feat(host/wire): mid-session shard-payload renegotiation, driven by the MTU verdict

Phases 1-2 of design/shard-payload-reneg.md, on top of the Phase 0
per-frame geometry. The leg-1 watcher stops merely diagnosing the
constrained path and heals the CURRENT session; the same machinery,
inverted, takes a proven jumbo LAN up to ~8.9 KB shards.

- Messages: MSG_SHARD_PAYLOAD_CHANGED (0x08, host→client, {shard_payload
  u16}) and MSG_SHARD_PAYLOAD_ACK (0x09, the echo). Asymmetric by
  design: a shrink re-keys the packetizer at the next AU immediately
  after sending (per-frame pinning makes ordering irrelevant; the ack is
  telemetry), a grow emits nothing above the old size until the ack —
  the ack is the gate even though client buffers are statically sized.
- Client: one dispatch arm in the shared pump control task (all client
  families) — validate against the advertised receive bounds, ack;
  out-of-bounds requests get SILENCE, not an ack, so a buggy host can
  never read a granted grow out of garbage.
- Host driver: the wire_mtu watcher grows a ShardReneg arm — on a
  below-ceiling verdict it still records the learned budget (session 2
  starts right) and now also shrinks session 1 at the ~3-10 s verdict
  mark; with the jumbo opt-in (PUNKTFUNK_JUMBO=1, or PUNKTFUNK_WIRE_MTU
  > 1500 — one knob, derived) it sends the ack-gated grow after a
  settled-at-sealed-jumbo proof and then stays alive as the revert
  guard: quinn's blackhole detection lowering current_mtu shrinks the
  wire back through the same path. The QUIC MTUD probe ceiling rises
  from 1472 to the sealed jumbo size with the opt-in (per-ENDPOINT: a
  few extra failed probes toward non-jumbo peers, zero cost otherwise).
- Apply point: Session::set_shard_payload drained in the send loop next
  to the adaptive-FEC target, gated on no open streamed AU (a streamed
  frame's shard-aligned tiling derives from the size it began with).
- Renegotiation is gated OFF for PyroWave sessions: their clients parse
  chunk-aligned AUs in windows of the Welcome value pinned at session
  start (read once over the C ABI), so a mid-stream re-key would corrupt
  the parse — those sessions keep the leg-1 next-session clamp. This
  also settles the plan's open question on the two wire_chunk consumers:
  both are PyroWave-only, so the gate covers them entirely.
- Legacy peers are inert both ways: no Hello advertisement → the host
  never constructs the driver; an old host never sends the message.

core: 296/296 --features quic + clippy -D warnings (macOS), fmt; the
regenerated header carries the new message ids (drift gate).

crates/punktfunk-core/src/client/pump/control_task.rs

@@ -243,6 +243,38 @@ impl ControlTask {
                             seq: offer.seq,
                             kinds: offer.kinds,
                         });
+                    } else if let Ok(chg) = crate::quic::ShardPayloadChanged::decode(&msg) {
+                        // Mid-session shard renegotiation (design/shard-payload-reneg.md): the
+                        // host re-keys the sealed video geometry. Per-frame pinning means there
+                        // is nothing to re-key on the receive path — the reassembler follows
+                        // each frame's own header and every buffer is statically sized for the
+                        // ceiling — so the dispatch is validate + ack. The ack is telemetry for
+                        // a shrink and the GATE for a grow (the host emits nothing above the
+                        // old size until it lands). Validate against our own receive bounds —
+                        // the same ceiling we advertised in `Hello::max_shard_payload` — and
+                        // answer an out-of-bounds request with SILENCE, not an ack: a buggy
+                        // host must never read a granted grow out of garbage.
+                        let n = chg.shard_payload as usize;
+                        if (crate::config::MIN_SHARD_PAYLOAD..=crate::config::max_shard_payload())
+                            .contains(&n)
+                            && n % 2 == 0
+                        {
+                            tracing::info!(
+                                shard_payload = n,
+                                "host re-keyed the wire shard payload — acking"
+                            );
+                            let ack = crate::quic::ShardPayloadAck {
+                                shard_payload: chg.shard_payload,
+                            };
+                            if io::write_msg(&mut ctrl_send, &ack.encode()).await.is_err() {
+                                break;
+                            }
+                        } else {
+                            tracing::warn!(
+                                shard_payload = n,
+                                "out-of-bounds shard-payload change — ignoring (no ack)"
+                            );
+                        }
                     } else if let Ok(shape) = crate::quic::CursorShape::decode(&msg) {
                         // Pointer bitmap changed (cursor channel, only when negotiated). try_send:
                         // an overflowing ring drops the newest shape — the next change resends.

crates/punktfunk-core/src/config.rs

@@ -373,16 +373,54 @@ pub fn shard_payload_for_udp_budget(udp_budget: usize, peer: core::net::IpAddr)
     p.clamp(MIN_SHARD_PAYLOAD, mtu1500_shard_payload_for(peer))
 }
 
+/// The family's IP+UDP header bytes between an on-wire IP MTU and its UDP payload budget —
+/// 28 for IPv4 (and IPv4-mapped), 48 for IPv6.
+fn ip_udp_overhead(peer: core::net::IpAddr) -> usize {
+    match peer {
+        core::net::IpAddr::V4(_) => 28,
+        core::net::IpAddr::V6(v6) if v6.to_ipv4_mapped().is_some() => 28,
+        core::net::IpAddr::V6(_) => 48,
+    }
+}
+
 /// [`shard_payload_for_udp_budget`] for an operator-supplied ON-WIRE IP MTU (the number
 /// `netsh interface ipv4 show subinterfaces` / `ip link` shows): subtracts the family's IP+UDP
 /// headers first — 28 for IPv4 (and IPv4-mapped), 48 for IPv6.
 pub fn shard_payload_for_wire_mtu(wire_mtu: usize, peer: core::net::IpAddr) -> usize {
-    let ip_udp = match peer {
-        core::net::IpAddr::V4(_) => 28,
-        core::net::IpAddr::V6(v6) if v6.to_ipv4_mapped().is_some() => 28,
-        core::net::IpAddr::V6(_) => 48,
-    };
-    shard_payload_for_udp_budget(wire_mtu.saturating_sub(ip_udp), peer)
+    shard_payload_for_udp_budget(wire_mtu.saturating_sub(ip_udp_overhead(peer)), peer)
+}
+
+/// The operator's jumbo-frames opt-in (design/shard-payload-reneg.md Phase 2): the target
+/// on-wire IP MTU, or `None` = no opt-in (nothing above the 1500-default wire is ever probed
+/// or grown to). One knob, one code path: a `PUNKTFUNK_WIRE_MTU` above the standard 1500
+/// derives the target from the operator's number; `PUNKTFUNK_JUMBO=1` is the fixed 9000
+/// profile for operators who don't want to think in MTUs. Raising the wire above 1500 is
+/// only ever an ACK-GATED mid-session grow toward a client that advertised
+/// [`max_shard_payload`] headroom — sessions still START at the family default.
+pub fn jumbo_wire_mtu() -> Option<usize> {
+    if let Ok(v) = std::env::var("PUNKTFUNK_WIRE_MTU") {
+        if let Ok(mtu) = v.trim().parse::<usize>() {
+            if mtu > 1500 {
+                return Some(mtu);
+            }
+        }
+    }
+    match std::env::var("PUNKTFUNK_JUMBO") {
+        Ok(v) if v.trim() == "1" => Some(9000),
+        _ => None,
+    }
+}
+
+/// The jumbo sibling of [`shard_payload_for_wire_mtu`]: the largest even shard payload whose
+/// sealed datagram fits `wire_mtu`, clamped to the RECEIVE ceiling ([`max_shard_payload`])
+/// instead of the family 1500-default — the up-leg's grow target. Still floored at
+/// [`MIN_SHARD_PAYLOAD`].
+pub fn jumbo_shard_payload_for(wire_mtu: usize, peer: core::net::IpAddr) -> usize {
+    let p = wire_mtu
+        .saturating_sub(ip_udp_overhead(peer))
+        .saturating_sub(HEADER_LEN + CRYPTO_OVERHEAD);
+    let p = p - p % 2; // FEC requires even shards
+    p.clamp(MIN_SHARD_PAYLOAD, max_shard_payload())
 }
 
 /// Everything needed to construct a [`Session`](crate::session::Session).
@@ -626,6 +664,29 @@ mod tests {
         assert_eq!(shard_payload_for_wire_mtu(1280, v6), 1168);
     }
 
+    /// Jumbo grow-target sizing (the up-leg, design/shard-payload-reneg.md): even, sealed
+    /// fits the wire, clamped to the RECEIVE ceiling instead of the family 1500-default —
+    /// and the standard 9000 profile lands on the exact documented value.
+    #[test]
+    fn jumbo_shard_payload_math() {
+        use core::net::IpAddr;
+        let v4: IpAddr = "192.168.1.50".parse().unwrap();
+        let v6: IpAddr = "fd00::50".parse().unwrap();
+        // 9000 − 28 (IPv4+UDP) − 64 (header+crypto) = 8908 even; sealed 8972 ≤ the 9216
+        // datagram ceiling. The v6 sibling: 9000 − 48 − 64 = 8888.
+        assert_eq!(jumbo_shard_payload_for(9000, v4), 8908);
+        assert_eq!(sealed_datagram_bytes(8908), 8972);
+        assert!(sealed_datagram_bytes(8908) <= MAX_DATAGRAM_BYTES);
+        assert_eq!(jumbo_shard_payload_for(9000, v6), 8888);
+        // An operator MTU larger than the receive path clamps to the ceiling, smaller ones
+        // track the wire, and degenerate ones floor at MIN_SHARD_PAYLOAD.
+        assert_eq!(jumbo_shard_payload_for(64_000, v4), max_shard_payload());
+        let p = jumbo_shard_payload_for(4000, v4);
+        assert_eq!(p % 2, 0);
+        assert!(sealed_datagram_bytes(p) <= 4000 - 28);
+        assert_eq!(jumbo_shard_payload_for(100, v4), MIN_SHARD_PAYLOAD);
+    }
+
     /// Family selection: genuine v6 remotes get the v6 size; v4 — including the IPv4-mapped v6
     /// form a dual-stack `[::]` socket reports for a v4 client — keeps the v4 size.
     #[test]

crates/punktfunk-core/src/quic/control.rs

@@ -55,6 +55,36 @@ pub struct RfiRequest {
     pub last_frame: u32,
 }
 
+/// `host → client`, any time after [`Start`]: the video data plane's sealed shard payload
+/// changes mid-session (design/shard-payload-reneg.md Phase 1). Sent ONLY to a client whose
+/// [`Hello::max_shard_payload`] advertised per-frame geometry (0/absent = legacy — the host
+/// must never send this), and never above that advertised ceiling. Asymmetric semantics:
+///
+/// - **Shrink** (the mid-session MTU heal): the host may re-key its packetizer at the next
+///   AU boundary immediately after sending — per-frame pinning on the client makes the
+///   control-vs-datagram reorder race irrelevant and a smaller shard always fits existing
+///   buffers. The [`ShardPayloadAck`] is telemetry.
+/// - **Grow** (jumbo): the host must not emit a single sealed datagram above the OLD size
+///   until the ack arrives — the ack IS the gate, even when the client's buffers would
+///   happen to fit (the rule must not erode if the buffer strategy changes later).
+///
+/// No `effective_frame_index`: per-frame pinning makes it redundant — every video packet
+/// carries its own `shard_bytes` and the receiver follows each frame's pin.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct ShardPayloadChanged {
+    /// The new sealed shard payload in bytes (even, within the client's advertised bounds).
+    pub shard_payload: u16,
+}
+
+/// `client → host`: answer to [`ShardPayloadChanged`] — echoes the value the client applied.
+/// Only sent for an in-bounds request; an out-of-bounds one is dropped WITHOUT an ack (a
+/// buggy host must not read silence-then-garbage as a granted grow). The host treats the
+/// echoed value as the grant for a pending grow.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct ShardPayloadAck {
+    pub shard_payload: u16,
+}
+
 /// `client → host`, periodic: the client's observed data-plane loss, so the host can size FEC to
 /// the link instead of a flat percentage (adaptive FEC). `loss_ppm` is parts-per-million of shards
 /// that arrived missing-but-recovered (plus a bump when frames went unrecoverable) over the report
@@ -200,6 +230,10 @@ pub const MSG_SET_BITRATE: u8 = 0x05;
 pub const MSG_BITRATE_CHANGED: u8 = 0x06;
 /// Type byte of [`RfiRequest`].
 pub const MSG_RFI_REQUEST: u8 = 0x07;
+/// Type byte of [`ShardPayloadChanged`].
+pub const MSG_SHARD_PAYLOAD_CHANGED: u8 = 0x08;
+/// Type byte of [`ShardPayloadAck`].
+pub const MSG_SHARD_PAYLOAD_ACK: u8 = 0x09;
 /// Type byte of [`ProbeRequest`].
 pub const MSG_PROBE_REQUEST: u8 = 0x20;
 /// Type byte of [`ProbeResult`].
@@ -306,6 +340,46 @@ impl RfiRequest {
     }
 }
 
+impl ShardPayloadChanged {
+    pub fn encode(&self) -> Vec<u8> {
+        // magic[0..4] type[4] shard_payload[5..7]
+        let mut b = Vec::with_capacity(7);
+        b.extend_from_slice(CTL_MAGIC);
+        b.push(MSG_SHARD_PAYLOAD_CHANGED);
+        b.extend_from_slice(&self.shard_payload.to_le_bytes());
+        b
+    }
+
+    pub fn decode(b: &[u8]) -> Result<ShardPayloadChanged> {
+        if b.len() != 7 || &b[0..4] != CTL_MAGIC || b[4] != MSG_SHARD_PAYLOAD_CHANGED {
+            return Err(PunktfunkError::InvalidArg("bad ShardPayloadChanged"));
+        }
+        Ok(ShardPayloadChanged {
+            shard_payload: u16::from_le_bytes(b[5..7].try_into().unwrap()),
+        })
+    }
+}
+
+impl ShardPayloadAck {
+    pub fn encode(&self) -> Vec<u8> {
+        // magic[0..4] type[4] shard_payload[5..7]
+        let mut b = Vec::with_capacity(7);
+        b.extend_from_slice(CTL_MAGIC);
+        b.push(MSG_SHARD_PAYLOAD_ACK);
+        b.extend_from_slice(&self.shard_payload.to_le_bytes());
+        b
+    }
+
+    pub fn decode(b: &[u8]) -> Result<ShardPayloadAck> {
+        if b.len() != 7 || &b[0..4] != CTL_MAGIC || b[4] != MSG_SHARD_PAYLOAD_ACK {
+            return Err(PunktfunkError::InvalidArg("bad ShardPayloadAck"));
+        }
+        Ok(ShardPayloadAck {
+            shard_payload: u16::from_le_bytes(b[5..7].try_into().unwrap()),
+        })
+    }
+}
+
 impl LossReport {
     pub fn encode(&self) -> Vec<u8> {
         // magic[0..4] type[4] loss_ppm[5..9]
@@ -1146,6 +1220,33 @@ mod tests {
         assert!(SetBitrate::decode(&LossReport { loss_ppm: 7 }.encode()).is_err());
     }
 
+    #[test]
+    fn shard_payload_messages_roundtrip() {
+        for shard_payload in [512u16, 1216, 1408, 8908] {
+            let chg = ShardPayloadChanged { shard_payload };
+            assert_eq!(ShardPayloadChanged::decode(&chg.encode()).unwrap(), chg);
+            let ack = ShardPayloadAck { shard_payload };
+            assert_eq!(ShardPayloadAck::decode(&ack.encode()).unwrap(), ack);
+            // Identical payload shape — the type byte alone must keep the pair disjoint (a
+            // change echoed back must never re-decode as a change).
+            assert!(ShardPayloadChanged::decode(&ack.encode()).is_err());
+            assert!(ShardPayloadAck::decode(&chg.encode()).is_err());
+        }
+        // Exact length — no trailing bytes, no truncation.
+        let bytes = ShardPayloadChanged { shard_payload: 512 }.encode();
+        assert!(ShardPayloadChanged::decode(&[bytes.as_slice(), &[0]].concat()).is_err());
+        assert!(ShardPayloadChanged::decode(&bytes[..bytes.len() - 1]).is_err());
+        // Disjoint from the neighboring ids either side (0x07 RfiRequest / 0x20 ProbeRequest).
+        assert!(ShardPayloadChanged::decode(
+            &RfiRequest {
+                first_frame: 1,
+                last_frame: 2
+            }
+            .encode()
+        )
+        .is_err());
+    }
+
     #[test]
     fn probe_messages_roundtrip() {
         let req = ProbeRequest {

crates/punktfunk-core/src/quic/endpoint.rs

@@ -59,7 +59,25 @@ fn stream_transport_idle(idle: std::time::Duration) -> Arc<quinn::TransportConfi
     // PINGs quinn already expects to lose above a constrained hop — a lost probe settles the
     // search lower, exactly as it did before.
     let mut mtud = quinn::MtuDiscoveryConfig::default();
-    mtud.upper_bound(crate::config::video_datagram_udp_ceiling() as u16);
+    // Jumbo opt-in (design/shard-payload-reneg.md Phase 2): with `PUNKTFUNK_JUMBO=1` /
+    // `PUNKTFUNK_WIRE_MTU` > 1500 set, discovery probes up to the sealed JUMBO datagram
+    // size so a settled connection can PROVE a jumbo path — the actual grow stays
+    // client-ack-gated (`native/wire_mtu.rs`). The ceiling is per-ENDPOINT, not
+    // per-connection: with the opt-in set, connections to non-jumbo peers spend a few extra
+    // failed probes (one PTO each) settling lower; zero cost for anyone who doesn't opt in.
+    // Derived with the IPv4 overhead — a v6 peer's sealed jumbo target is smaller, so the
+    // ceiling covers it and discovery settles at the v6 path's own budget.
+    let probe_ceiling = match crate::config::jumbo_wire_mtu() {
+        Some(mtu) => {
+            let shard = crate::config::jumbo_shard_payload_for(
+                mtu,
+                std::net::IpAddr::V4(std::net::Ipv4Addr::UNSPECIFIED),
+            );
+            crate::config::sealed_datagram_bytes(shard) as u16
+        }
+        None => crate::config::video_datagram_udp_ceiling() as u16,
+    };
+    mtud.upper_bound(probe_ceiling);
     t.mtu_discovery_config(Some(mtud));
     Arc::new(t)
 }

crates/punktfunk-host/src/native.rs

@@ -1091,6 +1091,30 @@ async fn serve_session(
     // just never fires then.
     let (cursor_shape_tx, cursor_shape_rx) =
         tokio::sync::mpsc::unbounded_channel::<punktfunk_core::quic::CursorShape>();
+    // Mid-session shard renegotiation (design/shard-payload-reneg.md Phase 2): the wire-MTU
+    // watcher decides (constrained-path shrink / ack-gated jumbo grow), the control task
+    // writes the `ShardPayloadChanged` and routes the acks back, and the data-plane loop
+    // applies `Session::set_shard_payload` between AUs (drained next to `bitrate_rx`).
+    // Channels are wired unconditionally (they just never fire); the DRIVER exists only for
+    // a client that advertised `Hello::max_shard_payload` on a non-chunk-aligned session —
+    // PyroWave clients parse chunk-aligned AUs in windows of the `Welcome` value pinned at
+    // session start (read once over the C ABI), so those sessions keep the leg-1
+    // next-session clamp instead of a mid-stream re-key.
+    let (shard_change_tx, shard_change_rx) = tokio::sync::mpsc::unbounded_channel::<u16>();
+    let (shard_ack_tx, shard_ack_rx) = tokio::sync::mpsc::unbounded_channel::<u16>();
+    let (shard_apply_tx, shard_apply_rx) = std::sync::mpsc::channel::<usize>();
+    let shard_reneg = (hello.max_shard_payload > 0 && codec != crate::encode::Codec::PyroWave)
+        .then_some(wire_mtu::ShardReneg {
+            client_ceiling: hello.max_shard_payload,
+            change_tx: shard_change_tx,
+            ack_rx: shard_ack_rx,
+            apply_tx: shard_apply_tx,
+        });
+    // The session is real: watch this connection's MTU discovery settle and turn it into a
+    // path verdict (WARN + learned clamp for the next session on a constrained path; clears
+    // a stale clamp on a healthy one) — and, with the driver above, heal or grow THIS
+    // session mid-stream. Bounded ~10 s task unless a jumbo grow leaves it as revert guard.
+    wire_mtu::spawn_watch(conn.clone(), welcome.shard_payload as usize, shard_reneg);
     // Negotiated cursor forwarding: the HOST_CAP_CURSOR bit the Welcome advertised, read back
     // rather than recomputed (`handshake::cursor_forward` computed it once, with the encoder
     // blend-capability gate — re-running it here could drift, and would re-probe).
@@ -1146,6 +1170,8 @@ async fn serve_session(
         probe_result_rx,
         reconfig_result_rx,
         retarget_rx,
+        shard_change_rx,
+        shard_ack_tx,
         cursor_shape_rx,
         cursor_client_draws,
         clip_enabled,
@@ -1579,6 +1605,7 @@ async fn serve_session(
                         keyframe: keyframe_rx,
                         rfi: rfi_rx,
                         bitrate_rx,
+                        shard_rx: shard_apply_rx,
                         compositor,
                         gamescope_route,
                         bitrate_kbps,

crates/punktfunk-host/src/native/control.rs

@@ -43,6 +43,11 @@ pub(super) async fn run(
     // Host-initiated bitrate re-target (a rebuild re-resolved an Automatic rate): forwarded to
     // the client as a `BitrateChanged` so its controller's climb base tracks the real encoder.
     mut retarget_rx: tokio::sync::mpsc::UnboundedReceiver<u32>,
+    // Mid-session shard renegotiation (design/shard-payload-reneg.md): the wire-MTU watcher
+    // asks for a `ShardPayloadChanged` here (this task is the control stream's sole writer),
+    // and the client's `ShardPayloadAck`s flow back on `shard_ack_tx` — the grow gate.
+    mut shard_change_rx: tokio::sync::mpsc::UnboundedReceiver<u16>,
+    shard_ack_tx: tokio::sync::mpsc::UnboundedSender<u16>,
     mut cursor_shape_rx: tokio::sync::mpsc::UnboundedReceiver<punktfunk_core::quic::CursorShape>,
     cursor_client_draws: Arc<AtomicBool>,
     clip_enabled: Arc<AtomicBool>,
@@ -56,6 +61,9 @@ pub(super) async fn run(
     // Set once `clip_offer_rx` closes (coordinator gone / inert handle) so its `select!` branch
     // stops firing on a perpetually-ready `None`.
     let mut clip_offer_closed = false;
+    // Same discipline for the wire-MTU watcher's channel — its bounded lifetime ends mid-session
+    // on every healthy path.
+    let mut shard_change_closed = false;
     let mut active = initial_mode;
     // Host-side switch rate limit (a backstop against a hostile/broken client spamming
     // Reconfigure into pipeline-rebuild churn — the drain-to-newest in the data plane already
@@ -214,6 +222,16 @@ pub(super) async fn run(
                     if bitrate_tx.send(resolved).is_err() {
                         break; // data plane gone
                     }
+                } else if let Ok(ack) = punktfunk_core::quic::ShardPayloadAck::decode(&msg) {
+                    // Mid-session shard renegotiation: the client applied (or granted) a
+                    // geometry change. Forward to the wire-MTU watcher — for a grow this IS
+                    // the gate that lets the packetizer go above the old size. A dropped
+                    // send just means the watcher already ended (shrink acks are telemetry).
+                    tracing::info!(
+                        shard_payload = ack.shard_payload,
+                        "client acked shard-payload change"
+                    );
+                    let _ = shard_ack_tx.send(ack.shard_payload);
                 } else if let Ok(req) = ProbeRequest::decode(&msg) {
                     tracing::info!(
                         target_kbps = req.target_kbps,
@@ -317,6 +335,19 @@ pub(super) async fn run(
                     break;
                 }
             }
+            n = shard_change_rx.recv(), if !shard_change_closed => {
+                // Mid-session shard renegotiation: the wire-MTU watcher decided (shrink on a
+                // constrained-path verdict / ack-gated jumbo grow). Only ever fires toward a
+                // client that advertised `Hello::max_shard_payload` — the watcher owns that
+                // gate. `None` = the watcher's bounded lifetime ended (normal, NOT a session
+                // end): disable this branch, exactly the `clip_offer_closed` pattern — a
+                // closed mpsc yields `None` perpetually and would busy-spin the select.
+                let Some(n) = n else { shard_change_closed = true; continue };
+                let msg = punktfunk_core::quic::ShardPayloadChanged { shard_payload: n };
+                if io::write_msg(&mut ctrl_send, &msg.encode()).await.is_err() {
+                    break;
+                }
+            }
             shape = cursor_shape_rx.recv() => {
                 // Cursor-forward bridge (M2): the encode loop diffed a new pointer bitmap.
                 // Rare (shape changes are human-paced); ≤ ~58 KiB fits the u16 frame by

crates/punktfunk-host/src/native/handshake.rs

@@ -663,10 +663,9 @@ pub(super) async fn negotiate(
     let start =
         Start::decode(&io::read_msg(recv).await?).map_err(|e| anyhow!("Start decode: {e:?}"))?;
     bringup.mark("start");
-    // The session is real: watch this connection's MTU discovery settle and turn it into a
-    // path verdict (WARN + learned clamp for the next session on a constrained path; clears a
-    // stale clamp on a healthy one). Bounded ~10 s task, ends by itself.
-    wire_mtu::spawn_watch(conn.clone(), welcome.shard_payload as usize);
+    // The wire-MTU watch (`wire_mtu::spawn_watch`) is spawned by `serve_session` after the
+    // control-task channels exist — it now also DRIVES the mid-session shard renegotiation
+    // (design/shard-payload-reneg.md), which needs the control stream's writer.
     Ok::<_, anyhow::Error>((
         hello,
         welcome,

crates/punktfunk-host/src/native/stream.rs

@@ -762,6 +762,9 @@ fn send_loop(
     slice_wire: bool,
     burst_cap: Option<usize>,
     fec_target: Arc<AtomicU8>,
+    // Mid-session shard-payload re-keys from the wire-MTU watcher (validated + ack-gated
+    // there) — applied between AUs only (design/shard-payload-reneg.md Phase 1).
+    shard_rx: std::sync::mpsc::Receiver<usize>,
     stats: SendStats,
     // `Some` = the client advertised VIDEO_CAP_HOST_TIMING: emit one 0xCF datagram per AU right
     // after its last packet left the socket (capture→sent, the whole host pipeline incl. pacing).
@@ -818,6 +821,25 @@ fn send_loop(
         }
         // Adaptive FEC: pick up any new recovery target the control task set from client LossReports.
         apply_fec_target(&mut session, &fec_target);
+        // Mid-session shard renegotiation: apply a re-key from the wire-MTU watcher — between
+        // AUs only, NEVER with a streamed AU open (its shard-aligned tiling derives from the
+        // size it began with; same gate as the probe burst above). Drain to the newest; the
+        // protocol side (client advertisement, ack-gated grow) was enforced by the watcher.
+        if streamed.is_none() {
+            let mut want_shard = None;
+            while let Ok(s) = shard_rx.try_recv() {
+                want_shard = Some(s);
+            }
+            if let Some(s) = want_shard {
+                match session.set_shard_payload(s) {
+                    Ok(()) => tracing::info!(shard_payload = s, "wire shard payload re-keyed"),
+                    // Can't fire for a watcher-driven value (it validates the same bounds) —
+                    // belt-and-suspenders for a future driver.
+                    Err(e) => tracing::warn!(shard_payload = s, error = ?e,
+                        "shard re-key refused by session validation"),
+                }
+            }
+        }
         // Short timeout so we keep re-checking `stop` + probes when no frames are flowing.
         match frame_rx.recv_timeout(std::time::Duration::from_millis(50)) {
             Ok(send_msg) => {
@@ -1171,6 +1193,11 @@ pub(super) struct SessionContext {
     /// Accepted mid-stream bitrate changes (adaptive bitrate, already clamped) — the encoder
     /// alone is rebuilt in place at the new rate; capture + virtual output are untouched.
     pub(super) bitrate_rx: std::sync::mpsc::Receiver<u32>,
+    /// Mid-session shard-payload changes from the wire-MTU watcher (already validated +
+    /// protocol-gated there; a grow arrives only after the client's ack). Applied between
+    /// AUs via [`Session::set_shard_payload`] — the packetizer re-keys, capture/encoder/
+    /// virtual output are untouched (design/shard-payload-reneg.md Phase 1).
+    pub(super) shard_rx: std::sync::mpsc::Receiver<usize>,
     /// The resolved compositor backend (moot on Windows — `vdisplay::open` ignores it there).
     pub(super) compositor: crate::vdisplay::Compositor,
     /// This session's resolved gamescope sub-mode, or `None` for every other backend. Carried here
@@ -1385,6 +1412,7 @@ pub(super) fn virtual_stream(ctx: SessionContext, prepared: Option<PreparedDispl
         keyframe,
         rfi,
         bitrate_rx,
+        shard_rx,
         compositor,
         gamescope_route,
         mut bitrate_kbps,
@@ -1771,6 +1799,7 @@ pub(super) fn virtual_stream(ctx: SessionContext, prepared: Option<PreparedDispl
                     slice_wire,
                     burst_cap,
                     fec_target,
+                    shard_rx,
                     send_stats,
                     timing_conn,
                     phase_send,

crates/punktfunk-host/src/native/wire_mtu.rs

@@ -30,10 +30,30 @@ use std::net::IpAddr;
 use std::sync::{Mutex, OnceLock};
 
 use punktfunk_core::config::{
-    mtu1500_shard_payload_for, sealed_datagram_bytes, shard_payload_for_udp_budget,
-    shard_payload_for_wire_mtu, video_datagram_udp_ceiling,
+    jumbo_shard_payload_for, jumbo_wire_mtu, mtu1500_shard_payload_for, sealed_datagram_bytes,
+    shard_payload_for_udp_budget, shard_payload_for_wire_mtu, video_datagram_udp_ceiling,
 };
 
+/// Everything the MID-SESSION renegotiation driver needs (design/shard-payload-reneg.md
+/// Phase 2) — `None` at [`spawn_watch`] makes the watcher observe-and-learn only (leg-1
+/// behavior). Constructed ONLY when the client's `Hello::max_shard_payload` advertised
+/// per-frame geometry AND the session's wire is not chunk-aligned: a PyroWave client parses
+/// chunk-aligned AUs in windows of the `Welcome` value pinned at session start (Apple
+/// `Stage2Pipeline` / `pf-client-core` video.rs read it once over the C ABI), so re-keying
+/// such a session mid-stream would corrupt its parse — those sessions keep the leg-1
+/// next-session clamp instead.
+pub(super) struct ShardReneg {
+    /// The client's advertised receive ceiling (bytes of shard; > 0 by construction).
+    pub client_ceiling: u16,
+    /// → control task (the control stream's sole writer): send `ShardPayloadChanged{n}`.
+    pub change_tx: tokio::sync::mpsc::UnboundedSender<u16>,
+    /// ← control task: the client's `ShardPayloadAck`s (the grow gate).
+    pub ack_rx: tokio::sync::mpsc::UnboundedReceiver<u16>,
+    /// → data plane: apply [`Session::set_shard_payload`] between AUs
+    /// (drained next to `bitrate_rx` in the encode loop).
+    pub apply_tx: std::sync::mpsc::Sender<usize>,
+}
+
 /// Measured UDP-payload budget per peer IP, learned from live control connections whose MTU
 /// discovery settled below the video-datagram ceiling. In-memory only: a host restart
 /// re-learns in one session, and entries self-correct (a later ceiling-hit erases, a lower
@@ -96,15 +116,26 @@ fn resolve(env_wire_mtu: Option<usize>, learned_udp_budget: Option<u16>, peer: I
 }
 
 /// Sample the control connection's discovered MTU after the search has settled and turn it
-/// into a verdict. Spawned once per negotiated session; the task ends by itself after the
-/// final sample (bounded ~10 s lifetime, holding only a cheap `Connection` handle).
-pub(super) fn spawn_watch(conn: quinn::Connection, session_shard_payload: usize) {
+/// into a verdict — and, with a [`ShardReneg`] driver, act on it MID-SESSION
+/// (design/shard-payload-reneg.md Phase 2): a below-ceiling verdict shrinks the live wire at
+/// the ~3–10 s mark (session 1 heals instead of staying black), and a settled-at-jumbo
+/// verdict grows it, ack-gated, when the operator opted in. Spawned once per negotiated
+/// session; without a grow the task ends after the final sample (bounded ~10 s lifetime,
+/// holding only a cheap `Connection` handle) — after a grow it stays as the revert guard
+/// until the connection closes.
+pub(super) fn spawn_watch(
+    conn: quinn::Connection,
+    session_shard_payload: usize,
+    reneg: Option<ShardReneg>,
+) {
     tokio::spawn(async move {
         let peer = conn.remote_address().ip();
         let ceiling = video_datagram_udp_ceiling() as u16;
         // Discovery finishes in a handful of RTTs on a LAN (well under the first sample) but
         // needs a loss timeout per failed probe on a constrained path — the second sample
-        // covers that with margin. Max, because discovery only ever raises `current_mtu`.
+        // covers that with margin. Max, because discovery only ever raises `current_mtu`
+        // (the post-grow revert guard below re-reads it live, where blackhole detection CAN
+        // lower it again).
         let mut settled = 0u16;
         for wait_s in [3u64, 7] {
             tokio::time::sleep(std::time::Duration::from_secs(wait_s)).await;
@@ -113,6 +144,9 @@ pub(super) fn spawn_watch(conn: quinn::Connection, session_shard_payload: usize)
                 break;
             }
         }
+        // The wire this session is CURRENTLY sealed at — moves on a mid-session shrink/grow.
+        let mut current = session_shard_payload;
+        let mut reneg = reneg;
         if settled >= ceiling {
             // The path carries full-size video datagrams — erase any stale learned clamp so
             // the next session returns to the default wire.
@@ -120,34 +154,116 @@ pub(super) fn spawn_watch(conn: quinn::Connection, session_shard_payload: usize)
                 tracing::info!(peer = %peer,
                     "wire MTU: path re-measured at full size — learned clamp cleared");
             }
-            return;
-        }
-        // A closed connection stops discovering, so a session that ended before the final
-        // sample proves nothing (a healthy high-RTT path could still be mid-search): learn
-        // only from a connection that stayed alive through the whole window.
-        if conn.close_reason().is_some() {
-            return;
-        }
-        learned().lock().unwrap().insert(peer, settled);
-        if sealed_datagram_bytes(session_shard_payload) <= settled as usize {
-            // This session was already clamped small enough — the path is still constrained
-            // (keep the record fresh) but video fits, so no alarm.
-            tracing::info!(peer = %peer, discovered_udp_mtu = settled,
-                "wire MTU: constrained path re-measured; this session's video is sized to fit");
         } else {
-            tracing::warn!(
-                peer = %peer,
-                discovered_udp_mtu = settled,
-                needed_udp_mtu = ceiling,
-                "wire MTU: this path CANNOT carry full-size video datagrams — the control \
-                 plane works but every video packet is oversized for a hop, which streams as \
-                 an endless black screen with zero reported loss. Typical cause: a VPN/overlay \
-                 adapter (Tailscale / Cloudflare WARP / ZeroTier) claiming the LAN route, or a \
-                 lowered NIC MTU — compare `ping <client> -f -l 1450` vs `-l 1200` and check \
-                 `netsh interface ipv4 show subinterfaces` (Windows) / `ip link` (Linux). The \
-                 measured budget is recorded: the NEXT session from this client sizes video to \
-                 fit automatically. To pin it for all sessions set PUNKTFUNK_WIRE_MTU."
-            );
+            // A closed connection stops discovering, so a session that ended before the final
+            // sample proves nothing (a healthy high-RTT path could still be mid-search): learn
+            // only from a connection that stayed alive through the whole window.
+            if conn.close_reason().is_some() {
+                return;
+            }
+            learned().lock().unwrap().insert(peer, settled);
+            if sealed_datagram_bytes(current) <= settled as usize {
+                // This session was already clamped small enough — the path is still constrained
+                // (keep the record fresh) but video fits, so no alarm.
+                tracing::info!(peer = %peer, discovered_udp_mtu = settled,
+                    "wire MTU: constrained path re-measured; this session's video is sized to fit");
+            } else {
+                tracing::warn!(
+                    peer = %peer,
+                    discovered_udp_mtu = settled,
+                    needed_udp_mtu = ceiling,
+                    "wire MTU: this path CANNOT carry full-size video datagrams — the control \
+                     plane works but every video packet is oversized for a hop, which streams as \
+                     an endless black screen with zero reported loss. Typical cause: a VPN/overlay \
+                     adapter (Tailscale / Cloudflare WARP / ZeroTier) claiming the LAN route, or a \
+                     lowered NIC MTU — compare `ping <client> -f -l 1450` vs `-l 1200` and check \
+                     `netsh interface ipv4 show subinterfaces` (Windows) / `ip link` (Linux). The \
+                     measured budget is recorded: the NEXT session from this client sizes video to \
+                     fit automatically. To pin it for all sessions set PUNKTFUNK_WIRE_MTU."
+                );
+                // Phase 2 down-leg: heal THIS session at the verdict mark. Shrink is sent
+                // and applied immediately — per-frame pinning on the client makes ordering
+                // irrelevant and smaller always fits; the ack is telemetry. The learned
+                // record above still makes session 2 START right.
+                if let Some(r) = reneg.as_ref() {
+                    let target = shard_payload_for_udp_budget(settled as usize, peer);
+                    if target < current
+                        && r.change_tx.send(target as u16).is_ok()
+                        && r.apply_tx.send(target).is_ok()
+                    {
+                        tracing::info!(
+                            peer = %peer,
+                            shard_payload = target,
+                            was = current,
+                            "wire MTU: video re-keyed mid-session to fit the constrained path \
+                             — the stream heals now instead of on the next connect"
+                        );
+                        current = target;
+                    }
+                }
+            }
+        }
+        // Phase 2 up-leg: jumbo grow — operator opt-in (PUNKTFUNK_JUMBO / PUNKTFUNK_WIRE_MTU
+        // > 1500, which also raised the endpoint's probe ceiling so `settled` can even reach
+        // here), client-advertised headroom, and a settled-at-jumbo proof. The grow is
+        // ACK-GATED: not one sealed datagram above the old size leaves before the client's
+        // ack, even though its buffers are statically sized — the rule must not erode.
+        let (Some(mtu), Some(r)) = (jumbo_wire_mtu(), reneg.as_mut()) else {
+            return;
+        };
+        let target = jumbo_shard_payload_for(mtu, peer).min(r.client_ceiling as usize);
+        let target = target - target % 2;
+        if target <= current || (settled as usize) < sealed_datagram_bytes(target) {
+            return;
+        }
+        if r.change_tx.send(target as u16).is_err() {
+            return;
+        }
+        let acked = tokio::time::timeout(std::time::Duration::from_secs(5), async {
+            while let Some(v) = r.ack_rx.recv().await {
+                if v as usize == target {
+                    return true;
+                }
+            }
+            false
+        })
+        .await
+        .unwrap_or(false);
+        if !acked {
+            tracing::warn!(peer = %peer, shard_payload = target,
+                "wire MTU: jumbo grow not acked — staying at the current wire");
+            return;
+        }
+        if r.apply_tx.send(target).is_err() {
+            return;
+        }
+        tracing::info!(
+            peer = %peer,
+            shard_payload = target,
+            was = current,
+            wire_mtu = mtu,
+            "wire MTU: jumbo grow acked and applied — packets-per-frame cut ~6×"
+        );
+        current = target;
+        // Revert guard: a mis-proven jumbo hop must self-correct instead of blackholing.
+        // quinn's PMTU blackhole detection lowers `current_mtu` when the big packets start
+        // vanishing; sample it and shrink back through the same path the down-leg uses.
+        loop {
+            tokio::time::sleep(std::time::Duration::from_secs(5)).await;
+            if conn.close_reason().is_some() {
+                return;
+            }
+            let mtu_now = conn.stats().path.current_mtu;
+            if (mtu_now as usize) < sealed_datagram_bytes(current) {
+                let back = shard_payload_for_udp_budget(mtu_now as usize, peer);
+                tracing::warn!(peer = %peer, discovered_udp_mtu = mtu_now,
+                    shard_payload = back, was = current,
+                    "wire MTU: jumbo path stopped fitting — reverting the wire to match");
+                if r.change_tx.send(back as u16).is_err() || r.apply_tx.send(back).is_err() {
+                    return;
+                }
+                current = back;
+            }
         }
     });
 }

include/punktfunk_core.h

@@ -792,6 +792,16 @@
 #define MSG_RFI_REQUEST 7
 #endif
 
+#if defined(PUNKTFUNK_FEATURE_QUIC)
+// Type byte of [`ShardPayloadChanged`].
+#define MSG_SHARD_PAYLOAD_CHANGED 8
+#endif
+
+#if defined(PUNKTFUNK_FEATURE_QUIC)
+// Type byte of [`ShardPayloadAck`].
+#define MSG_SHARD_PAYLOAD_ACK 9
+#endif
+
 #if defined(PUNKTFUNK_FEATURE_QUIC)
 // Type byte of [`ProbeRequest`].
 #define MSG_PROBE_REQUEST 32