perf(latency): tier-0 attribution + tier-1 send-path levers from the latency plan

design/latency-reduction-2026-07.md T0.1/T0.2/T1.2/T1.3:

- T1.2 rate-capped front-loaded pacing: the paced overflow's budget is now
  min(0.9x slack, overflow wire time at ~3x the live encoder bitrate)
  (PUNKTFUNK_PACE_FACTOR, 0 = legacy deadline-only spread). A 300 KB-1 MB
  frame's tail leaves in ~2-5 ms instead of smearing across ~15 ms at 60 fps;
  GameStream schedule byte-identical (pins unchanged).
- T1.3 data-first wire order: packetize emits every block's data shards before
  any parity (per-block parity pools keep all blocks' parity alive for the
  second pass), so lossless completion stops waiting behind the parity tail.
  EOF = last emitted packet; receiver already order-agnostic.
- T0.1 staged 0xCF: HostTiming gains an append-extensible per-stage tail
  (queue/encode/pace us; seal+channel-wait derived as residual) - no cap bit
  needed, old peers read the 13-byte prefix. Joined client-side into
  Stats::host_{queue,encode,xfer,pace}_ms, the OSD detailed tier, and the
  probe's report.
- T0.2 true on-glass present timing: VK_KHR_present_id/present_wait enabled
  when supported; a PresentTimer waiter thread resolves each present id to
  real visibility, replacing the submit-time display stamp (which undercounts
  by up to a refresh and hides a silent-FIFO standing queue).

Validated on .21: core 185 + host 185 tests, pf-presenter 19, clippy
-D warnings across all five touched crates; loss-harness recovery curve
unchanged; C ABI harness round-trips.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

crates/punktfunk-core/src/packet/packetize.rs

@@ -32,10 +32,13 @@ pub struct Packetizer {
     /// Every other data shard is a `shard_payload`-sized slice straight into the frame buffer —
     /// blocks are consecutive shard ranges, so only the frame's last shard can be partial.
     tail: Vec<u8>,
-    /// Reusable parity buffers for [`ErasureCoder::encode_into`] (plan Phase 1.4): grows once
-    /// to the session's high-water recovery count, then every block's parity is generated
-    /// into it with zero allocations.
-    recovery: Vec<Vec<u8>>,
+    /// Reusable PER-BLOCK parity buffers for [`ErasureCoder::encode_into`] (plan Phase 1.4):
+    /// one pool per block index, each growing once to its high-water recovery count, then
+    /// every frame's parity is generated into them with zero allocations. Per-block (not one
+    /// shared pool) because the data-first wire order (latency plan T1.3) emits every block's
+    /// DATA shards before any block's parity — all blocks' parity must stay alive until the
+    /// frame's second emission pass.
+    recovery: Vec<Vec<Vec<u8>>>,
 }
 
 impl Packetizer {
@@ -103,6 +106,15 @@ impl Packetizer {
     /// ([`Session::seal_frame`](crate::session::Session::seal_frame)). An `emit` error aborts
     /// the frame mid-way (packet numbering has already advanced — callers treat it as fatal).
     ///
+    /// Wire order is DATA-FIRST (latency plan T1.3): every block's data shards in block order,
+    /// then every block's parity shards in block order. In the lossless case a frame completes
+    /// the moment its last DATA shard arrives, so the completion-gating packet no longer sits
+    /// behind the parity tail of the paced spread (~fec% of the spread saved). The receiver is
+    /// order-agnostic (headers are self-describing; the reassembler completes each block on
+    /// `data + recovery ≥ k`), so this is not a wire-format change. `FLAG_SOF` stays on the
+    /// first emitted packet (block 0, shard 0); `FLAG_EOF` marks the last emitted packet —
+    /// the final parity shard, or the final data shard of a FEC-free frame.
+    ///
     /// `frame_index`: `Some(i)` stamps the AU with the caller's index — the punktfunk/1 encode
     /// loop numbers video AUs itself so the encoder's RFI bookkeeping (LTR marks, DPB timestamps)
     /// is 1:1 with what the client sees, surviving encoder rebuilds/resets that restart internal
@@ -152,7 +164,6 @@ impl Packetizer {
             self.tail[..rem].copy_from_slice(&frame[full_shards * payload..]);
         }
         let tail = &self.tail;
-        let recovery_pool = &mut self.recovery;
         let shard_at = |s: usize| -> &[u8] {
             if s < full_shards {
                 &frame[s * payload..(s + 1) * payload]
@@ -160,41 +171,30 @@ impl Packetizer {
                 tail.as_slice()
             }
         };
+        // Per-block shard geometry (deterministic — recomputed in both passes).
+        let block_data_count = |b: usize| ((b + 1) * max_block).min(total_data) - b * max_block;
 
+        // One parity pool per block, reused across frames (steady-state zero-alloc).
+        if self.recovery.len() < block_count {
+            self.recovery.resize_with(block_count, Vec::new);
+        }
+
+        // Total parity across the frame decides where FLAG_EOF lands (the last emitted packet).
+        let mut total_recovery = 0usize;
         for b in 0..block_count {
-            let first = b * max_block;
-            let last = ((b + 1) * max_block).min(total_data);
-            let block_data_count = last - first;
-
-            // This block's data shards: references into `frame` (plus the staged tail).
-            let data_shards: Vec<&[u8]> = (first..last).map(shard_at).collect();
-
-            let recovery_count = self.fec.recovery_for(block_data_count);
-            coder.encode_into(&data_shards, recovery_count, recovery_pool)?;
-            let recovery = &*recovery_pool;
-            let total_shards = block_data_count + recovery_count;
-            if total_shards > u16::MAX as usize {
+            let k = block_data_count(b);
+            let m = self.fec.recovery_for(k);
+            if k + m > u16::MAX as usize {
                 return Err(PunktfunkError::Unsupported("block shard count exceeds u16"));
             }
+            total_recovery += m;
+        }
 
-            for shard_index in 0..total_shards {
-                let body: &[u8] = if shard_index < block_data_count {
-                    data_shards[shard_index]
-                } else {
-                    &recovery[shard_index - block_data_count]
-                };
-
-                let seq = self.next_seq;
-                self.next_seq = self.next_seq.wrapping_add(1);
-
-                let mut flags = FLAG_PIC;
-                if b == 0 && shard_index == 0 {
-                    flags |= FLAG_SOF;
-                }
-                if b + 1 == block_count && shard_index + 1 == total_shards {
-                    flags |= FLAG_EOF;
-                }
-
+        let mut emit_one =
+            |next_seq: &mut u32, b: usize, shard_index: usize, body: &[u8], flags: u8| {
+                let seq = *next_seq;
+                *next_seq = next_seq.wrapping_add(1);
+                let k = block_data_count(b);
                 let hdr = PacketHeader {
                     pts_ns,
                     frame_index,
@@ -203,8 +203,8 @@ impl Packetizer {
                     user_flags,
                     block_index: b as u16,
                     block_count: block_count as u16,
-                    data_shards: block_data_count as u16,
-                    recovery_shards: recovery_count as u16,
+                    data_shards: k as u16,
+                    recovery_shards: self.fec.recovery_for(k) as u16,
                     shard_index: shard_index as u16,
                     shard_bytes: payload as u16,
                     magic: PUNKTFUNK_MAGIC,
@@ -212,9 +212,48 @@ impl Packetizer {
                     fec_scheme: coder.scheme() as u8,
                     flags,
                 };
-                emit(&hdr, body)?;
+                emit(&hdr, body)
+            };
+        let mut next_seq = self.next_seq;
+
+        // Pass 1 — per block: generate parity into the block's pool, emit the DATA shards.
+        for b in 0..block_count {
+            let first = b * max_block;
+            let k = block_data_count(b);
+
+            // This block's data shards: references into `frame` (plus the staged tail).
+            let data_shards: Vec<&[u8]> = (first..first + k).map(shard_at).collect();
+            let recovery_count = self.fec.recovery_for(k);
+            coder.encode_into(&data_shards, recovery_count, &mut self.recovery[b])?;
+
+            for (shard_index, body) in data_shards.iter().enumerate() {
+                let mut flags = FLAG_PIC;
+                if b == 0 && shard_index == 0 {
+                    flags |= FLAG_SOF;
+                }
+                if total_recovery == 0 && b + 1 == block_count && shard_index + 1 == k {
+                    flags |= FLAG_EOF;
+                }
+                emit_one(&mut next_seq, b, shard_index, body, flags)?;
             }
         }
+
+        // Pass 2 — per block: emit the parity shards (the frame's tail on the wire).
+        let mut parity_left = total_recovery;
+        for b in 0..block_count {
+            let k = block_data_count(b);
+            let recovery_count = self.fec.recovery_for(k);
+            for r in 0..recovery_count {
+                parity_left -= 1;
+                let mut flags = FLAG_PIC;
+                if parity_left == 0 {
+                    flags |= FLAG_EOF;
+                }
+                let body: &[u8] = &self.recovery[b][r];
+                emit_one(&mut next_seq, b, k + r, body, flags)?;
+            }
+        }
+        self.next_seq = next_seq;
         Ok(())
     }
 }

crates/punktfunk-core/src/packet/tests.rs

@@ -370,16 +370,94 @@ fn e2e_roundtrip(
 /// 100 bytes / 16 = 7 shards → blocks of (4 data + 2 rec) and (3 data + 2 rec).
 #[test]
 fn e2e_multiblock_loss_reorder_dup_gf16() {
-    // Packet order: blk0 = idx 0..6 (4 data + 2 rec), blk1 = idx 6..11 (3 data + 2 rec).
+    // Data-first wire order (T1.3): blk0 data = idx 0..4, blk1 data = idx 4..7,
+    // blk0 rec = idx 7..9, blk1 rec = idx 9..11.
     // Kill 2 data in block 0 and 1 data in block 1 — all within the 50% budget.
-    e2e_roundtrip(FecScheme::Gf16, 100, 50, &[0, 2, 7], false);
-    e2e_roundtrip(FecScheme::Gf16, 100, 50, &[0, 2, 7], true);
+    e2e_roundtrip(FecScheme::Gf16, 100, 50, &[0, 2, 5], false);
+    e2e_roundtrip(FecScheme::Gf16, 100, 50, &[0, 2, 5], true);
 }
 
 #[test]
 fn e2e_multiblock_loss_reorder_dup_gf8() {
-    e2e_roundtrip(FecScheme::Gf8, 100, 50, &[1, 3, 8], false);
-    e2e_roundtrip(FecScheme::Gf8, 100, 50, &[1, 3, 8], true);
+    e2e_roundtrip(FecScheme::Gf8, 100, 50, &[1, 3, 6], false);
+    e2e_roundtrip(FecScheme::Gf8, 100, 50, &[1, 3, 6], true);
+}
+
+/// T1.3 pin: the wire order is DATA-FIRST — every block's data shards in block order, then
+/// every block's parity in block order — so the lossless-completion-gating packet (the last
+/// data shard) never sits behind parity in the paced spread. SOF on the first emitted packet,
+/// EOF on the last (a parity shard whenever the frame carries FEC).
+#[test]
+fn packetize_emits_all_data_before_any_parity() {
+    use zerocopy::FromBytes;
+    let cfg = e2e_config(FecScheme::Gf16, 50);
+    let coder = coder_for(FecScheme::Gf16);
+    let mut pk = Packetizer::new(&cfg);
+    // 100 B / 16 → 7 data shards → blocks (4 data + 2 rec) + (3 data + 2 rec).
+    let src: Vec<u8> = (0..100).map(|i| (i * 31 + 3) as u8).collect();
+    let pkts = pk.packetize(&src, 1, 0, coder.as_ref()).unwrap();
+    assert_eq!(pkts.len(), 11);
+    let hdrs: Vec<PacketHeader> = pkts
+        .iter()
+        .map(|p| PacketHeader::read_from_bytes(&p[..HEADER_LEN]).unwrap())
+        .collect();
+    // (block_index, shard_index) in emission order.
+    let layout: Vec<(u16, u16)> = hdrs
+        .iter()
+        .map(|h| (h.block_index, h.shard_index))
+        .collect();
+    assert_eq!(
+        layout,
+        vec![
+            (0, 0),
+            (0, 1),
+            (0, 2),
+            (0, 3), // blk0 data
+            (1, 0),
+            (1, 1),
+            (1, 2), // blk1 data
+            (0, 4),
+            (0, 5), // blk0 parity
+            (1, 3),
+            (1, 4), // blk1 parity
+        ],
+        "data-first wire order"
+    );
+    // A shard is parity iff shard_index >= data_shards; no parity may precede any data.
+    let first_parity = hdrs
+        .iter()
+        .position(|h| h.shard_index >= h.data_shards)
+        .unwrap();
+    assert!(
+        hdrs[first_parity..]
+            .iter()
+            .all(|h| h.shard_index >= h.data_shards),
+        "no data shard after the first parity shard"
+    );
+    // Stream seqs stay strictly sequential in emission order (the nonce contract).
+    for (i, w) in hdrs.windows(2).enumerate() {
+        assert_eq!(w[1].stream_seq, w[0].stream_seq + 1, "seq gap at {i}");
+    }
+    assert_eq!(hdrs[0].flags & FLAG_SOF, FLAG_SOF, "SOF on first packet");
+    assert_eq!(
+        hdrs.last().unwrap().flags & FLAG_EOF,
+        FLAG_EOF,
+        "EOF on last (parity) packet"
+    );
+    assert_eq!(
+        hdrs.iter().filter(|h| h.flags & FLAG_EOF != 0).count(),
+        1,
+        "exactly one EOF"
+    );
+
+    // FEC-free frame: EOF falls on the last data shard instead.
+    let cfg0 = e2e_config(FecScheme::Gf16, 0);
+    let mut pk0 = Packetizer::new(&cfg0);
+    let pkts0 = pk0.packetize(&src, 2, 0, coder.as_ref()).unwrap();
+    assert_eq!(pkts0.len(), 7, "no parity at 0% FEC");
+    let last = PacketHeader::read_from_bytes(&pkts0.last().unwrap()[..HEADER_LEN]).unwrap();
+    assert_eq!(last.flags & FLAG_EOF, FLAG_EOF, "EOF on last data shard");
+    assert!(last.shard_index < last.data_shards, "last packet is data");
 }
 
 /// Zero losses, in order: the pure fast path (no codec call, recovered == 0) must still