone();
+    let fm_signal = fm.clone();
+    let fm_media = fm.clone();
+    let fm_rtt = fm.clone();
+    let peer_fp_signal = peer_fp.clone();
+    let peer_fp_media = peer_fp.clone();
+    let label_signal = peer_label.clone();
+    let label_rtt = peer_label.clone();
+
+    let signal_task = async move {
+        loop {
+            match signal_transport.recv_signal().await {
+                Ok(Some(msg)) => {
+                    handle_signal(&fm_signal, &peer_fp_signal, &label_signal, msg).await;
+                }
+                Ok(None) => break,
+                Err(e) => {
+                    error!(peer = %label_signal, "federation signal error: {e}");
+                    break;
+                }
+            }
+        }
+    };
+
+    let peer_label_media = peer_label.clone();
+    let media_task = async move {
+        let mut media_count: u64 = 0;
+        loop {
+            match media_transport.connection().read_datagram().await {
+                Ok(data) => {
+                    media_count += 1;
+                    if media_count == 1 || media_count % 250 == 0 {
+                        info!(peer = %peer_label_media, media_count, len = data.len(), "federation: received datagram");
+                    }
+                    handle_datagram(&fm_media, &peer_fp_media, data).await;
+                }
+                Err(e) => {
+                    info!(peer = %peer_label_media, "federation media task ended: {e}");
+                    break;
+                }
+            }
+        }
+    };
+
+    // RTT monitor: periodically sample QUIC RTT for this peer
+    let rtt_task = async move {
+        loop {
+            tokio::time::sleep(Duration::from_secs(5)).await;
+            let rtt_ms = rtt_transport.connection().stats().path.rtt.as_millis() as f64;
+        }
+    };
+
+    tokio::select! {
+        _ = signal_task => {}
+        _ = media_task => {}
+        _ = rtt_task => {}
+    }
+
+    // Cleanup: remove peer link + metrics
+    fm.metrics.federation_peer_status.with_label_values(&[&peer_label]).set(0);
+    {
+        let mut links = fm.peer_links.lock().await;
+        links.remove(&peer_fp);
+    }
+    info!(peer = %peer_label, "federation link ended");
+
+    Ok(())
+}
+
+/// Handle an incoming federation signal.
+async fn handle_signal(
+    fm: &Arc<FederationManager>,
+    peer_fp: &str,
+    peer_label: &str,
+    msg: SignalMessage,
+) {
+    // Update last_seen for this peer
+    {
+        let mut links = fm.peer_links.lock().await;
+        if let Some(link) = links.get_mut(peer_fp) {
+            link.last_seen = Instant::now();
+        }
+    }
+
+    match msg {
+        SignalMessage::GlobalRoomActive { room, participants } => {
+            if fm.is_global_room(&room) {
+                info!(peer = %peer_label, room = %room, remote_participants = participants.len(), "peer has global room active");
+                let mut links = fm.peer_links.lock().await;
+                if let Some(link) = links.get_mut(peer_fp) {
+                    link.active_rooms.insert(room.clone());
+                }
+                // Update active rooms metric
+                let total: usize = links.values().map(|l| l.active_rooms.len()).sum();
+                fm.metrics.federation_active_rooms.set(total as i64);
+                if let Some(link) = links.get_mut(peer_fp) {
+                    // Tag remote participants with their relay label
+                    let tagged: Vec<_> = participants.iter().map(|p| {
+                        let mut tagged = p.clone();
+                        if tagged.relay_label.is_none() {
+                            tagged.relay_label = Some(link.label.clone());
+                        }
+                        tagged
+                    }).collect();
+                    link.remote_participants.insert(room.clone(), tagged);
+                }
+                // Propagate to other peers (with relay labels preserved)
+                let tagged_for_propagation = if let Some(link) = links.get(peer_fp) {
+                    let label = link.label.clone();
+                    participants.iter().map(|p| {
+                        let mut t = p.clone();
+                        if t.relay_label.is_none() {
+                            t.relay_label = Some(label.clone());
+                        }
+                        t
+                    }).collect::<Vec<_>>()
+                } else {
+                    participants.clone()
+                };
+                for (fp, link) in links.iter() {
+                    if fp != peer_fp {
+                        let _ = link.transport.send_signal(&SignalMessage::GlobalRoomActive {
+                            room: room.clone(),
+                            participants: tagged_for_propagation.clone(),
+                        }).await;
+                    }
+                }
+                drop(links);
+
+                // Broadcast updated RoomUpdate to local clients in this room
+                // Find the local room name (may be hashed or raw)
+                let mgr = fm.room_mgr.lock().await;
+                for local_room in mgr.active_rooms() {
+                    if fm.is_global_room(&local_room) && fm.resolve_global_room(&local_room) == fm.resolve_global_room(&room) {
+                        // Build merged participant list: local + all remote (deduped)
+                        let mut all_participants = mgr.local_participant_list(&local_room);
+                        let links = fm.peer_links.lock().await;
+                        for link in links.values() {
+                            if let Some(canonical) = fm.resolve_global_room(&local_room) {
+                                if let Some(remote) = link.remote_participants.get(canonical) {
+                                    all_participants.extend(remote.iter().cloned());
+                                }
+                                // Also check raw room name, but only if different from canonical
+                                if canonical != local_room {
+                                    if let Some(remote) = link.remote_participants.get(&local_room) {
+                                        all_participants.extend(remote.iter().cloned());
+                                    }
+                                }
+                            }
+                        }
+                        // Deduplicate by fingerprint
+                        let mut seen = HashSet::new();
+                        all_participants.retain(|p| seen.insert(p.fingerprint.clone()));
+                        let update = SignalMessage::RoomUpdate {
+                            count: all_participants.len() as u32,
+                            participants: all_participants,
+                        };
+                        let senders = mgr.local_senders(&local_room);
+                        drop(links);
+                        drop(mgr);
+                        room::broadcast_signal(&senders, &update).await;
+                        break;
+                    }
+                }
+            }
+        }
+        SignalMessage::GlobalRoomInactive { room } => {
+            info!(peer = %peer_label, room = %room, "peer global room now inactive");
+            let mut links = fm.peer_links.lock().await;
+            if let Some(link) = links.get_mut(peer_fp) {
+                link.active_rooms.remove(&room);
+                // Clear remote participants for this peer+room
+                link.remote_participants.remove(&room);
+                // Also try canonical name
+                if let Some(canonical) = fm.resolve_global_room(&room) {
+                    link.remote_participants.remove(canonical);
+                }
+            }
+
+            // Update active rooms metric
+            let total: usize = links.values().map(|l| l.active_rooms.len()).sum();
+            fm.metrics.federation_active_rooms.set(total as i64);
+
+            // Build remaining remote participants (from all peers except the one going inactive)
+            let remaining_remote: Vec<wzp_proto::packet::RoomParticipant> = {
+                let canonical = fm.resolve_global_room(&room);
+                let mut result = Vec::new();
+                for (fp, link) in links.iter() {
+                    if fp == peer_fp { continue; }
+                    if let Some(c) = canonical {
+                        if let Some(remote) = link.remote_participants.get(c) {
+                            result.extend(remote.iter().cloned());
+                        }
+                    }
+                }
+                let mut seen = HashSet::new();
+                result.retain(|p| seen.insert(p.fingerprint.clone()));
+                result
+            };
+
+            // Propagate to other peers: send updated GlobalRoomActive with revised list,
+            // or GlobalRoomInactive if no participants remain anywhere
+            let local_active = {
+                let mgr = fm.room_mgr.lock().await;
+                mgr.active_rooms().iter().any(|r| fm.resolve_global_room(r) == fm.resolve_global_room(&room))
+            };
+            let has_remaining = !remaining_remote.is_empty() || local_active;
+
+            // Collect peer transports to send to (avoid holding lock across await)
+            let peer_sends: Vec<_> = links.iter()
+                .filter(|(fp, _)| *fp != peer_fp)
+                .map(|(_, link)| link.transport.clone())
+                .collect();
+            drop(links);
+
+            if has_remaining {
+                // Send updated participant list to other peers
+                let mut updated_participants = remaining_remote.clone();
+                if local_active {
+                    let mgr = fm.room_mgr.lock().await;
+                    for local_room in mgr.active_rooms() {
+                        if fm.resolve_global_room(&local_room) == fm.resolve_global_room(&room) {
+                            updated_participants.extend(mgr.local_participant_list(&local_room));
+                            break;
+                        }
+                    }
+                }
+                let msg = SignalMessage::GlobalRoomActive {
+                    room: room.clone(),
+                    participants: updated_participants,
+                };
+                for transport in &peer_sends {
+                    let _ = transport.send_signal(&msg).await;
+                }
+            } else {
+                // No participants left anywhere — propagate inactive
+                let msg = SignalMessage::GlobalRoomInactive { room: room.clone() };
+                for transport in &peer_sends {
+                    let _ = transport.send_signal(&msg).await;
+                }
+            }
+
+            // Broadcast updated RoomUpdate to local clients (remote participant removed)
+            let mgr = fm.room_mgr.lock().await;
+            for local_room in mgr.active_rooms() {
+                if fm.is_global_room(&local_room) && fm.resolve_global_room(&local_room) == fm.resolve_global_room(&room) {
+                    let mut all_participants = mgr.local_participant_list(&local_room);
+                    all_participants.extend(remaining_remote.iter().cloned());
+                    // Deduplicate by fingerprint
+                    let mut seen = HashSet::new();
+                    all_participants.retain(|p| seen.insert(p.fingerprint.clone()));
+                    let update = SignalMessage::RoomUpdate {
+                        count: all_participants.len() as u32,
+                        participants: all_participants,
+                    };
+                    let senders = mgr.local_senders(&local_room);
+                    drop(mgr);
+                    room::broadcast_signal(&senders, &update).await;
+                    info!(room = %room, "broadcast updated presence (remote participant removed)");
+                    break;
+                }
+            }
+        }
+        _ => {} // ignore other signals
+    }
+}
+
+/// Handle an incoming federation datagram (room-hash-tagged media).
+async fn handle_datagram(
+    fm: &Arc<FederationManager>,
+    source_peer_fp: &str,
+    data: Bytes,
+) {
+    if data.len() < 12 { return; } // 8-byte hash + min packet
+
+    let mut rh = [0u8; 8];
+    rh.copy_from_slice(&data[..8]);
+    let media_bytes = data.slice(8..);
+
+    let pkt = match wzp_proto::MediaPacket::from_bytes(media_bytes.clone()) {
+        Some(pkt) => pkt,
+        None => {
+            fm.event_log.emit(Event::new("federation_ingress_malformed").len(data.len()));
+            return;
+        }
+    };
+
+    // Event log: federation ingress
+    let peer_label = {
+        let links = fm.peer_links.lock().await;
+        links.get(source_peer_fp).map(|l| l.label.clone()).unwrap_or_default()
+    };
+    fm.event_log.emit(Event::new("federation_ingress").packet(&pkt).peer(&peer_label));
+
+    // Count inbound federation packet + update last_seen
+    fm.metrics.federation_packets_forwarded
+        .with_label_values(&[source_peer_fp, "in"]).inc();
+    {
+        let mut links = fm.peer_links.lock().await;
+        if let Some(link) = links.get_mut(source_peer_fp) {
+            link.last_seen = Instant::now();
+        }
+    }
+
+    // Dedup: drop packets we've already seen (multi-path duplicates).
+    // Key uses a hash of the actual payload bytes — unique per Opus frame,
+    // so different senders with the same seq/timestamp never collide.
+    let payload_hash = {
+        let mut h = 0u64;
+        for (i, &b) in media_bytes.iter().take(16).enumerate() {
+            h ^= (b as u64) << ((i % 8) * 8);
+        }
+        h
+    };
+    {
+        let mut dedup = fm.dedup.lock().await;
+        if dedup.is_dup(&rh, pkt.header.seq, payload_hash) {
+            fm.event_log.emit(Event::new("dedup_drop").seq(pkt.header.seq).peer(&peer_label));
+            return;
+        }
+    }
+
+    // Find room by hash — check local rooms AND global room config
+    let room_name = {
+        let mgr = fm.room_mgr.lock().await;
+        let active = mgr.active_rooms();
+        // First: check local rooms (has participants)
+        active.iter().find(|r| room_hash(r) == rh).cloned()
+            .or_else(|| active.iter().find(|r| fm.global_room_hash(r) == rh).cloned())
+            // Second: check global room config (hub relay may have no local participants)
+            .or_else(|| {
+                fm.global_rooms.iter().find(|name| room_hash(name) == rh).cloned()
+            })
+    };
+
+    let room_name = match room_name {
+        Some(r) => r,
+        None => {
+            fm.event_log.emit(Event::new("room_not_found").seq(pkt.header.seq).peer(&peer_label));
+            return;
+        }
+    };
+
+    // Rate limit per room
+    if FEDERATION_RATE_LIMIT_PPS > 0 {
+        let mut limiters = fm.rate_limiters.lock().await;
+        let limiter = limiters.entry(room_name.clone())
+            .or_insert_with(|| RateLimiter::new(FEDERATION_RATE_LIMIT_PPS));
+        if !limiter.allow() {
+            fm.event_log.emit(Event::new("rate_limit_drop").room(&room_name).seq(pkt.header.seq));
+            return;
+        }
+    }
+
+    // Deliver to all local participants — forward the raw bytes as-is.
+    // The original sender's MediaPacket is preserved exactly (no re-serialization).
+    let locals = {
+        let mgr = fm.room_mgr.lock().await;
+        mgr.local_senders(&room_name)
+    };
+    for sender in &locals {
+        match sender {
+            room::ParticipantSender::Quic(t) => {
+                if let Err(e) = t.send_raw_datagram(&media_bytes) {
+                    fm.event_log.emit(Event::new("local_deliver_error").room(&room_name).seq(pkt.header.seq).reason(&e.to_string()));
+                    warn!("federation local delivery error: {e}");
+                }
+            }
+            room::ParticipantSender::WebSocket(_) => { let _ = sender.send_raw(&pkt.payload).await; }
+        }
+    }
+    fm.event_log.emit(Event::new("local_deliver").room(&room_name).seq(pkt.header.seq).to_count(locals.len()));
+
+    // Multi-hop: forward to ALL other connected peers (not the source)
+    // Don't filter by active_rooms — the receiving peer decides whether to deliver
+    let links = fm.peer_links.lock().await;
+    for (fp, link) in links.iter() {
+        if fp != source_peer_fp {
+            let mut tagged = Vec::with_capacity(8 + media_bytes.len());
+            tagged.extend_from_slice(&rh);
+            tagged.extend_from_slice(&media_bytes);
+            let _ = link.transport.send_raw_datagram(&tagged);
+        }
+    }
+}

crates/wzp-relay/src/handshake.rs

@@ -78,31 +78,26 @@ pub async fn accept_handshake(
     };
     transport.send_signal(&answer).await?;
 
-    // Derive caller fingerprint from their identity public key (first 8 bytes as hex)
-    let caller_fp = caller_identity_pub[..8]
-        .iter()
-        .map(|b| format!("{b:02x}"))
-        .collect::<String>();
+    // Derive caller fingerprint: SHA-256(Ed25519 pub)[:16], formatted as xxxx:xxxx:...
+    // Must match the format used in signal registration and presence.
+    let caller_fp = {
+        use sha2::{Sha256, Digest};
+        let hash = Sha256::digest(&caller_identity_pub);
+        let fp = wzp_crypto::Fingerprint([
+            hash[0], hash[1], hash[2], hash[3], hash[4], hash[5], hash[6], hash[7],
+            hash[8], hash[9], hash[10], hash[11], hash[12], hash[13], hash[14], hash[15],
+        ]);
+        fp.to_string()
+    };
 
     Ok((session, chosen_profile, caller_fp, caller_alias))
 }
 
 /// Select the best quality profile from those the caller supports.
 fn choose_profile(supported: &[QualityProfile]) -> QualityProfile {
-    // Prefer higher-quality profiles. Use GOOD as default if supported list is empty.
-    if supported.is_empty() {
-        return QualityProfile::GOOD;
-    }
-    // Pick the profile with the highest bitrate.
-    supported
-        .iter()
-        .max_by(|a, b| {
-            a.total_bitrate_kbps()
-                .partial_cmp(&b.total_bitrate_kbps())
-                .unwrap_or(std::cmp::Ordering::Equal)
-        })
-        .copied()
-        .unwrap_or(QualityProfile::GOOD)
+    // Cap at GOOD (24k) for now — studio tiers (32k/48k/64k) not yet tested
+    // for federation reliability (large packets may exceed path MTU).
+    QualityProfile::GOOD
 }
 
 #[cfg(test)]

crates/wzp-relay/src/lib.rs

@@ -8,7 +8,11 @@
 //! quality transitions.
 
 pub mod auth;
+pub mod call_registry;
 pub mod config;
+pub mod event_log;
+pub mod federation;
+pub mod signal_hub;
 pub mod handshake;
 pub mod metrics;
 pub mod pipeline;

crates/wzp-relay/src/main.rs

@@ -15,7 +15,7 @@ use std::time::Duration;
 use tokio::sync::Mutex;
 use tracing::{error, info, warn};
 
-use wzp_proto::MediaTransport;
+use wzp_proto::{MediaTransport, SignalMessage};
 use wzp_relay::config::RelayConfig;
 use wzp_relay::metrics::RelayMetrics;
 use wzp_relay::pipeline::{PipelineConfig, RelayPipeline};
@@ -23,12 +23,54 @@ use wzp_relay::presence::PresenceRegistry;
 use wzp_relay::room::{self, RoomManager};
 use wzp_relay::session_mgr::SessionManager;
 
-fn parse_args() -> RelayConfig {
-    let mut config = RelayConfig::default();
+/// Parsed CLI result — config + identity path.
+struct CliResult {
+    config: RelayConfig,
+    identity_path: Option<String>,
+    config_file: Option<String>,
+    config_needs_create: bool,
+}
+
+fn parse_args() -> CliResult {
     let args: Vec<String> = std::env::args().collect();
+
+    // First pass: extract --config and --identity
+    let mut config_file = None;
+    let mut identity_path = None;
     let mut i = 1;
     while i < args.len() {
         match args[i].as_str() {
+            "--config" | "-c" => { i += 1; config_file = args.get(i).cloned(); }
+            "--identity" | "-i" => { i += 1; identity_path = args.get(i).cloned(); }
+            _ => {}
+        }
+        i += 1;
+    }
+
+    // Track if we need to create the config after identity is known
+    let config_needs_create = config_file.as_ref().map(|p| !std::path::Path::new(p).exists()).unwrap_or(false);
+
+    let mut config = if let Some(ref path) = config_file {
+        if config_needs_create {
+            // Will be re-created with personalized info after identity is loaded
+            RelayConfig::default()
+        } else {
+            wzp_relay::config::load_config(path)
+                .unwrap_or_else(|e| {
+                    eprintln!("failed to load config from {path}: {e}");
+                    std::process::exit(1);
+                })
+        }
+    } else {
+        RelayConfig::default()
+    };
+
+    // CLI flags override config file values
+    let mut i = 1;
+    while i < args.len() {
+        match args[i].as_str() {
+            "--config" | "-c" => { i += 1; } // already handled
+            "--identity" | "-i" => { i += 1; } // already handled
             "--listen" => {
                 i += 1;
                 config.listen_addr = args.get(i).expect("--listen requires an address")
@@ -81,6 +123,28 @@ fn parse_args() -> RelayConfig {
                     args.get(i).expect("--static-dir requires a directory path").to_string(),
                 );
             }
+            "--global-room" => {
+                i += 1;
+                config.global_rooms.push(wzp_relay::config::GlobalRoomConfig {
+                    name: args.get(i).expect("--global-room requires a room name").to_string(),
+                });
+            }
+            "--debug-tap" => {
+                i += 1;
+                config.debug_tap = Some(
+                    args.get(i).expect("--debug-tap requires a room name (or '*' for all)").to_string(),
+                );
+            }
+            "--event-log" => {
+                i += 1;
+                config.event_log = Some(
+                    args.get(i).expect("--event-log requires a file path").to_string(),
+                );
+            }
+            "--version" | "-V" => {
+                println!("wzp-relay {}", env!("WZP_BUILD_HASH"));
+                std::process::exit(0);
+            }
             "--mesh-status" => {
                 // Print mesh table from a fresh registry and exit.
                 // In practice this is useful after the relay has been running;
@@ -90,9 +154,11 @@ fn parse_args() -> RelayConfig {
                 std::process::exit(0);
             }
             "--help" | "-h" => {
-                eprintln!("Usage: wzp-relay [--listen <addr>] [--remote <addr>] [--auth-url <url>] [--metrics-port <port>] [--probe <addr>]... [--probe-mesh] [--mesh-status]");
+                eprintln!("Usage: wzp-relay [--config <path>] [--listen <addr>] [--remote <addr>] [--auth-url <url>] [--metrics-port <port>] [--probe <addr>]... [--probe-mesh] [--mesh-status]");
                 eprintln!();
                 eprintln!("Options:");
+                eprintln!("  -c, --config <path>    Load config from TOML file (creates example if missing)");
+                eprintln!("  -i, --identity <path>  Identity file path (creates if missing, uses OsRng)");
                 eprintln!("  --listen <addr>        Listen address (default: 0.0.0.0:4433)");
                 eprintln!("  --remote <addr>        Remote relay for forwarding (disables room mode)");
                 eprintln!("  --auth-url <url>       featherChat auth endpoint (e.g., https://chat.example.com/v1/auth/validate)");
@@ -102,6 +168,8 @@ fn parse_args() -> RelayConfig {
                 eprintln!("  --probe-mesh           Enable mesh mode (mark config flag, probes all --probe targets).");
                 eprintln!("  --mesh-status          Print mesh health table and exit (diagnostic).");
                 eprintln!("  --trunking             Enable trunk batching for outgoing media in room mode.");
+                eprintln!("  --global-room <name>   Declare a room as global (bridged across federation). Repeatable.");
+                eprintln!("  --debug-tap <room>     Log packet headers for a room ('*' for all rooms).");
                 eprintln!("  --ws-port <port>       WebSocket listener port for browser clients (e.g., 8080).");
                 eprintln!("  --static-dir <dir>     Directory to serve static files from (HTML/JS/WASM).");
                 eprintln!();
@@ -116,7 +184,7 @@ fn parse_args() -> RelayConfig {
         }
         i += 1;
     }
-    config
+    CliResult { config, identity_path, config_file, config_needs_create }
 }
 
 struct RelayStats {
@@ -184,10 +252,29 @@ async fn run_downstream(
     }
 }
 
+/// Detect a non-loopback IP address from local interfaces.
+/// Prefers public IPs over private (10.x, 172.16-31.x, 192.168.x).
+fn detect_public_ip() -> Option<String> {
+    use std::net::UdpSocket;
+    // Connect to a public address to find our outbound IP (doesn't actually send anything)
+    if let Ok(socket) = UdpSocket::bind("0.0.0.0:0") {
+        if socket.connect("8.8.8.8:80").is_ok() {
+            if let Ok(addr) = socket.local_addr() {
+                return Some(addr.ip().to_string());
+            }
+        }
+    }
+    None
+}
+
+/// Build-time git hash, set by build.rs or env.
+const BUILD_GIT_HASH: &str = env!("WZP_BUILD_HASH");
+
 #[tokio::main]
 async fn main() -> anyhow::Result<()> {
-    let config = parse_args();
+    let CliResult { mut config, identity_path, config_file, config_needs_create } = parse_args();
     tracing_subscriber::fmt().init();
+    info!(version = BUILD_GIT_HASH, "wzp-relay build");
     rustls::crypto::ring::default_provider()
         .install_default()
         .expect("failed to install rustls crypto provider");
@@ -207,43 +294,88 @@ async fn main() -> anyhow::Result<()> {
         tokio::spawn(wzp_relay::metrics::serve_metrics(port, m, p, rr));
     }
 
-    // Load or generate relay identity — persisted in ~/.wzp/relay-identity
+    // Load or generate relay identity
     let relay_seed = {
-        let config_dir = dirs::home_dir()
-            .unwrap_or_else(|| std::path::PathBuf::from("."))
-            .join(".wzp");
-        let identity_path = config_dir.join("relay-identity");
-        if identity_path.exists() {
-            if let Ok(hex) = std::fs::read_to_string(&identity_path) {
+        let id_path = match identity_path {
+            Some(ref p) => std::path::PathBuf::from(p),
+            None => dirs::home_dir()
+                .unwrap_or_else(|| std::path::PathBuf::from("."))
+                .join(".wzp")
+                .join("relay-identity"),
+        };
+        if id_path.exists() {
+            if let Ok(hex) = std::fs::read_to_string(&id_path) {
                 if let Ok(s) = wzp_crypto::Seed::from_hex(hex.trim()) {
-                    info!("loaded relay identity from {}", identity_path.display());
+                    info!("loaded relay identity from {}", id_path.display());
                     s
                 } else {
-                    warn!("corrupt relay identity file, generating new");
+                    warn!("corrupt identity file {}, generating new", id_path.display());
                     let s = wzp_crypto::Seed::generate();
                     let hex: String = s.0.iter().map(|b| format!("{b:02x}")).collect();
-                    let _ = std::fs::write(&identity_path, &hex);
+                    let _ = std::fs::write(&id_path, &hex);
                     s
                 }
             } else {
                 let s = wzp_crypto::Seed::generate();
                 let hex: String = s.0.iter().map(|b| format!("{b:02x}")).collect();
-                let _ = std::fs::write(&identity_path, &hex);
+                let _ = std::fs::write(&id_path, &hex);
                 s
             }
         } else {
             let s = wzp_crypto::Seed::generate();
-            let _ = std::fs::create_dir_all(&config_dir);
+            if let Some(parent) = id_path.parent() {
+                let _ = std::fs::create_dir_all(parent);
+            }
             let hex: String = s.0.iter().map(|b| format!("{b:02x}")).collect();
-            let _ = std::fs::write(&identity_path, &hex);
-            info!("generated relay identity at {}", identity_path.display());
+            let _ = std::fs::write(&id_path, &hex);
+            info!("generated relay identity at {}", id_path.display());
             s
         }
     };
     let relay_fp = relay_seed.derive_identity().public_identity().fingerprint;
     info!(addr = %config.listen_addr, fingerprint = %relay_fp, "WarzonePhone relay starting");
 
-    let (server_config, _cert) = wzp_transport::server_config();
+    let (server_config, cert_der) = wzp_transport::server_config_from_seed(&relay_seed.0);
+    let tls_fp = wzp_transport::tls_fingerprint(&cert_der);
+    info!(tls_fingerprint = %tls_fp, "TLS certificate (deterministic from relay identity)");
+
+    // Create personalized config file if it was missing
+    let public_ip = detect_public_ip();
+    if config_needs_create {
+        if let Some(ref path) = config_file {
+            let info = wzp_relay::config::RelayInfo {
+                listen_addr: config.listen_addr.to_string(),
+                tls_fingerprint: tls_fp.clone(),
+                public_ip: public_ip.clone(),
+            };
+            if let Err(e) = wzp_relay::config::load_or_create_config(path, Some(&info)) {
+                warn!("failed to create config: {e}");
+            }
+        }
+    }
+
+    // Print federation hint with our public IP + listen port + TLS fingerprint
+    let listen_port = config.listen_addr.port();
+    if let Some(ip) = &public_ip {
+        info!("federation: to peer with this relay, add to relay.toml:");
+        info!("  [[peers]]");
+        info!("  url = \"{ip}:{listen_port}\"");
+        info!("  fingerprint = \"{tls_fp}\"");
+    }
+
+    // Log configured peers and trusted relays
+    if !config.peers.is_empty() {
+        info!(count = config.peers.len(), "federation peers configured");
+        for p in &config.peers {
+            info!(url = %p.url, label = ?p.label, "  peer");
+        }
+    }
+    if !config.trusted.is_empty() {
+        info!(count = config.trusted.len(), "trusted relays configured");
+        for t in &config.trusted {
+            info!(fingerprint = %t.fingerprint, label = ?t.label, "  trusted");
+        }
+    }
     let endpoint = wzp_transport::create_endpoint(config.listen_addr, Some(server_config))?;
 
     // Forward mode
@@ -261,9 +393,41 @@ async fn main() -> anyhow::Result<()> {
     // Room manager (room mode only)
     let room_mgr = Arc::new(Mutex::new(RoomManager::new()));
 
+    // Event log for protocol analysis
+    let event_log = wzp_relay::event_log::start_event_log(
+        config.event_log.as_ref().map(std::path::PathBuf::from)
+    );
+
+    // Federation manager
+    let global_room_set: std::collections::HashSet<String> = config.global_rooms.iter()
+        .map(|g| g.name.clone())
+        .collect();
+
+    let federation_mgr = if !config.peers.is_empty() || !config.trusted.is_empty() || !global_room_set.is_empty() {
+        let fm = Arc::new(wzp_relay::federation::FederationManager::new(
+            config.peers.clone(),
+            config.trusted.clone(),
+            global_room_set.clone(),
+            room_mgr.clone(),
+            endpoint.clone(),
+            tls_fp.clone(),
+            metrics.clone(),
+            event_log.clone(),
+        ));
+        let fm_run = fm.clone();
+        tokio::spawn(async move { fm_run.run().await });
+        Some(fm)
+    } else {
+        None
+    };
+
     // Session manager — enforces max concurrent sessions
     let session_mgr = Arc::new(Mutex::new(SessionManager::new(config.max_sessions)));
 
+    // Signal hub + call registry for direct 1:1 calls
+    let signal_hub = Arc::new(Mutex::new(wzp_relay::signal_hub::SignalHub::new()));
+    let call_registry = Arc::new(Mutex::new(wzp_relay::call_registry::CallRegistry::new()));
+
     // Spawn inter-relay health probes via ProbeMesh coordinator
     if !config.probe_targets.is_empty() {
         let mesh = wzp_relay::probe::ProbeMesh::new(
@@ -298,6 +462,15 @@ async fn main() -> anyhow::Result<()> {
     } else {
         info!("auth disabled — any client can connect (use --auth-url to enable)");
     }
+    if !config.global_rooms.is_empty() {
+        info!(count = config.global_rooms.len(), "global rooms configured");
+        for g in &config.global_rooms {
+            info!(name = %g.name, "  global room");
+        }
+    }
+    if let Some(ref tap) = config.debug_tap {
+        info!(filter = %tap, "debug tap enabled — logging packet headers");
+    }
 
     info!("Listening for connections...");
 
@@ -314,8 +487,13 @@ async fn main() -> anyhow::Result<()> {
         let relay_seed_bytes = relay_seed.0;
         let metrics = metrics.clone();
         let trunking_enabled = config.trunking_enabled;
+        let debug_tap = config.debug_tap.as_ref().map(|filter| room::DebugTap { room_filter: filter.clone() });
         let presence = presence.clone();
         let route_resolver = route_resolver.clone();
+        let federation_mgr = federation_mgr.clone();
+        let signal_hub = signal_hub.clone();
+        let call_registry = call_registry.clone();
+        let listen_addr_str = config.listen_addr.to_string();
 
         tokio::spawn(async move {
             let addr = connection.remote_address();
@@ -331,12 +509,22 @@ async fn main() -> anyhow::Result<()> {
             let transport = Arc::new(wzp_transport::QuinnTransport::new(connection));
 
             // Ping connections: client just measures QUIC connect RTT.
-            // No handshake, no streams — client closes immediately after connecting.
             if room_name == "ping" {
                 info!(%addr, "ping connection (RTT probe)");
                 return;
             }
 
+            // Version query: respond with build hash over a uni stream.
+            if room_name == "version" {
+                if let Ok(mut send) = transport.connection().open_uni().await {
+                    let _ = send.write_all(BUILD_GIT_HASH.as_bytes()).await;
+                    let _ = send.finish();
+                    // Wait for client to read before closing
+                    tokio::time::sleep(std::time::Duration::from_millis(100)).await;
+                }
+                return;
+            }
+
             // Probe connections use SNI "_probe" to identify themselves.
             // They skip auth + handshake and just do Ping->Pong + presence gossip.
             if room_name == "_probe" {
@@ -423,6 +611,294 @@ async fn main() -> anyhow::Result<()> {
                 return;
             }
 
+            // Federation connections use SNI "_federation"
+            if room_name == "_federation" {
+                if let Some(ref fm) = federation_mgr {
+                    // Wait for FederationHello to identify the connecting relay
+                    let hello_fp = match tokio::time::timeout(
+                        std::time::Duration::from_secs(5),
+                        transport.recv_signal(),
+                    ).await {
+                        Ok(Ok(Some(wzp_proto::SignalMessage::FederationHello { tls_fingerprint }))) => tls_fingerprint,
+                        _ => {
+                            warn!(%addr, "federation: no hello received, closing");
+                            return;
+                        }
+                    };
+
+                    if let Some(label) = fm.check_inbound_trust(addr, &hello_fp) {
+                        let peer_config = wzp_relay::config::PeerConfig {
+                            url: addr.to_string(),
+                            fingerprint: hello_fp,
+                            label: Some(label.clone()),
+                        };
+                        let fm = fm.clone();
+                        info!(%addr, label = %label, "inbound federation accepted (trusted)");
+                        fm.handle_inbound(transport, peer_config).await;
+                    } else {
+                        warn!(%addr, fp = %hello_fp, "unknown relay wants to federate");
+                        info!("  to accept, add to relay.toml:");
+                        info!("  [[trusted]]");
+                        info!("  fingerprint = \"{hello_fp}\"");
+                        info!("  label = \"Relay at {addr}\"");
+                    }
+                } else {
+                    info!(%addr, "federation connection rejected (no federation configured)");
+                }
+                return;
+            }
+
+            // Direct calling: persistent signaling connection
+            if room_name == "_signal" {
+                info!(%addr, "signal connection");
+
+                // Optional auth
+                let auth_fp: Option<String> = if let Some(ref url) = auth_url {
+                    match transport.recv_signal().await {
+                        Ok(Some(SignalMessage::AuthToken { token })) => {
+                            match wzp_relay::auth::validate_token(url, &token).await {
+                                Ok(client) => Some(client.fingerprint),
+                                Err(e) => {
+                                    error!(%addr, "signal auth failed: {e}");
+                                    return;
+                                }
+                            }
+                        }
+                        _ => { warn!(%addr, "signal: expected AuthToken"); return; }
+                    }
+                } else {
+                    None
+                };
+
+                // Wait for RegisterPresence
+                let (client_fp, client_alias) = match tokio::time::timeout(
+                    std::time::Duration::from_secs(10),
+                    transport.recv_signal(),
+                ).await {
+                    Ok(Ok(Some(SignalMessage::RegisterPresence { identity_pub, signature: _, alias }))) => {
+                        // Compute fingerprint: SHA-256(Ed25519 pub key)[:16], same as Fingerprint type
+                        let fp = {
+                            use sha2::{Sha256, Digest};
+                            let hash = Sha256::digest(&identity_pub);
+                            let fingerprint = wzp_crypto::Fingerprint([
+                                hash[0], hash[1], hash[2], hash[3], hash[4], hash[5], hash[6], hash[7],
+                                hash[8], hash[9], hash[10], hash[11], hash[12], hash[13], hash[14], hash[15],
+                            ]);
+                            fingerprint.to_string()
+                        };
+                        let fp = auth_fp.unwrap_or(fp);
+                        (fp, alias)
+                    }
+                    _ => {
+                        warn!(%addr, "signal: no RegisterPresence received");
+                        return;
+                    }
+                };
+
+                // Register in signal hub + presence
+                {
+                    let mut hub = signal_hub.lock().await;
+                    hub.register(client_fp.clone(), transport.clone(), client_alias.clone());
+                }
+                {
+                    let mut reg = presence.lock().await;
+                    reg.register_local(&client_fp, client_alias.clone(), None);
+                }
+
+                // Send ack
+                let _ = transport.send_signal(&SignalMessage::RegisterPresenceAck {
+                    success: true,
+                    error: None,
+                }).await;
+
+                info!(%addr, fingerprint = %client_fp, alias = ?client_alias, "signal client registered");
+
+                // Signal recv loop
+                loop {
+                    match transport.recv_signal().await {
+                        Ok(Some(msg)) => {
+                            match msg {
+                                SignalMessage::DirectCallOffer { ref target_fingerprint, ref call_id, ref caller_alias, .. } => {
+                                    let target_fp = target_fingerprint.clone();
+                                    let call_id = call_id.clone();
+
+                                    // Check if target is online
+                                    let online = {
+                                        let hub = signal_hub.lock().await;
+                                        hub.is_online(&target_fp)
+                                    };
+                                    if !online {
+                                        info!(%addr, target = %target_fp, "call target not online");
+                                        let _ = transport.send_signal(&SignalMessage::Hangup {
+                                            reason: wzp_proto::HangupReason::Normal,
+                                        }).await;
+                                        continue;
+                                    }
+
+                                    // Create call in registry
+                                    {
+                                        let mut reg = call_registry.lock().await;
+                                        reg.create_call(call_id.clone(), client_fp.clone(), target_fp.clone());
+                                    }
+
+                                    // Forward offer to callee
+                                    info!(caller = %client_fp, callee = %target_fp, call_id = %call_id, "routing direct call offer");
+                                    let hub = signal_hub.lock().await;
+                                    if let Err(e) = hub.send_to(&target_fp, &msg).await {
+                                        warn!("failed to forward call offer: {e}");
+                                    }
+
+                                    // Send ringing to caller
+                                    drop(hub);
+                                    let _ = transport.send_signal(&SignalMessage::CallRinging {
+                                        call_id: call_id.clone(),
+                                    }).await;
+                                }
+
+                                SignalMessage::DirectCallAnswer { ref call_id, ref accept_mode, .. } => {
+                                    let call_id = call_id.clone();
+                                    let mode = *accept_mode;
+
+                                    let peer_fp = {
+                                        let reg = call_registry.lock().await;
+                                        reg.peer_fingerprint(&call_id, &client_fp).map(|s| s.to_string())
+                                    };
+
+                                    let Some(peer_fp) = peer_fp else {
+                                        warn!(call_id = %call_id, "answer for unknown call");
+                                        continue;
+                                    };
+
+                                    if mode == wzp_proto::CallAcceptMode::Reject {
+                                        info!(call_id = %call_id, "call rejected");
+                                        let mut reg = call_registry.lock().await;
+                                        reg.end_call(&call_id);
+                                        drop(reg);
+                                        let hub = signal_hub.lock().await;
+                                        let _ = hub.send_to(&peer_fp, &SignalMessage::Hangup {
+                                            reason: wzp_proto::HangupReason::Normal,
+                                        }).await;
+                                    } else {
+                                        // Accept — create private room
+                                        let room = format!("call-{call_id}");
+                                        {
+                                            let mut reg = call_registry.lock().await;
+                                            reg.set_active(&call_id, mode, room.clone());
+                                        }
+                                        info!(call_id = %call_id, room = %room, mode = ?mode, "call accepted, creating room");
+
+                                        // Forward answer to caller
+                                        {
+                                            let hub = signal_hub.lock().await;
+                                            let _ = hub.send_to(&peer_fp, &msg).await;
+                                        }
+
+                                        // Send CallSetup to both parties
+                                        // Use the address the client connected to (their remote addr
+                                        // is our perspective, but we need our listen addr).
+                                        // Replace 0.0.0.0 with the client's destination IP.
+                                        let relay_addr_for_setup = if listen_addr_str.starts_with("0.0.0.0:") {
+                                            let port = &listen_addr_str[8..];
+                                            // Use the local IP from the client's connection
+                                            let local_ip = addr.ip();
+                                            if local_ip.is_loopback() {
+                                                format!("127.0.0.1:{port}")
+                                            } else {
+                                                format!("{local_ip}:{port}")
+                                            }
+                                        } else {
+                                            listen_addr_str.clone()
+                                        };
+                                        let setup = SignalMessage::CallSetup {
+                                            call_id: call_id.clone(),
+                                            room: room.clone(),
+                                            relay_addr: relay_addr_for_setup,
+                                        };
+                                        {
+                                            let hub = signal_hub.lock().await;
+                                            let _ = hub.send_to(&peer_fp, &setup).await;
+                                            let _ = hub.send_to(&client_fp, &setup).await;
+                                        }
+                                    }
+                                }
+
+                                SignalMessage::Hangup { .. } => {
+                                    // Forward hangup to all active calls for this user
+                                    let calls = {
+                                        let reg = call_registry.lock().await;
+                                        reg.calls_for_fingerprint(&client_fp)
+                                            .iter()
+                                            .map(|c| (c.call_id.clone(), if c.caller_fingerprint == client_fp {
+                                                c.callee_fingerprint.clone()
+                                            } else {
+                                                c.caller_fingerprint.clone()
+                                            }))
+                                            .collect::<Vec<_>>()
+                                    };
+                                    for (call_id, peer_fp) in &calls {
+                                        let hub = signal_hub.lock().await;
+                                        let _ = hub.send_to(peer_fp, &msg).await;
+                                        drop(hub);
+                                        let mut reg = call_registry.lock().await;
+                                        reg.end_call(call_id);
+                                    }
+                                }
+
+                                SignalMessage::Ping { timestamp_ms } => {
+                                    let _ = transport.send_signal(&SignalMessage::Pong { timestamp_ms }).await;
+                                }
+
+                                other => {
+                                    warn!(%addr, "signal: unexpected message: {:?}", std::mem::discriminant(&other));
+                                }
+                            }
+                        }
+                        Ok(None) => {
+                            info!(%addr, "signal connection closed");
+                            break;
+                        }
+                        Err(e) => {
+                            warn!(%addr, "signal recv error: {e}");
+                            break;
+                        }
+                    }
+                }
+
+                // Cleanup: unregister + end active calls
+                let active_calls = {
+                    let reg = call_registry.lock().await;
+                    reg.calls_for_fingerprint(&client_fp)
+                        .iter()
+                        .map(|c| (c.call_id.clone(), if c.caller_fingerprint == client_fp {
+                            c.callee_fingerprint.clone()
+                        } else {
+                            c.caller_fingerprint.clone()
+                        }))
+                        .collect::<Vec<_>>()
+                };
+                for (call_id, peer_fp) in &active_calls {
+                    let hub = signal_hub.lock().await;
+                    let _ = hub.send_to(peer_fp, &SignalMessage::Hangup {
+                        reason: wzp_proto::HangupReason::Normal,
+                    }).await;
+                    drop(hub);
+                    let mut reg = call_registry.lock().await;
+                    reg.end_call(call_id);
+                }
+
+                {
+                    let mut hub = signal_hub.lock().await;
+                    hub.unregister(&client_fp);
+                }
+                {
+                    let mut reg = presence.lock().await;
+                    reg.unregister_local(&client_fp);
+                }
+
+                transport.close().await.ok();
+                return;
+            }
+
             // Auth check: if --auth-url is set, expect first signal message to be a token
             // Auth: if --auth-url is set, expect AuthToken as first signal
             let authenticated_fp: Option<String> = if let Some(ref url) = auth_url {
@@ -489,6 +965,28 @@ async fn main() -> anyhow::Result<()> {
             // Use the caller's identity fingerprint from the handshake
             let participant_fp = authenticated_fp.clone().unwrap_or(caller_fp);
 
+            // ACL: call rooms (call-*) are restricted to the two authorized participants.
+            // Only the relay's call orchestrator creates these rooms — random clients can't join.
+            if room_name.starts_with("call-") {
+                let call_id = &room_name[5..]; // strip "call-" prefix
+                let authorized = {
+                    let reg = call_registry.lock().await;
+                    match reg.get(call_id) {
+                        Some(call) => {
+                            call.caller_fingerprint == participant_fp
+                                || call.callee_fingerprint == participant_fp
+                        }
+                        None => false, // unknown call — reject
+                    }
+                };
+                if !authorized {
+                    warn!(%addr, room = %room_name, fp = %participant_fp, "rejected: not authorized for this call room");
+                    transport.close().await.ok();
+                    return;
+                }
+                info!(%addr, room = %room_name, fp = %participant_fp, "authorized for call room");
+            }
+
             // Register in presence registry
             {
                 let mut reg = presence.lock().await;
@@ -541,6 +1039,20 @@ async fn main() -> anyhow::Result<()> {
 
                 metrics.active_sessions.inc();
 
+                // Call rooms: enforce 2-participant limit
+                if room_name.starts_with("call-") {
+                    let mgr = room_mgr.lock().await;
+                    if mgr.room_size(&room_name) >= 2 {
+                        drop(mgr);
+                        warn!(%addr, room = %room_name, "call room full (max 2 participants)");
+                        metrics.active_sessions.dec();
+                        let mut smgr = session_mgr.lock().await;
+                        smgr.remove_session(session_id);
+                        transport.close().await.ok();
+                        return;
+                    }
+                }
+
                 let participant_id = {
                     let mut mgr = room_mgr.lock().await;
                     match mgr.join(
@@ -553,7 +1065,25 @@ async fn main() -> anyhow::Result<()> {
                         Ok((id, update, senders)) => {
                             metrics.active_rooms.set(mgr.list().len() as i64);
                             drop(mgr); // release lock before async broadcast
-                            room::broadcast_signal(&senders, &update).await;
+
+                            // Merge federated participants into RoomUpdate if this is a global room
+                            let merged_update = if let Some(ref fm) = federation_mgr {
+                                if fm.is_global_room(&room_name) {
+                                    if let SignalMessage::RoomUpdate { count: _, participants: mut local_parts } = update {
+                                        let remote = fm.get_remote_participants(&room_name).await;
+                                        local_parts.extend(remote);
+                                        // Deduplicate by fingerprint
+                                        let mut seen = std::collections::HashSet::new();
+                                        local_parts.retain(|p| seen.insert(p.fingerprint.clone()));
+                                        SignalMessage::RoomUpdate {
+                                            count: local_parts.len() as u32,
+                                            participants: local_parts,
+                                        }
+                                    } else { update }
+                                } else { update }
+                            } else { update };
+
+                            room::broadcast_signal(&senders, &merged_update).await;
                             id
                         }
                         Err(e) => {
@@ -571,6 +1101,25 @@ async fn main() -> anyhow::Result<()> {
                     .iter()
                     .map(|b| format!("{b:02x}"))
                     .collect();
+                // Set up federation media channel if this is a global room
+                let (federation_tx, federation_room_hash) = if let Some(ref fm) = federation_mgr {
+                    let is_global = fm.is_global_room(&room_name);
+                    if is_global {
+                        let canonical_hash = fm.global_room_hash(&room_name);
+                        let (tx, rx) = tokio::sync::mpsc::channel(256);
+                        let fm_clone = fm.clone();
+                        tokio::spawn(async move {
+                            wzp_relay::federation::run_federation_media_egress(fm_clone, rx).await;
+                        });
+                        info!(room = %room_name, canonical = ?fm.resolve_global_room(&room_name), "federation egress created (global room)");
+                        (Some(tx), Some(canonical_hash))
+                    } else {
+                        (None, None)
+                    }
+                } else {
+                    (None, None)
+                };
+
                 room::run_participant(
                     room_mgr.clone(),
                     room_name,
@@ -579,6 +1128,9 @@ async fn main() -> anyhow::Result<()> {
                     metrics.clone(),
                     &session_id_str,
                     trunking_enabled,
+                    debug_tap,
+                    federation_tx,
+                    federation_room_hash,
                 ).await;
 
                 // Participant disconnected — clean up presence + per-session metrics

crates/wzp-relay/src/metrics.rs

@@ -16,6 +16,13 @@ pub struct RelayMetrics {
     pub bytes_forwarded: IntCounter,
     pub auth_attempts: IntCounterVec,
     pub handshake_duration: Histogram,
+    // Federation metrics
+    pub federation_peer_status: IntGaugeVec,
+    pub federation_peer_rtt_ms: GaugeVec,
+    pub federation_packets_forwarded: IntCounterVec,
+    pub federation_packets_deduped: IntCounter,
+    pub federation_packets_rate_limited: IntCounter,
+    pub federation_active_rooms: IntGauge,
     // Per-session metrics
     pub session_buffer_depth: IntGaugeVec,
     pub session_loss_pct: GaugeVec,
@@ -60,6 +67,28 @@ impl RelayMetrics {
         )
         .expect("metric");
 
+        let federation_peer_status = IntGaugeVec::new(
+            Opts::new("wzp_federation_peer_status", "Peer connection status (0=disconnected, 1=connected)"),
+            &["peer"],
+        ).expect("metric");
+        let federation_peer_rtt_ms = GaugeVec::new(
+            Opts::new("wzp_federation_peer_rtt_ms", "QUIC RTT to federated peer in milliseconds"),
+            &["peer"],
+        ).expect("metric");
+        let federation_packets_forwarded = IntCounterVec::new(
+            Opts::new("wzp_federation_packets_forwarded_total", "Packets forwarded to/from federated peers"),
+            &["peer", "direction"],
+        ).expect("metric");
+        let federation_packets_deduped = IntCounter::with_opts(
+            Opts::new("wzp_federation_packets_deduped_total", "Duplicate federation packets dropped"),
+        ).expect("metric");
+        let federation_packets_rate_limited = IntCounter::with_opts(
+            Opts::new("wzp_federation_packets_rate_limited_total", "Federation packets dropped by rate limiter"),
+        ).expect("metric");
+        let federation_active_rooms = IntGauge::with_opts(
+            Opts::new("wzp_federation_active_rooms", "Number of federated rooms currently active"),
+        ).expect("metric");
+
         let session_buffer_depth = IntGaugeVec::new(
             Opts::new(
                 "wzp_relay_session_jitter_buffer_depth",
@@ -107,6 +136,12 @@ impl RelayMetrics {
         registry.register(Box::new(bytes_forwarded.clone())).expect("register");
         registry.register(Box::new(auth_attempts.clone())).expect("register");
         registry.register(Box::new(handshake_duration.clone())).expect("register");
+        registry.register(Box::new(federation_peer_status.clone())).expect("register");
+        registry.register(Box::new(federation_peer_rtt_ms.clone())).expect("register");
+        registry.register(Box::new(federation_packets_forwarded.clone())).expect("register");
+        registry.register(Box::new(federation_packets_deduped.clone())).expect("register");
+        registry.register(Box::new(federation_packets_rate_limited.clone())).expect("register");
+        registry.register(Box::new(federation_active_rooms.clone())).expect("register");
         registry.register(Box::new(session_buffer_depth.clone())).expect("register");
         registry.register(Box::new(session_loss_pct.clone())).expect("register");
         registry.register(Box::new(session_rtt_ms.clone())).expect("register");
@@ -120,6 +155,12 @@ impl RelayMetrics {
             bytes_forwarded,
             auth_attempts,
             handshake_duration,
+            federation_peer_status,
+            federation_peer_rtt_ms,
+            federation_packets_forwarded,
+            federation_packets_deduped,
+            federation_packets_rate_limited,
+            federation_active_rooms,
             session_buffer_depth,
             session_loss_pct,
             session_rtt_ms,

crates/wzp-relay/src/room.rs

@@ -18,6 +18,38 @@ use wzp_proto::MediaTransport;
 use crate::metrics::RelayMetrics;
 use crate::trunk::TrunkBatcher;
 
+/// Debug tap: logs packet metadata for matching rooms.
+#[derive(Clone)]
+pub struct DebugTap {
+    /// Room name filter ("*" = all rooms, or specific room name/hash).
+    pub room_filter: String,
+}
+
+impl DebugTap {
+    pub fn matches(&self, room_name: &str) -> bool {
+        self.room_filter == "*" || self.room_filter == room_name
+    }
+
+    pub fn log_packet(&self, room: &str, dir: &str, addr: &std::net::SocketAddr, pkt: &wzp_proto::MediaPacket, fan_out: usize) {
+        let h = &pkt.header;
+        info!(
+            target: "debug_tap",
+            room = %room,
+            dir = dir,
+            addr = %addr,
+            seq = h.seq,
+            codec = ?h.codec_id,
+            ts = h.timestamp,
+            fec_block = h.fec_block,
+            fec_sym = h.fec_symbol,
+            repair = h.is_repair,
+            len = pkt.payload.len(),
+            fan_out,
+            "TAP"
+        );
+    }
+}
+
 /// Unique participant ID within a room.
 pub type ParticipantId = u64;
 
@@ -27,6 +59,22 @@ fn next_id() -> ParticipantId {
     NEXT_PARTICIPANT_ID.fetch_add(1, Ordering::Relaxed)
 }
 
+/// Events emitted by RoomManager for federation to observe.
+#[derive(Clone, Debug)]
+pub enum RoomEvent {
+    /// First local participant joined this room.
+    LocalJoin { room: String },
+    /// Last local participant left this room.
+    LocalLeave { room: String },
+}
+
+/// Outbound federation media from a local participant.
+pub struct FederationMediaOut {
+    pub room_name: String,
+    pub room_hash: [u8; 8],
+    pub data: Bytes,
+}
+
 /// How to send data to a participant — either via QUIC transport or WebSocket channel.
 #[derive(Clone)]
 pub enum ParticipantSender {
@@ -132,6 +180,7 @@ impl Room {
             .map(|p| wzp_proto::packet::RoomParticipant {
                 fingerprint: p.fingerprint.clone().unwrap_or_default(),
                 alias: p.alias.clone(),
+                relay_label: None, // local participant
             })
             .collect()
     }
@@ -141,17 +190,6 @@ impl Room {
         self.participants.iter().map(|p| p.sender.clone()).collect()
     }
 
-    /// Update a participant's alias. Returns true if the participant was found.
-    fn set_alias(&mut self, id: ParticipantId, alias: String) -> bool {
-        if let Some(p) = self.participants.iter_mut().find(|p| p.id == id) {
-            info!(participant = id, %alias, "alias updated");
-            p.alias = Some(alias);
-            true
-        } else {
-            false
-        }
-    }
-
     fn is_empty(&self) -> bool {
         self.participants.is_empty()
     }
@@ -168,24 +206,35 @@ pub struct RoomManager {
     /// When `None`, rooms are open (no auth mode). When `Some`, only listed
     /// fingerprints can join the corresponding room.
     acl: Option<HashMap<String, HashSet<String>>>,
+    /// Channel for room lifecycle events (federation subscribes).
+    event_tx: tokio::sync::broadcast::Sender<RoomEvent>,
 }
 
 impl RoomManager {
     pub fn new() -> Self {
+        let (event_tx, _) = tokio::sync::broadcast::channel(64);
         Self {
             rooms: HashMap::new(),
             acl: None,
+            event_tx,
         }
     }
 
     /// Create a room manager with ACL enforcement enabled.
     pub fn with_acl() -> Self {
+        let (event_tx, _) = tokio::sync::broadcast::channel(64);
         Self {
             rooms: HashMap::new(),
             acl: Some(HashMap::new()),
+            event_tx,
         }
     }
 
+    /// Subscribe to room lifecycle events (for federation).
+    pub fn subscribe_events(&self) -> tokio::sync::broadcast::Receiver<RoomEvent> {
+        self.event_tx.subscribe()
+    }
+
     /// Grant a fingerprint access to a room.
     pub fn allow(&mut self, room_name: &str, fingerprint: &str) {
         if let Some(ref mut acl) = self.acl {
@@ -224,8 +273,13 @@ impl RoomManager {
             warn!(room = room_name, fingerprint = ?fingerprint, "unauthorized room join attempt");
             return Err("not authorized for this room".to_string());
         }
+        let was_empty = !self.rooms.contains_key(room_name)
+            || self.rooms.get(room_name).map_or(true, |r| r.is_empty());
         let room = self.rooms.entry(room_name.to_string()).or_insert_with(Room::new);
         let id = room.add(addr, sender, fingerprint.map(|s| s.to_string()), alias.map(|s| s.to_string()));
+        if was_empty {
+            let _ = self.event_tx.send(RoomEvent::LocalJoin { room: room_name.to_string() });
+        }
         let update = wzp_proto::SignalMessage::RoomUpdate {
             count: room.len() as u32,
             participants: room.participant_list(),
@@ -246,12 +300,34 @@ impl RoomManager {
         Ok(id)
     }
 
+    /// Get list of active room names.
+    pub fn active_rooms(&self) -> Vec<String> {
+        self.rooms.keys().cloned().collect()
+    }
+
+    /// Get participant list for a room (fingerprint + alias).
+    pub fn local_participant_list(&self, room_name: &str) -> Vec<wzp_proto::packet::RoomParticipant> {
+        self.rooms.get(room_name)
+            .map(|room| room.participant_list())
+            .unwrap_or_default()
+    }
+
+    /// Get all senders for participants in a room (for federation inbound media delivery).
+    pub fn local_senders(&self, room_name: &str) -> Vec<ParticipantSender> {
+        self.rooms.get(room_name)
+            .map(|room| room.participants.iter()
+                .map(|p| p.sender.clone())
+                .collect())
+            .unwrap_or_default()
+    }
+
     /// Leave a room. Returns (room_update_msg, remaining_senders) for broadcasting, or None if room is now empty.
     pub fn leave(&mut self, room_name: &str, participant_id: ParticipantId) -> Option<(wzp_proto::SignalMessage, Vec<ParticipantSender>)> {
         if let Some(room) = self.rooms.get_mut(room_name) {
             room.remove(participant_id);
             if room.is_empty() {
                 self.rooms.remove(room_name);
+                let _ = self.event_tx.send(RoomEvent::LocalLeave { room: room_name.to_string() });
                 info!(room = room_name, "room closed (empty)");
                 return None;
             }
@@ -266,26 +342,6 @@ impl RoomManager {
         }
     }
 
-    /// Update a participant's alias and return a RoomUpdate + senders for broadcasting.
-    pub fn set_alias(
-        &mut self,
-        room_name: &str,
-        participant_id: ParticipantId,
-        alias: String,
-    ) -> Option<(wzp_proto::SignalMessage, Vec<ParticipantSender>)> {
-        if let Some(room) = self.rooms.get_mut(room_name) {
-            if room.set_alias(participant_id, alias) {
-                let update = wzp_proto::SignalMessage::RoomUpdate {
-                    count: room.len() as u32,
-                    participants: room.participant_list(),
-                };
-                let senders = room.all_senders();
-                return Some((update, senders));
-            }
-        }
-        None
-    }
-
     /// Get senders for all OTHER participants in a room.
     pub fn others(
         &self,
@@ -381,6 +437,9 @@ pub async fn run_participant(
     metrics: Arc<RelayMetrics>,
     session_id: &str,
     trunking_enabled: bool,
+    debug_tap: Option<DebugTap>,
+    federation_tx: Option<tokio::sync::mpsc::Sender<FederationMediaOut>>,
+    federation_room_hash: Option<[u8; 8]>,
 ) {
     if trunking_enabled {
         run_participant_trunked(
@@ -389,7 +448,7 @@ pub async fn run_participant(
         .await;
     } else {
         run_participant_plain(
-            room_mgr, room_name, participant_id, transport, metrics, session_id,
+            room_mgr, room_name, participant_id, transport, metrics, session_id, debug_tap, federation_tx, federation_room_hash,
         )
         .await;
     }
@@ -403,168 +462,164 @@ async fn run_participant_plain(
     transport: Arc<wzp_transport::QuinnTransport>,
     metrics: Arc<RelayMetrics>,
     session_id: &str,
+    debug_tap: Option<DebugTap>,
+    federation_tx: Option<tokio::sync::mpsc::Sender<FederationMediaOut>>,
+    federation_room_hash: Option<[u8; 8]>,
 ) {
     let addr = transport.connection().remote_address();
+    let mut packets_forwarded = 0u64;
+    let mut last_recv_instant = std::time::Instant::now();
+    let mut max_recv_gap_ms = 0u64;
+    let mut max_forward_ms = 0u64;
+    let mut send_errors = 0u64;
+    let mut last_log_instant = std::time::Instant::now();
 
-    // Media forwarding task (with debug logging from Android fixes)
-    let media_room_mgr = room_mgr.clone();
-    let media_room_name = room_name.clone();
-    let media_transport = transport.clone();
-    let media_metrics = metrics.clone();
-    let media_session_id = session_id.to_string();
-    let media_task = async move {
-        let mut packets_forwarded = 0u64;
-        let mut last_recv_instant = std::time::Instant::now();
-        let mut max_recv_gap_ms = 0u64;
-        let mut max_forward_ms = 0u64;
-        let mut send_errors = 0u64;
-        let mut last_log_instant = std::time::Instant::now();
+    info!(
+        room = %room_name,
+        participant = participant_id,
+        %addr,
+        session = session_id,
+        "forwarding loop started (plain)"
+    );
 
-        info!(
-            room = %media_room_name,
-            participant = participant_id,
-            %addr,
-            session = %media_session_id,
-            "forwarding loop started (plain)"
-        );
-
-        loop {
-            let pkt = match media_transport.recv_media().await {
-                Ok(Some(pkt)) => pkt,
-                Ok(None) => {
-                    info!(%addr, participant = participant_id, forwarded = packets_forwarded, "disconnected (stream ended)");
-                    break;
+    loop {
+        let recv_start = std::time::Instant::now();
+        let pkt = match transport.recv_media().await {
+            Ok(Some(pkt)) => pkt,
+            Ok(None) => {
+                info!(%addr, participant = participant_id, forwarded = packets_forwarded, "disconnected (stream ended)");
+                break;
+            }
+            Err(e) => {
+                let msg = e.to_string();
+                if msg.contains("timed out") || msg.contains("reset") || msg.contains("closed") {
+                    info!(%addr, participant = participant_id, forwarded = packets_forwarded, "connection closed: {e}");
+                } else {
+                    error!(%addr, participant = participant_id, forwarded = packets_forwarded, "recv error: {e}");
                 }
-                Err(e) => {
-                    let msg = e.to_string();
-                    if msg.contains("timed out") || msg.contains("reset") || msg.contains("closed") {
-                        info!(%addr, participant = participant_id, forwarded = packets_forwarded, "connection closed: {e}");
-                    } else {
-                        error!(%addr, participant = participant_id, forwarded = packets_forwarded, "recv error: {e}");
-                    }
-                    break;
-                }
-            };
-
-            let recv_gap_ms = last_recv_instant.elapsed().as_millis() as u64;
-            last_recv_instant = std::time::Instant::now();
-            if recv_gap_ms > max_recv_gap_ms {
-                max_recv_gap_ms = recv_gap_ms;
-            }
-            if recv_gap_ms > 200 {
-                warn!(
-                    room = %media_room_name,
-                    participant = participant_id,
-                    recv_gap_ms,
-                    seq = pkt.header.seq,
-                    "large recv gap"
-                );
+                break;
             }
+        };
 
-            if let Some(ref report) = pkt.quality_report {
-                media_metrics.update_session_quality(&media_session_id, report);
-            }
+        let recv_gap_ms = last_recv_instant.elapsed().as_millis() as u64;
+        last_recv_instant = std::time::Instant::now();
+        if recv_gap_ms > max_recv_gap_ms {
+            max_recv_gap_ms = recv_gap_ms;
+        }
+        // Log if recv gap is suspiciously large (>200ms = missed ~10 packets)
+        if recv_gap_ms > 200 {
+            warn!(
+                room = %room_name,
+                participant = participant_id,
+                recv_gap_ms,
+                seq = pkt.header.seq,
+                "large recv gap"
+            );
+        }
 
-            let lock_start = std::time::Instant::now();
-            let others = {
-                let mgr = media_room_mgr.lock().await;
-                mgr.others(&media_room_name, participant_id)
-            };
-            let lock_ms = lock_start.elapsed().as_millis() as u64;
-            if lock_ms > 10 {
-                warn!(room = %media_room_name, participant = participant_id, lock_ms, "slow room_mgr lock");
-            }
+        // Update per-session quality metrics if a quality report is present
+        if let Some(ref report) = pkt.quality_report {
+            metrics.update_session_quality(session_id, report);
+        }
 
-            let fwd_start = std::time::Instant::now();
-            let pkt_bytes = pkt.payload.len() as u64;
-            for other in &others {
-                match other {
-                    ParticipantSender::Quic(t) => {
-                        if let Err(e) = t.send_media(&pkt).await {
-                            send_errors += 1;
-                            if send_errors <= 5 || send_errors % 100 == 0 {
-                                warn!(
-                                    room = %media_room_name,
-                                    participant = participant_id,
-                                    peer = %t.connection().remote_address(),
-                                    total_send_errors = send_errors,
-                                    "send_media error: {e}"
-                                );
-                            }
+        // Get current list of other participants
+        let lock_start = std::time::Instant::now();
+        let others = {
+            let mgr = room_mgr.lock().await;
+            mgr.others(&room_name, participant_id)
+        };
+        let lock_ms = lock_start.elapsed().as_millis() as u64;
+        if lock_ms > 10 {
+            warn!(
+                room = %room_name,
+                participant = participant_id,
+                lock_ms,
+                "slow room_mgr lock"
+            );
+        }
+
+        // Debug tap: log packet metadata
+        if let Some(ref tap) = debug_tap {
+            if tap.matches(&room_name) {
+                tap.log_packet(&room_name, "in", &addr, &pkt, others.len());
+            }
+        }
+
+        // Forward to all others
+        let fwd_start = std::time::Instant::now();
+        let pkt_bytes = pkt.payload.len() as u64;
+        for other in &others {
+            match other {
+                ParticipantSender::Quic(t) => {
+                    if let Err(e) = t.send_media(&pkt).await {
+                        send_errors += 1;
+                        if send_errors <= 5 || send_errors % 100 == 0 {
+                            warn!(
+                                room = %room_name,
+                                participant = participant_id,
+                                peer = %t.connection().remote_address(),
+                                total_send_errors = send_errors,
+                                "send_media error: {e}"
+                            );
                         }
                     }
-                    ParticipantSender::WebSocket(_) => {
-                        let _ = other.send_raw(&pkt.payload).await;
-                    }
                 }
-            }
-            let fwd_ms = fwd_start.elapsed().as_millis() as u64;
-            if fwd_ms > max_forward_ms { max_forward_ms = fwd_ms; }
-            if fwd_ms > 50 {
-                warn!(room = %media_room_name, participant = participant_id, fwd_ms, fan_out = others.len(), "slow forward");
-            }
-
-            let fan_out = others.len() as u64;
-            media_metrics.packets_forwarded.inc_by(fan_out);
-            media_metrics.bytes_forwarded.inc_by(pkt_bytes * fan_out);
-            packets_forwarded += 1;
-
-            if last_log_instant.elapsed() >= Duration::from_secs(5) {
-                let room_size = {
-                    let mgr = media_room_mgr.lock().await;
-                    mgr.room_size(&media_room_name)
-                };
-                info!(
-                    room = %media_room_name,
-                    participant = participant_id,
-                    forwarded = packets_forwarded,
-                    room_size, fan_out, max_recv_gap_ms, max_forward_ms, send_errors,
-                    "participant stats"
-                );
-                max_recv_gap_ms = 0;
-                max_forward_ms = 0;
-                last_log_instant = std::time::Instant::now();
-            }
-        }
-    };
-
-    // Signal handling task — processes SetAlias and other in-call signals
-    let signal_room_mgr = room_mgr.clone();
-    let signal_room_name = room_name.clone();
-    let signal_transport = transport.clone();
-    let signal_task = async move {
-        loop {
-            match signal_transport.recv_signal().await {
-                Ok(Some(wzp_proto::SignalMessage::SetAlias { alias })) => {
-                    info!(%addr, participant = participant_id, %alias, "SetAlias received");
-                    let mut mgr = signal_room_mgr.lock().await;
-                    if let Some((update, senders)) =
-                        mgr.set_alias(&signal_room_name, participant_id, alias)
-                    {
-                        drop(mgr);
-                        broadcast_signal(&senders, &update).await;
-                    }
-                }
-                Ok(Some(wzp_proto::SignalMessage::Hangup { .. })) => {
-                    info!(%addr, participant = participant_id, "hangup received");
-                    break;
-                }
-                Ok(Some(msg)) => {
-                    info!(%addr, participant = participant_id, "signal: {:?}", std::mem::discriminant(&msg));
-                }
-                Ok(None) => break,
-                Err(e) => {
-                    warn!(%addr, participant = participant_id, "signal recv error: {e}");
-                    break;
+                ParticipantSender::WebSocket(_) => {
+                    let _ = other.send_raw(&pkt.payload).await;
                 }
             }
         }
-    };
 
-    // Run both in parallel — exit when either finishes (disconnection)
-    tokio::select! {
-        _ = media_task => {}
-        _ = signal_task => {}
+        // Federation: forward to active peer relays via channel
+        if let Some(ref fed_tx) = federation_tx {
+            let data = pkt.to_bytes();
+            let _ = fed_tx.try_send(FederationMediaOut {
+                room_name: room_name.clone(),
+                room_hash: federation_room_hash.unwrap_or_else(|| crate::federation::room_hash(&room_name)),
+                data,
+            });
+        }
+
+        let fwd_ms = fwd_start.elapsed().as_millis() as u64;
+        if fwd_ms > max_forward_ms {
+            max_forward_ms = fwd_ms;
+        }
+        if fwd_ms > 50 {
+            warn!(
+                room = %room_name,
+                participant = participant_id,
+                fwd_ms,
+                fan_out = others.len(),
+                "slow forward"
+            );
+        }
+
+        let fan_out = others.len() as u64;
+        metrics.packets_forwarded.inc_by(fan_out);
+        metrics.bytes_forwarded.inc_by(pkt_bytes * fan_out);
+        packets_forwarded += 1;
+
+        // Periodic stats log every 5 seconds
+        if last_log_instant.elapsed() >= Duration::from_secs(5) {
+            let room_size = {
+                let mgr = room_mgr.lock().await;
+                mgr.room_size(&room_name)
+            };
+            info!(
+                room = %room_name,
+                participant = participant_id,
+                forwarded = packets_forwarded,
+                room_size,
+                fan_out,
+                max_recv_gap_ms,
+                max_forward_ms,
+                send_errors,
+                "participant stats"
+            );
+            max_recv_gap_ms = 0;
+            max_forward_ms = 0;
+            last_log_instant = std::time::Instant::now();
+        }
     }
 
     // Clean up — leave room and broadcast update to remaining participants

crates/wzp-relay/src/signal_hub.rs

@@ -0,0 +1,105 @@
+//! Persistent signaling connection manager.
+//!
+//! Tracks clients connected via `_signal` SNI. Routes call signals
+//! (DirectCallOffer, DirectCallAnswer, Hangup) between registered users.
+
+use std::collections::HashMap;
+use std::sync::Arc;
+use std::time::Instant;
+
+use tracing::{info, warn};
+use wzp_proto::{MediaTransport, SignalMessage};
+use wzp_transport::QuinnTransport;
+
+/// A client connected via `_signal` for direct calling.
+pub struct SignalClient {
+    pub fingerprint: String,
+    pub alias: Option<String>,
+    pub transport: Arc<QuinnTransport>,
+    pub connected_at: Instant,
+}
+
+/// Manages persistent signaling connections.
+pub struct SignalHub {
+    clients: HashMap<String, SignalClient>,
+}
+
+impl SignalHub {
+    pub fn new() -> Self {
+        Self {
+            clients: HashMap::new(),
+        }
+    }
+
+    /// Register a new signaling client.
+    pub fn register(&mut self, fp: String, transport: Arc<QuinnTransport>, alias: Option<String>) {
+        info!(fingerprint = %fp, alias = ?alias, "signal client registered");
+        self.clients.insert(fp.clone(), SignalClient {
+            fingerprint: fp,
+            alias,
+            transport,
+            connected_at: Instant::now(),
+        });
+    }
+
+    /// Unregister a signaling client. Returns the client if found.
+    pub fn unregister(&mut self, fp: &str) -> Option<SignalClient> {
+        let client = self.clients.remove(fp);
+        if client.is_some() {
+            info!(fingerprint = %fp, "signal client unregistered");
+        }
+        client
+    }
+
+    /// Look up a client by fingerprint.
+    pub fn get(&self, fp: &str) -> Option<&SignalClient> {
+        self.clients.get(fp)
+    }
+
+    /// Check if a fingerprint is online.
+    pub fn is_online(&self, fp: &str) -> bool {
+        self.clients.contains_key(fp)
+    }
+
+    /// Send a signal message to a client by fingerprint.
+    pub async fn send_to(&self, fp: &str, msg: &SignalMessage) -> Result<(), String> {
+        match self.clients.get(fp) {
+            Some(client) => {
+                client.transport.send_signal(msg).await
+                    .map_err(|e| format!("send to {fp}: {e}"))
+            }
+            None => Err(format!("{fp} not online")),
+        }
+    }
+
+    /// Number of connected signaling clients.
+    pub fn online_count(&self) -> usize {
+        self.clients.len()
+    }
+
+    /// List all online fingerprints.
+    pub fn online_fingerprints(&self) -> Vec<&str> {
+        self.clients.keys().map(|s| s.as_str()).collect()
+    }
+
+    /// Get alias for a fingerprint.
+    pub fn alias(&self, fp: &str) -> Option<&str> {
+        self.clients.get(fp).and_then(|c| c.alias.as_deref())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn register_unregister() {
+        let mut hub = SignalHub::new();
+        assert_eq!(hub.online_count(), 0);
+        assert!(!hub.is_online("alice"));
+
+        // Can't easily construct QuinnTransport in a unit test,
+        // so we just test the HashMap logic conceptually.
+        // Integration tests cover the full flow.
+    }
+}

crates/wzp-transport/Cargo.toml

@@ -16,6 +16,9 @@ async-trait = { workspace = true }
 serde_json = "1"
 rustls = { version = "0.23", default-features = false, features = ["ring", "std"] }
 rcgen = "0.13"
+ed25519-dalek = { workspace = true }
+hkdf = { workspace = true }
+sha2 = { workspace = true }
 
 [dev-dependencies]
 tokio = { workspace = true, features = ["rt-multi-thread", "macros"] }

crates/wzp-transport/src/config.rs

@@ -6,20 +6,74 @@ use std::time::Duration;
 use quinn::crypto::rustls::QuicClientConfig;
 use quinn::crypto::rustls::QuicServerConfig;
 
-/// Create a server configuration with a self-signed certificate (for testing).
+/// Create a server configuration with a self-signed certificate (random keypair).
 ///
-/// Tunes QUIC transport parameters for lossy VoIP:
-/// - 30s idle timeout
-/// - 5s keep-alive interval
-/// - DATAGRAM extension enabled
-/// - Conservative flow control for bandwidth-constrained links
+/// The certificate changes on every call. Use `server_config_from_seed` for
+/// a deterministic certificate that survives relay restarts.
 pub fn server_config() -> (quinn::ServerConfig, Vec<u8>) {
     let cert_key = rcgen::generate_simple_self_signed(vec!["localhost".to_string()])
         .expect("failed to generate self-signed cert");
     let cert_der = rustls::pki_types::CertificateDer::from(cert_key.cert);
     let key_der =
         rustls::pki_types::PrivateKeyDer::try_from(cert_key.key_pair.serialize_der()).unwrap();
+    build_server_config(cert_der, key_der)
+}
 
+/// Create a server configuration with a deterministic self-signed certificate
+/// derived from a 32-byte seed. Same seed = same cert = same TLS fingerprint.
+pub fn server_config_from_seed(seed: &[u8; 32]) -> (quinn::ServerConfig, Vec<u8>) {
+    use ed25519_dalek::pkcs8::EncodePrivateKey;
+    use ed25519_dalek::SigningKey;
+    use hkdf::Hkdf;
+    use sha2::Sha256;
+
+    // Derive Ed25519 key bytes from seed via HKDF
+    let hk = Hkdf::<Sha256>::new(None, seed);
+    let mut ed_bytes = [0u8; 32];
+    hk.expand(b"wzp-tls-ed25519", &mut ed_bytes)
+        .expect("HKDF expand failed");
+
+    // Create Ed25519 signing key and export as PKCS8 DER
+    let signing_key = SigningKey::from_bytes(&ed_bytes);
+    let pkcs8_doc = signing_key.to_pkcs8_der()
+        .expect("failed to encode Ed25519 key as PKCS8");
+    let key_der_for_rcgen = rustls::pki_types::PrivateKeyDer::try_from(pkcs8_doc.as_bytes().to_vec())
+        .expect("failed to wrap PKCS8 DER");
+
+    // Create rcgen KeyPair from DER
+    let key_pair = rcgen::KeyPair::from_der_and_sign_algo(
+        &key_der_for_rcgen,
+        &rcgen::PKCS_ED25519,
+    )
+    .expect("failed to create KeyPair from seed-derived Ed25519 key");
+
+    // Build self-signed cert with this deterministic keypair
+    let params = rcgen::CertificateParams::new(vec!["localhost".to_string()])
+        .expect("failed to create CertificateParams");
+    let cert = params.self_signed(&key_pair).expect("failed to self-sign cert");
+    let cert_der = rustls::pki_types::CertificateDer::from(cert.der().to_vec());
+    let key_der = rustls::pki_types::PrivateKeyDer::try_from(key_pair.serialize_der())
+        .expect("failed to serialize key DER");
+
+    build_server_config(cert_der, key_der)
+}
+
+/// Compute a hex-formatted SHA-256 fingerprint of a DER-encoded certificate.
+///
+/// Format: `xx:xx:xx:xx:...` (32 bytes = 64 hex chars with colons).
+pub fn tls_fingerprint(cert_der: &[u8]) -> String {
+    use sha2::{Sha256, Digest};
+    let hash = Sha256::digest(cert_der);
+    hash.iter()
+        .map(|b| format!("{b:02x}"))
+        .collect::<Vec<_>>()
+        .join(":")
+}
+
+fn build_server_config(
+    cert_der: rustls::pki_types::CertificateDer<'static>,
+    key_der: rustls::pki_types::PrivateKeyDer<'static>,
+) -> (quinn::ServerConfig, Vec<u8>) {
     let mut server_crypto = rustls::ServerConfig::builder()
         .with_no_client_auth()
         .with_single_cert(vec![cert_der.clone()], key_der)

crates/wzp-transport/src/lib.rs

@@ -22,7 +22,7 @@ pub mod path_monitor;
 pub mod quic;
 pub mod reliable;
 
-pub use config::{client_config, server_config};
+pub use config::{client_config, server_config, server_config_from_seed, tls_fingerprint};
 pub use connection::{accept, connect, create_endpoint};
 pub use path_monitor::PathMonitor;
 pub use quic::QuinnTransport;

crates/wzp-transport/src/quic.rs

@@ -33,6 +33,13 @@ impl QuinnTransport {
         &self.connection
     }
 
+    /// Send raw bytes as a QUIC datagram (no MediaPacket framing).
+    pub fn send_raw_datagram(&self, data: &[u8]) -> Result<(), TransportError> {
+        self.connection
+            .send_datagram(bytes::Bytes::copy_from_slice(data))
+            .map_err(|e| TransportError::Internal(format!("datagram: {e}")))
+    }
+
     /// Close the QUIC connection immediately (synchronous, no async needed).
     /// The relay will detect the close and remove this participant from the room.
     pub fn close_now(&self) {
@@ -136,7 +143,7 @@ impl MediaTransport for QuinnTransport {
             }
         };
 
-        match datagram::deserialize_media(data) {
+        match datagram::deserialize_media(data.clone()) {
             Some(packet) => {
                 // Record receive observation
                 {
@@ -149,8 +156,10 @@ impl MediaTransport for QuinnTransport {
                 Ok(Some(packet))
             }
             None => {
-                tracing::warn!("received malformed media datagram");
-                Ok(None)
+                tracing::warn!(len = data.len(), "skipping malformed media datagram, continuing");
+                // Don't return Ok(None) — that signals connection closed.
+                // Recurse to read the next datagram instead.
+                Box::pin(self.recv_media()).await
             }
         }
     }