feat(recovery): clean mid-stream loss recovery — freeze-until-reanchor + AMD LTR-RFI

Removes the "gray frames with motion" artifact on Vulkan-Video clients and lets
AMD/NVENC hosts re-anchor after loss WITHOUT a 20-40x IDR spike.

Client (pf-client-core): after a reference loss the hardware decoder conceals the
missing-reference deltas (on RADV, a gray plate with new motion painted over) and
returns Ok. The pump now freezes on the last good picture until a clean re-anchor
instead of showing the concealment — lifting on a real IDR, an intra-refresh
recovery mark (2nd wave boundary), or an LTR-RFI recovery anchor (1st). The
frame_index gap is the early, precise loss signal and drives an RFI request.

Host recovery signals (inert unless the backend supports them):
- USER_FLAG_RECOVERY_POINT — intra-refresh wave boundary (NVENC constrained GDR).
- USER_FLAG_RECOVERY_ANCHOR — AMD LTR reference-frame-invalidation recovery frame.

AMD LTR-RFI (encode/windows/amf.rs) — the AMD twin of NVENC RFI. AMF's AVC/HEVC API
has no constrained-intra property (intra-refresh cannot heal; PSNR-proven), so the
only clean-recovery lever is user LTR: mark frames as long-term references, and on
loss force the next frame to re-reference the newest known-good one — a clean
P-frame, not an IDR. Two rotating LTR slots, ~0.5s mark cadence, on by default for
AVC/HEVC (PUNKTFUNK_NO_AMF_LTR disables). invalidate_ref_frames picks the newest LTR
before the loss; a range older than the live slots falls back to a keyframe.

Protocol (punktfunk-core): RfiRequest control message + NativeClient::request_rfi().
Host: RfiRequest dispatch -> invalidate_ref_frames (IDR fallback); an RFI success
anchors the keyframe cooldown so the client's frames_dropped echo of the same loss
is coalesced away rather than emitting a redundant IDR.

Spike: synthetic NV12 GPU source for headless AMF encoder testing.

Validated: core rfi_request_roundtrip; pf-client-core 31 unit tests
(incl. an_rfi_anchor_lifts_immediately); punktfunk-host builds + 271 tests on Linux;
punktfunk-host builds clean on Windows; real AMD iGPU spike (invalidate at frame 90
forced re-reference to LTR frame 60 — 180 frames, keyframes=1, no recovery IDR).

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

clients/windows/src/app/hosts.rs

@@ -189,14 +189,21 @@ fn status_row(online: Option<bool>, badge: &str, kind: Pill) -> Element {
 
 /// The in-tile rename editor (ContentDialog can't hold a text field): name box + save/cancel.
 /// No tap-to-connect while editing — a click into the box would bubble `Tapped` to the region.
+/// `initial` seeds the text box's displayed value and is CONSTANT for the life of the edit — the
+/// field is uncontrolled, its live value kept in `live` (read at Save). Driving a *controlled* box
+/// from an always-deferred `AsyncSetState` round-trip fights the caret on fast typing and can drop
+/// the last char if Save is clicked before the write lands; an uncontrolled box + a ref sidesteps
+/// both (and skips a full-page re-render per keystroke). See the seed block in `hosts_page`.
 fn rename_editor(
-    draft: &str,
+    initial: &str,
     fp: String,
+    live: HookRef<String>,
     set_rename: AsyncSetState<Option<(String, String)>>,
 ) -> Element {
     let commit = {
-        let (fp, draft, sr) = (fp.clone(), draft.to_string(), set_rename.clone());
+        let (fp, live, sr) = (fp.clone(), live.clone(), set_rename.clone());
         move || {
+            let draft = live.borrow();
             let name = draft.trim();
             if !name.is_empty() {
                 let mut known = KnownHosts::load();
@@ -209,12 +216,12 @@ fn rename_editor(
         }
     };
     let on_changed = {
-        let sr = set_rename.clone();
-        move |s: String| sr.call(Some((fp.clone(), s)))
+        let live = live.clone();
+        move |s: String| live.set(s)
     };
     card(
         vstack((
-            text_box(draft)
+            text_box(initial)
                 .placeholder_text("Host name")
                 .on_text_changed(on_changed),
             hstack((
@@ -240,6 +247,14 @@ pub(crate) fn hosts_page(props: &HostsProps, cx: &mut RenderCx) -> Element {
     let set_screen = &props.svc.set_screen;
     let set_status = &props.svc.set_status;
     let (manual, set_manual) = cx.use_state(String::new());
+    // The Add-host field's live value, read by Connect at click time. This page's `use_state` is
+    // unreliable as the click's source of truth: while the modal is open the page usually has no
+    // reason to re-render (you open it precisely because the host ISN'T being discovered, so no
+    // discovery tick fires), and the top-down reconcile skips this unchanged-props subtree — so a
+    // sync `set_manual` write never re-renders the Connect button to re-capture the address, and it
+    // would connect to the empty mount-time value. Mirror every keystroke into this stable ref (the
+    // pair-screen PIN pattern). `manual` still drives the text box's displayed value.
+    let manual_live = cx.use_ref(String::new());
     // "Add host" modal open state lives in ROOT (see `HostsProps`).
     let show_add = props.show_add;
     let set_show_add = &props.set_show_add;
@@ -249,6 +264,18 @@ pub(crate) fn hosts_page(props: &HostsProps, cx: &mut RenderCx) -> Element {
     let rename = props.rename.clone();
     let set_forget = &props.set_forget;
     let set_rename = &props.set_rename;
+    // The live rename draft, read at Save time (see `rename_editor`). Root `rename` carries only the
+    // INITIAL name, so it no longer round-trips per keystroke. Seed the draft each time the rename
+    // TARGET changes (start, cancel, or a switch to another host).
+    let rename_draft = cx.use_ref(String::new());
+    let rename_seed = cx.use_ref(Option::<String>::None);
+    {
+        let active = rename.as_ref().map(|(fp, _)| fp.clone());
+        if *rename_seed.borrow() != active {
+            rename_draft.set(rename.as_ref().map(|(_, n)| n.clone()).unwrap_or_default());
+            rename_seed.set(active);
+        }
+    }
     let hover = Hover {
         current: props.hover.clone(),
         set: props.set_hover.clone(),
@@ -393,8 +420,13 @@ pub(crate) fn hosts_page(props: &HostsProps, cx: &mut RenderCx) -> Element {
         for k in &known.hosts {
             // Rust 2021 (no let-chains): match the "this tile is being renamed" case explicitly.
             if matches!(&rename, Some((fp, _)) if fp == &k.fp_hex) {
-                let (fp, draft) = rename.clone().unwrap();
-                tiles.push(rename_editor(&draft, fp, set_rename.clone()));
+                let (fp, initial) = rename.clone().unwrap();
+                tiles.push(rename_editor(
+                    &initial,
+                    fp,
+                    rename_draft.clone(),
+                    set_rename.clone(),
+                ));
                 continue;
             }
             let target = Target {
@@ -595,14 +627,15 @@ pub(crate) fn hosts_page(props: &HostsProps, cx: &mut RenderCx) -> Element {
     // field). The scrim border fills the cell and is hit-testable, so it blocks the page behind;
     // it closes only via Cancel/Connect (a scrim tap would bubble `Tapped` up from the card too).
     let connect_manual = {
-        let (ctx2, ss, st, text, sa) = (
+        let (ctx2, ss, st, live, sa) = (
             ctx.clone(),
             set_screen.clone(),
             set_status.clone(),
-            manual.clone(),
+            manual_live.clone(),
             set_show_add.clone(),
         );
         move || {
+            let text = live.borrow();
             let text = text.trim();
             if text.is_empty() {
                 return;
@@ -640,7 +673,13 @@ pub(crate) fn hosts_page(props: &HostsProps, cx: &mut RenderCx) -> Element {
             text_box(manual)
                 .header("Address")
                 .placeholder_text("192.168.1.20  or  my-pc.local")
-                .on_text_changed(move |s| set_manual.call(s))
+                .on_text_changed({
+                    let live = manual_live.clone();
+                    move |s: String| {
+                        live.set(s.clone());
+                        set_manual.call(s);
+                    }
+                })
                 .margin(edges(0.0, 6.0, 0.0, 0.0)),
             hstack((
                 button("Connect")

clients/windows/src/app/pair.rs

@@ -14,21 +14,28 @@ pub(crate) fn pair_page(props: &Svc, cx: &mut RenderCx) -> Element {
     let set_screen = &props.set_screen;
     let set_status = &props.set_status;
     let (code, set_code) = cx.use_state(String::new());
+    // The PIN's live value, read directly by the click handler. This page's props (`Svc`) never
+    // change, and root wraps every screen in an animated `border` that compares equal once the
+    // entrance tween settles — so the top-down reconcile `can_skip_update`s this subtree and never
+    // re-renders the pair component off its *local* `use_state`. A button rebuilt only at mount
+    // would forever capture the empty mount-time PIN (pairing then fails as a "wrong PIN"). Mirror
+    // every keystroke into this stable ref instead, so the click reads exactly what was typed.
+    let live_pin = cx.use_ref(String::new());
     let target = ctx.shared.target.lock().unwrap().clone();
 
     let pair_btn = {
-        let (ctx2, ss, st, code2, target2) = (
+        let (ctx2, ss, st, live, target2) = (
             ctx.clone(),
             set_screen.clone(),
             set_status.clone(),
-            code.clone(),
+            live_pin.clone(),
             target.clone(),
         );
         button("Pair & Connect")
             .accent()
             .icon(Symbol::Accept)
             .on_click(move || {
-                let pin = code2.trim().to_string();
+                let pin = live.borrow().trim().to_string();
                 let (ctx3, ss, st, target3) =
                     (ctx2.clone(), ss.clone(), st.clone(), target2.clone());
                 std::thread::spawn(move || {
@@ -109,7 +116,16 @@ pub(crate) fn pair_page(props: &Svc, cx: &mut RenderCx) -> Element {
         text_box(code)
             .placeholder_text("PIN")
             .font_size(28.0)
-            .on_text_changed(move |s| set_code.call(s)),
+            .on_text_changed({
+                let live = live_pin.clone();
+                move |s: String| {
+                    // Record the live value for the click handler (the source of truth for the
+                    // PIN), and mirror it into `code` so the field stays correct if anything ever
+                    // does re-render this page (theme/DPI change).
+                    live.set(s.clone());
+                    set_code.call(s);
+                }
+            }),
         hstack((pair_btn, cancel_btn)).spacing(8.0),
         text_block(
             "Don\u{2019}t have a PIN? Request access instead and approve this device on the host \

crates/pf-client-core/src/session.rs

@@ -104,6 +104,81 @@ pub struct Stats {
 /// IDR (or a mid-GOP join) unfreezes almost immediately instead of never.
 const NO_OUTPUT_KEYFRAME_STREAK: u32 = 3;
 
+/// Longest the pump holds the last good frame waiting for a post-loss re-anchor keyframe before it
+/// gives up and resumes display. After a reference loss the hardware decoder does not error — it
+/// conceals the reference-missing deltas (on RADV, the DPB-and-output-COINCIDE path renders them as
+/// a gray plate with the new frame's motion painted over it) and returns Ok, so displaying them is
+/// the "gray frames mid-stream" artifact. We instead freeze on the last good picture until a fresh
+/// IDR re-anchors decode — the behaviour NVIDIA already shows (its DISTINCT output image + different
+/// concealment reads as a brief freeze, not gray). This cap only bounds the freeze when recovery
+/// genuinely stalls (host ignores the request, or an RFI recovery that never emits a keyframe), so a
+/// glitch can never become a permanent freeze. A recovery IDR round-trips well under this on any
+/// live link.
+const REANCHOR_FREEZE_MAX: Duration = Duration::from_millis(500);
+
+/// How many host intra-refresh recovery marks ([`USER_FLAG_RECOVERY_POINT`]) must arrive since the
+/// latest frame gap before the pump lifts its freeze on an IDR-free stream. TWO, not one: with a
+/// continuous rolling wave the host marks phase-fixed wave boundaries, so the FIRST boundary after a
+/// loss is only partially healed — stripes swept BEFORE the loss still reference the lost frame — and
+/// lifting there would flash a partially-stale picture. The SECOND boundary guarantees a full wave
+/// swept entirely after the loss, so the picture is clean. This stays correct under repeated loss
+/// because every new gap resets the count. The cost is up to ~2 wave periods of holding the last good
+/// frame — the deliberate "hold longer, never show garbage" trade.
+///
+/// [`USER_FLAG_RECOVERY_POINT`]: punktfunk_core::packet::USER_FLAG_RECOVERY_POINT
+const REANCHOR_MARKS_TO_LIFT: u32 = 2;
+
+/// Backstop patience while a host intra-refresh heal is visibly in progress. Each recovery mark
+/// pushes the freeze deadline out by this much, so a live mark stream (the host actively healing via
+/// its wave) keeps the client patiently holding the last good frame instead of tripping the IDR
+/// floor mid-heal. Must exceed the inter-mark interval (one wave period, ~0.5 s) with margin; if the
+/// marks STOP (heal stalled, or the host isn't running intra-refresh) the deadline lapses and the
+/// normal recovery-IDR floor fires, so a real stall still recovers.
+const RECOVERY_MARK_PATIENCE: Duration = Duration::from_millis(1500);
+
+/// Frames skipped when `got` arrives while `expected` was the next index, or `None` if `got` is
+/// contiguous (`== expected`) or a straggler we have already passed. Frame indices are u32 counters
+/// that wrap, so the "ahead" test is a wrapping subtraction split at the half-space: a small
+/// positive delta is a forward gap (missing frames whose dependents will decode against absent
+/// references); a delta in the top half is an index behind us.
+fn index_gap(expected: u32, got: u32) -> Option<u32> {
+    let ahead = got.wrapping_sub(expected);
+    (ahead != 0 && ahead < u32::MAX / 2).then_some(ahead)
+}
+
+/// Fold one decoded frame into the re-anchor state and decide whether it lifts the post-loss freeze.
+///
+/// `is_keyframe` — a real IDR (always a clean re-anchor). `has_anchor` — this AU carried
+/// [`USER_FLAG_RECOVERY_ANCHOR`](punktfunk_core::packet::USER_FLAG_RECOVERY_ANCHOR), the host's
+/// definitive single-frame re-anchor from an LTR-RFI recovery (a clean P-frame coded against a
+/// known-good reference), so it lifts on the FIRST occurrence exactly like an IDR — no two-mark wait.
+/// `has_mark` — this AU carried [`USER_FLAG_RECOVERY_POINT`](punktfunk_core::packet::USER_FLAG_RECOVERY_POINT),
+/// a host-signalled intra-refresh wave boundary (only *half* a re-anchor). `marks` — recovery marks
+/// seen since the latest gap.
+///
+/// Returns `(lift, new_marks)`: `lift` clears the freeze; `new_marks` is the running count (reset to 0
+/// on a lift). The two-mark rule ([`REANCHOR_MARKS_TO_LIFT`]) lives here so it is unit-tested
+/// independent of the pump's channel/decoder plumbing — the first wave boundary after a loss is only
+/// partially healed, so a single mark must NOT lift. An anchor (or IDR) is a *whole* re-anchor and
+/// lifts immediately.
+fn reanchor_after_frame(
+    is_keyframe: bool,
+    has_anchor: bool,
+    has_mark: bool,
+    marks: u32,
+) -> (bool, u32) {
+    let marks = if has_mark {
+        marks.saturating_add(1)
+    } else {
+        marks
+    };
+    if is_keyframe || has_anchor || marks >= REANCHOR_MARKS_TO_LIFT {
+        (true, 0)
+    } else {
+        (false, marks)
+    }
+}
+
 /// Frames the pump keeps waiting for their 0xCF host timing (pts → capture→received µs).
 /// ~2 s at 120 Hz — a timing arrives within a frame or two of its AU, and against an old
 /// host (no 0xCF at all) this just caps the dead-weight ring.
@@ -319,6 +394,20 @@ fn pump(
     // never drops, so the drop-count trigger below stays silent and the stream freezes
     // on the last good frame. A short streak forces a fresh IDR to re-anchor.
     let mut no_output_streak = 0u32;
+    // Freeze-until-reanchor: armed the moment we request a recovery keyframe (loss, decode error, or
+    // a no-output streak), it withholds the decoder's concealed frames from the presenter — which
+    // then redraws the last good picture — until a fresh keyframe re-anchors decode. See
+    // [`REANCHOR_FREEZE_MAX`] for why this exists and its backstop deadline.
+    let mut awaiting_reanchor = false;
+    let mut reanchor_deadline: Option<Instant> = None;
+    // Host intra-refresh recovery marks seen since the latest gap (see [`REANCHOR_MARKS_TO_LIFT`]).
+    // Reset to 0 whenever the freeze is (re-)armed, so a fresh loss always waits out two fresh marks.
+    let mut recovery_marks: u32 = 0;
+    // The frame_index we expect next (the host numbers frames consecutively). A jump means a frame
+    // went missing — the earliest, most reliable signal that the decoder is about to conceal, ~120 ms
+    // ahead of `frames_dropped` (the reassembler only declares a straggler lost once it ages out of
+    // the loss window, by which point the concealment already reached the screen).
+    let mut next_expected_index: Option<u32> = None;
 
     let end: Option<String> = loop {
         if stop.load(Ordering::SeqCst) {
@@ -334,9 +423,90 @@ fn pump(
                 // fps / goodput count every received AU (spec), decoded or not.
                 frames_n += 1;
                 bytes_n += frame.data.len() as u64;
+                // Reference-continuity gate: the host numbers frames consecutively, so a jump in
+                // frame_index means a frame is missing (lost, or an out-of-order straggler the
+                // reassembler emitted a newer frame ahead of) and this AU references a picture we
+                // never decoded. On RADV the decoder conceals that as a gray plate with the new
+                // motion on top — the reported artifact, and it shows most on high-motion frames (a
+                // full-screen pan bursts far more packets than a static desktop or a UFO-test's small
+                // moving sprite, so it is the frame that loses shards). Arm the freeze at the FIRST
+                // such frame — ~120 ms before `frames_dropped` would — so the gray never reaches the
+                // screen; recovery IDRs stay on the existing throttled path (see the arm below).
+                match next_expected_index {
+                    Some(exp) if frame.frame_index == exp => {
+                        next_expected_index = Some(exp.wrapping_add(1)); // contiguous
+                    }
+                    // A forward gap: hold the last good frame — but DO NOT ask for a keyframe here.
+                    // Hiding the concealment is free (the presenter redraws the last picture); an IDR
+                    // is not — at 4K120 it is a multi-megabyte frame and a visible stutter, and it can
+                    // re-trigger the very burst loss that caused this. The existing loss recovery below
+                    // (`frames_dropped`, host-coalesced + throttled) still requests it at exactly the
+                    // cadence it did before this change, so we add zero IDR pressure per pan. A
+                    // straggler behind us (`index_gap` → None) leaves the expectation put so the real
+                    // gap still trips.
+                    Some(exp) => {
+                        if let Some(gap) = index_gap(exp, frame.frame_index) {
+                            let now = Instant::now();
+                            awaiting_reanchor = true;
+                            recovery_marks = 0;
+                            reanchor_deadline = Some(now + REANCHOR_FREEZE_MAX);
+                            next_expected_index = Some(frame.frame_index.wrapping_add(1));
+                            // The gap carries the PRECISE lost range — [first missing, newest
+                            // received - 1] — so this is the one recovery signal that can drive true
+                            // reference-frame invalidation. Prefer an RFI request over a keyframe: an
+                            // RFI-capable host (AMD LTR / NVENC) re-references a known-good picture and
+                            // emits a clean P-frame tagged USER_FLAG_RECOVERY_ANCHOR (the freeze lifts
+                            // on ONE frame, no 20-40× IDR spike); an incapable/old host forces a
+                            // host-coalesced IDR instead, or ignores it (then the frames_dropped /
+                            // overdue keyframe paths below are the backstop). Throttled with those
+                            // paths (one recovery ask per 100 ms) so a burst of gaps — a full-screen
+                            // pan shedding shards — can't storm the control stream. This fires ~120 ms
+                            // before frames_dropped would, so recovery also starts sooner.
+                            if last_kf_req
+                                .is_none_or(|t| now.duration_since(t) >= Duration::from_millis(100))
+                            {
+                                last_kf_req = Some(now);
+                                let _ = connector
+                                    .request_rfi(exp, frame.frame_index.wrapping_sub(1));
+                            }
+                            tracing::trace!(gap, "frame gap — RFI recovery, holding last frame until re-anchor");
+                        }
+                    }
+                    None => next_expected_index = Some(frame.frame_index.wrapping_add(1)),
+                }
                 match decoder.decode(&frame.data) {
                     Ok(Some(image)) => {
                         no_output_streak = 0; // a decoded frame — the anchor holds
+                        // Host-signalled intra-refresh recovery mark: on an IDR-free intra-refresh
+                        // stream this wave-boundary flag is the only clean point the client can honor
+                        // (the decoder never flags the re-anchor — the coded frame stays `P`). A live
+                        // mark stream also means the host is actively healing, so push the backstop out
+                        // rather than trip a mid-heal IDR (see `RECOVERY_MARK_PATIENCE`).
+                        let has_mark =
+                            frame.flags & punktfunk_core::packet::USER_FLAG_RECOVERY_POINT != 0;
+                        // The host's definitive single-frame re-anchor: an LTR-RFI recovery frame (a
+                        // clean P-frame off a known-good reference), the AMD twin of an IDR re-anchor
+                        // but without the spike. It lifts on the FIRST occurrence.
+                        let has_anchor =
+                            frame.flags & punktfunk_core::packet::USER_FLAG_RECOVERY_ANCHOR != 0;
+                        if has_mark && awaiting_reanchor {
+                            reanchor_deadline = Some(Instant::now() + RECOVERY_MARK_PATIENCE);
+                        }
+                        // A fresh clean re-anchor lifts the freeze and shows this frame: a real intra
+                        // keyframe (IDR, always clean), an LTR-RFI recovery anchor (also whole), OR the
+                        // second recovery mark since the gap (the first wave boundary is only
+                        // half-healed — see `reanchor_after_frame`).
+                        let (lift, marks) = reanchor_after_frame(
+                            image.is_keyframe(),
+                            has_anchor,
+                            has_mark,
+                            recovery_marks,
+                        );
+                        recovery_marks = marks;
+                        if lift {
+                            awaiting_reanchor = false;
+                            reanchor_deadline = None;
+                        }
                         total_frames += 1;
                         dec_path = match &image {
                             DecodedImage::Cpu(_) => "software",
@@ -391,11 +561,20 @@ fn pump(
                             DecodedImage::VkFrame(v) => Some((v.timeline_sem, v.decode_done_value)),
                             _ => None,
                         };
-                        let _ = frame_tx.force_send(DecodedFrame {
-                            pts_ns: frame.pts_ns,
-                            decoded_ns,
-                            image,
-                        });
+                        if awaiting_reanchor {
+                            // Post-loss concealment: withhold this frame (it references a lost/gray
+                            // reference) so the presenter keeps redrawing the last good picture
+                            // rather than flashing the decoder's gray plate. Dropped here — the
+                            // hw-decode stat below still samples via `hw_fence` (raw handle + value,
+                            // valid past the guard). Cleared by the next keyframe or the backstop.
+                            tracing::trace!("holding last frame — awaiting post-loss re-anchor");
+                        } else {
+                            let _ = frame_tx.force_send(DecodedFrame {
+                                pts_ns: frame.pts_ns,
+                                decoded_ns,
+                                image,
+                            });
+                        }
                         // `decode` stage: received→decode COMPLETE, single clock.
                         match hw_fence {
                             Some((sem, value)) => {
@@ -424,6 +603,12 @@ fn pump(
                 // trip before asking again instead of flooding.
                 if no_output_streak >= NO_OUTPUT_KEYFRAME_STREAK {
                     let now = Instant::now();
+                    // Wedged on missing references: hold the last good frame until re-anchor
+                    // (armed even when the IDR request itself is throttled — the stream is broken
+                    // regardless of whether we ask again this iteration).
+                    awaiting_reanchor = true;
+                    recovery_marks = 0;
+                    reanchor_deadline = Some(now + REANCHOR_FREEZE_MAX);
                     if last_kf_req
                         .is_none_or(|t| now.duration_since(t) >= Duration::from_millis(100))
                     {
@@ -451,6 +636,9 @@ fn pump(
                 // through the same throttle as loss recovery below.
                 if decoder.take_keyframe_request() {
                     let now = Instant::now();
+                    awaiting_reanchor = true;
+                    recovery_marks = 0;
+                    reanchor_deadline = Some(now + REANCHOR_FREEZE_MAX);
                     if last_kf_req
                         .is_none_or(|t| now.duration_since(t) >= Duration::from_millis(100))
                     {
@@ -487,12 +675,33 @@ fn pump(
         if dropped > last_dropped {
             last_dropped = dropped;
             let now = Instant::now();
+            // A dropped AU means the frames after it reference a picture we never decoded — the
+            // decoder will conceal them (gray on RADV). Freeze on the last good frame until a fresh
+            // IDR re-anchors, so the concealment never reaches the screen.
+            awaiting_reanchor = true;
+            recovery_marks = 0;
+            reanchor_deadline = Some(now + REANCHOR_FREEZE_MAX);
             if last_kf_req.is_none_or(|t| now.duration_since(t) >= Duration::from_millis(100)) {
                 last_kf_req = Some(now);
                 let _ = connector.request_keyframe();
                 tracing::debug!(dropped, "requested keyframe (loss recovery)");
             }
         }
+        // Re-anchor overdue: the freeze has held the whole window with no keyframe — a lost recovery
+        // IDR, or a benign reorder that produced no `frames_dropped` and so requested none. Do NOT
+        // resume to gray (the one thing worse than a freeze): keep holding the last good frame and
+        // (re-)request a keyframe, throttled + host-coalesced, so a CLEAN re-anchor is what un-freezes
+        // us. A genuinely dead stream — host gone, link collapsed — is caught by the QUIC idle-timeout
+        // watchdog (returns to the menu), never by painting the decoder's concealment.
+        if awaiting_reanchor && reanchor_deadline.is_some_and(|d| Instant::now() >= d) {
+            let now = Instant::now();
+            reanchor_deadline = Some(now + REANCHOR_FREEZE_MAX);
+            if last_kf_req.is_none_or(|t| now.duration_since(t) >= Duration::from_millis(100)) {
+                last_kf_req = Some(now);
+                let _ = connector.request_keyframe();
+                tracing::debug!("re-anchor overdue — still holding, re-requesting keyframe");
+            }
+        }
 
         if window_start.elapsed() >= Duration::from_secs(1) {
             let secs = window_start.elapsed().as_secs_f32();
@@ -614,3 +823,107 @@ fn spawn_audio(
         .map_err(|e| tracing::warn!(error = %e, "audio thread failed to start — audio disabled"))
         .ok()
 }
+
+#[cfg(test)]
+mod tests {
+    use super::{index_gap, reanchor_after_frame, REANCHOR_MARKS_TO_LIFT};
+
+    // Simulate the pump's re-anchor state across a sequence of decoded frames: each `(is_keyframe,
+    // has_mark)` pair is folded through `reanchor_after_frame`, returning the frame index (0-based)
+    // at which the freeze first lifts, or `None` if it never does. `gap_before` reset points model a
+    // fresh loss re-arming the freeze (the pump zeroes the count at every gap/arm site).
+    fn lift_at(frames: &[(bool, bool)]) -> Option<usize> {
+        let mut marks = 0u32;
+        for (i, &(is_kf, has_mark)) in frames.iter().enumerate() {
+            // The intra-refresh-mark model never carries an LTR-RFI anchor (that path is exercised
+            // by `an_rfi_anchor_lifts_immediately`), so `has_anchor` is always false here.
+            let (lift, m) = reanchor_after_frame(is_kf, false, has_mark, marks);
+            marks = m;
+            if lift {
+                return Some(i);
+            }
+        }
+        None
+    }
+
+    #[test]
+    fn a_single_recovery_mark_does_not_lift() {
+        // The first wave boundary after a loss is only half-healed — one mark must hold the freeze.
+        assert_eq!(REANCHOR_MARKS_TO_LIFT, 2);
+        assert_eq!(lift_at(&[(false, true)]), None);
+        assert_eq!(lift_at(&[(false, false), (false, true), (false, false)]), None);
+    }
+
+    #[test]
+    fn the_second_recovery_mark_lifts() {
+        // Two marks = a full wave swept after the loss → clean re-anchor.
+        assert_eq!(lift_at(&[(false, true), (false, true)]), Some(1));
+        assert_eq!(
+            lift_at(&[(false, false), (false, true), (false, false), (false, true)]),
+            Some(3)
+        );
+    }
+
+    #[test]
+    fn a_real_keyframe_lifts_immediately() {
+        // An IDR is always a clean anchor — no marks needed.
+        assert_eq!(lift_at(&[(true, false)]), Some(0));
+        assert_eq!(lift_at(&[(false, true), (true, false)]), Some(1));
+    }
+
+    #[test]
+    fn a_fresh_gap_resets_the_mark_count() {
+        // The pump zeroes `recovery_marks` at each arm site, so one mark before a new gap plus one
+        // after must NOT lift — the model resets the running count to imitate that.
+        let mut marks = 0u32;
+        let (_, m) = reanchor_after_frame(false, false, true, marks); // mark #1 (pre-gap)
+        marks = m;
+        assert_eq!(marks, 1);
+        marks = 0; // a new gap re-arms the freeze → count reset
+        let (lift, m) = reanchor_after_frame(false, false, true, marks); // first mark of the new wave
+        assert!(!lift, "a single post-gap mark must not lift");
+        assert_eq!(m, 1);
+    }
+
+    #[test]
+    fn an_rfi_anchor_lifts_immediately() {
+        // An LTR-RFI recovery anchor is a WHOLE re-anchor (a clean P-frame off a known-good
+        // reference), so — like an IDR — it lifts on the FIRST occurrence, no two-mark wait.
+        let (lift, marks) = reanchor_after_frame(false, true, false, 0);
+        assert!(lift, "an RFI anchor must lift the freeze immediately");
+        assert_eq!(marks, 0, "a lift resets the running mark count");
+        // Even with zero prior marks and no keyframe, the anchor alone is sufficient.
+        let (lift, _) = reanchor_after_frame(false, true, true, 1);
+        assert!(lift, "an anchor lifts regardless of the pending mark count");
+    }
+
+    #[test]
+    fn contiguous_indices_are_not_a_gap() {
+        assert_eq!(index_gap(5, 5), None);
+        assert_eq!(index_gap(0, 0), None);
+    }
+
+    #[test]
+    fn a_forward_jump_reports_the_skip_count() {
+        assert_eq!(index_gap(5, 6), Some(1)); // one frame missing
+        assert_eq!(index_gap(5, 9), Some(4));
+    }
+
+    #[test]
+    fn a_straggler_behind_us_is_not_a_gap() {
+        // The reassembler emitted a newer frame first; the late one must not re-arm.
+        assert_eq!(index_gap(9, 5), None);
+        assert_eq!(index_gap(1, 0), None);
+    }
+
+    #[test]
+    fn the_index_counter_wraps_cleanly() {
+        // last frame = u32::MAX, so the next expected wraps to 0.
+        assert_eq!(index_gap(0, 0), None); // contiguous across the wrap
+        // waiting on u32::MAX, frame 0 arrived → MAX was skipped.
+        assert_eq!(index_gap(u32::MAX, 0), Some(1));
+        assert_eq!(index_gap(u32::MAX, 2), Some(3));
+        // an old frame arriving just after the wrap is still a straggler.
+        assert_eq!(index_gap(0, u32::MAX), None);
+    }
+}

crates/pf-client-core/src/video.rs

@@ -98,6 +98,10 @@ pub struct VkVideoFrame {
     pub width: u32,
     pub height: u32,
     pub color: ColorDesc,
+    /// Intra keyframe (IDR/I): the stream's re-anchor point. The pump resumes display on
+    /// one after suppressing the concealed frames a reference loss leaves in its wake (on
+    /// RADV a lost reference decodes to a gray plate with the new motion painted on top).
+    pub keyframe: bool,
     /// Keeps the cloned AVFrame (and through it the VkImage + frames context) alive
     /// until the presenter's fence proves the GPU reads done — same mechanism as the
     /// VAAPI path's DRM guard.
@@ -143,6 +147,44 @@ impl ColorDesc {
     }
 }
 
+/// True if the decoder tagged this frame as a full IDR keyframe — a guaranteed clean re-anchor
+/// after which the picture is loss-free, so the pump can lift a post-loss display freeze here.
+///
+/// Keys off `AV_FRAME_FLAG_KEY` (with `pict_type == I` as a belt for decoders that fill pict_type
+/// but not the flag). NOTE: FFmpeg's H.264/HEVC decode layer sets this flag **only for true IDR
+/// frames**, never for an *intra-refresh recovery point*. H.264 flags key only when a picture's
+/// `recovery_frame_cnt == 0` (a moving band uses `> 0`); HEVC clears the flag on every non-IRAP
+/// frame regardless of the recovery-point SEI. So an intra-refresh host (NVENC/AMF/QSV) heals the
+/// picture over N P-frames with no decoded frame ever flagged key — this function cannot detect
+/// that clean point, and the pump would freeze until the `REANCHOR_FREEZE_MAX` backstop (in
+/// `session.rs`) forces a real IDR. Detecting an intra-refresh re-anchor requires an out-of-band
+/// host wire signal on the AU that completes the wave; that is not yet plumbed.
+///
+/// # Safety
+/// `frame` must point to a valid `AVFrame` alive for the duration of the call.
+pub unsafe fn frame_is_keyframe(frame: *const ffmpeg::ffi::AVFrame) -> bool {
+    // SAFETY: caller guarantees a live AVFrame; plain field reads.
+    unsafe {
+        ((*frame).flags & ffmpeg::ffi::AV_FRAME_FLAG_KEY) != 0
+            || (*frame).pict_type == ffmpeg::ffi::AVPictureType::AV_PICTURE_TYPE_I
+    }
+}
+
+impl DecodedImage {
+    /// Whether the frame is an intra keyframe — see [`frame_is_keyframe`]. The pump uses
+    /// this as the stream's re-anchor signal after a loss.
+    pub fn is_keyframe(&self) -> bool {
+        match self {
+            DecodedImage::Cpu(f) => f.keyframe,
+            #[cfg(target_os = "linux")]
+            DecodedImage::Dmabuf(f) => f.keyframe,
+            DecodedImage::VkFrame(f) => f.keyframe,
+            #[cfg(windows)]
+            DecodedImage::D3d11(f) => f.keyframe,
+        }
+    }
+}
+
 /// The Y′CbCr→RGB conversion as three vec4 rows for a shader constant buffer / push-constant
 /// block: `rgb[i] = dot(r[i].xyz, yuv) + r[i].w` — bit-depth exact. The ONE coefficient
 /// implementation every presenter derives its CSC from (Vulkan push constants, the Windows
@@ -205,6 +247,8 @@ pub struct CpuFrame {
     /// pixels are full-range RGB), but a PQ/BT.2020 stream keeps its transfer + primaries
     /// baked in — the presenter tags the texture so GTK tone-maps it.
     pub color: ColorDesc,
+    /// Intra keyframe (IDR/I) — the pump's post-loss re-anchor signal. See [`VkVideoFrame`].
+    pub keyframe: bool,
 }
 
 /// A decoded frame still on the GPU: dmabuf fds + plane layout for
@@ -222,6 +266,8 @@ pub struct DmabufFrame {
     /// Signaling of the source frame — drives the `GdkDmabufTexture` color state (BT.709
     /// narrow for SDR, BT.2020 PQ for an HDR stream).
     pub color: ColorDesc,
+    /// Intra keyframe (IDR/I) — the pump's post-loss re-anchor signal. See [`VkVideoFrame`].
+    pub keyframe: bool,
     pub guard: DrmFrameGuard,
 }
 
@@ -644,6 +690,9 @@ impl SoftwareDecoder {
             stride: dst_linesize[0] as usize,
             rgba,
             color,
+            // `is_key()` reads the same intra flag `frame_is_keyframe` derives from pict_type
+            // for the hardware paths; ffmpeg-next handles the FFmpeg-version binding split.
+            keyframe: frame.is_key(),
         })
     }
 }
@@ -844,6 +893,7 @@ impl VaapiDecoder {
                 // SAFETY: `self.frame` is the live decoded AVFrame (unref'd only after
                 // this returns); plain CICP field reads.
                 color: ColorDesc::from_raw(self.frame),
+                keyframe: frame_is_keyframe(self.frame),
                 guard,
             })
         }
@@ -1363,6 +1413,7 @@ impl VulkanDecoder {
                 width: (*self.frame).width as u32,
                 height: (*self.frame).height as u32,
                 color: ColorDesc::from_raw(self.frame),
+                keyframe: frame_is_keyframe(self.frame),
                 guard: DrmFrameGuard(clone),
             })
         }

crates/pf-client-core/src/video_d3d11.rs

@@ -99,6 +99,9 @@ pub struct D3d11Frame {
     /// BT.709 full-range RGB — regardless of the stream's own CICP (a PQ stream was
     /// tone-mapped). The presenter keys SDR/HDR handling off this, so it always reads SDR.
     pub color: ColorDesc,
+    /// Intra keyframe (IDR/I) — the pump's post-loss re-anchor signal. See
+    /// `crate::video::VkVideoFrame`.
+    pub keyframe: bool,
     /// The ring slot's NT shared handle (`IDXGIResource1::CreateSharedHandle`), stable for the
     /// ring's lifetime. Raw `isize` so the frame crosses the pump→presenter channel.
     pub handle: isize,
@@ -692,6 +695,8 @@ impl D3d11vaDecoder {
                     matrix: 0,    // identity — RGB
                     full_range: true,
                 },
+                // SAFETY: `self.frame` is the live decoded AVFrame for this call.
+                keyframe: crate::video::frame_is_keyframe(self.frame),
                 handle,
                 generation,
             })

crates/punktfunk-core/src/client.rs

@@ -19,7 +19,8 @@ use crate::packet::FLAG_PROBE;
 use crate::quic::{
     accept_resync, endpoint, io, wall_clock_ns, window_loss_ppm, BitrateChanged, ClockEcho,
     ClockResync, ColorInfo, HdrMeta, Hello, HidOutput, LossReport, ProbeRequest, ProbeResult,
-    Reconfigure, Reconfigured, RequestKeyframe, ResyncStep, RichInput, SetBitrate, Start, Welcome,
+    Reconfigure, Reconfigured, RequestKeyframe, RfiRequest, ResyncStep, RichInput, SetBitrate, Start,
+    Welcome,
 };
 use crate::session::{Frame, Session};
 use crate::transport::UdpTransport;
@@ -49,6 +50,10 @@ enum CtrlRequest {
     Mode(Mode),
     Probe(ProbeRequest),
     Keyframe,
+    /// Reference-frame-invalidation recovery: the client saw a `frame_index` gap and reports the
+    /// invalidation range so an RFI-capable host re-references a known-good picture instead of
+    /// forcing a full IDR. See [`RfiRequest`].
+    Rfi(RfiRequest),
     Loss(LossReport),
     /// Adaptive bitrate: ask the host to re-target its encoder (kbps). Sent by the pump's
     /// [`BitrateController`] when the user's bitrate setting is Automatic.
@@ -868,6 +873,24 @@ impl NativeClient {
             .map_err(|_| PunktfunkError::Closed)
     }
 
+    /// Ask the host to recover from loss by **reference-frame invalidation** rather than a full IDR:
+    /// the client reports the range `[first_frame, last_frame]` of access units it can no longer trust
+    /// (from the first missing `frame_index` through the newest received). An RFI-capable host
+    /// re-references a known-good picture before `first_frame` (AMD LTR / NVENC RFI) and emits a clean
+    /// P-frame tagged [`crate::packet::USER_FLAG_RECOVERY_ANCHOR`]; a host that can't RFI forces an IDR
+    /// instead (same as [`request_keyframe`](Self::request_keyframe)). Non-blocking, fire-and-forget —
+    /// the recovered frame is the only ack; throttle it like the keyframe request. Prefer this over
+    /// `request_keyframe` on loss so AMD/RFI hosts avoid the IDR spike; the keyframe request remains
+    /// the backstop when the recovery frame itself is lost.
+    pub fn request_rfi(&self, first_frame: u32, last_frame: u32) -> Result<()> {
+        self.ctrl_tx
+            .try_send(CtrlRequest::Rfi(RfiRequest {
+                first_frame,
+                last_frame,
+            }))
+            .map_err(|_| PunktfunkError::Closed)
+    }
+
     /// Cumulative access units the host→client reassembler dropped as unrecoverable (FEC couldn't
     /// rebuild them). A video loop polls this and calls [`request_keyframe`](Self::request_keyframe)
     /// when it increases — the correct loss trigger under infinite GOP, where unrecoverable loss
@@ -1511,6 +1534,7 @@ async fn worker_main(args: WorkerArgs) {
                             CtrlRequest::Mode(m) => Reconfigure { mode: m }.encode(),
                             CtrlRequest::Probe(p) => p.encode(),
                             CtrlRequest::Keyframe => RequestKeyframe.encode(),
+                            CtrlRequest::Rfi(r) => r.encode(),
                             CtrlRequest::Loss(r) => r.encode(),
                             CtrlRequest::SetBitrate(k) => SetBitrate { bitrate_kbps: k }.encode(),
                             CtrlRequest::ClockResync => {

crates/punktfunk-core/src/packet.rs

@@ -35,6 +35,25 @@ pub const FLAG_SOF: u8 = 0x4;
 /// feeding them to the decoder. Punktfunk/1 only (GameStream never sets it).
 pub const FLAG_PROBE: u8 = 0x8;
 
+/// Application `user_flags` bit (the u32 [`PacketHeader::user_flags`] word, surfaced to the client
+/// as [`crate::session::Frame::flags`]) — NOT a transport packet flag. Marks the access unit that
+/// **completes an intra-refresh wave**: the picture is loss-free from here even though the frame is
+/// a coded `P` (no IDR, so the decoder never sets `AV_FRAME_FLAG_KEY`). The client lifts its
+/// post-loss display freeze on this bit as well as on a real keyframe — the only bitstream-invisible
+/// clean point it can honor without forcing a full IDR. Lives above the low nibble because the host
+/// reuses `FLAG_PIC`/`FLAG_SOF`/`FLAG_PROBE` bit values inside `user_flags`; `0x10` clears all four.
+pub const USER_FLAG_RECOVERY_POINT: u32 = 0x10;
+
+/// Application `user_flags` bit — a **definitive single-frame clean re-anchor**. Unlike
+/// [`USER_FLAG_RECOVERY_POINT`] (an intra-refresh wave boundary, where the first boundary after a loss
+/// is only half-healed so the client waits for the second), this marks an access unit the host coded
+/// to reference a **known-good** picture on purpose — an AMD **LTR reference-frame-invalidation**
+/// recovery frame (`ForceLTRReferenceBitfield`): a clean P-frame off a long-term reference the client
+/// already has, not an IDR. The picture is loss-free the instant this AU decodes, so the client lifts
+/// its post-loss freeze on the **first** such mark. Coded `P` (no IDR), so the decoder never sets
+/// `AV_FRAME_FLAG_KEY` — this host flag is the only signal.
+pub const USER_FLAG_RECOVERY_ANCHOR: u32 = 0x20;
+
 /// Crypto framing overhead [`Session`](crate::session::Session) adds when encrypting:
 /// an 8-byte sequence prefix plus the GCM tag.
 pub const CRYPTO_OVERHEAD: usize = 8 + crate::crypto::TAG_LEN;

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

@@ -355,6 +355,24 @@ pub struct Reconfigured {
 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
 pub struct RequestKeyframe;
 
+/// `client → host`: reference-frame-invalidation recovery — the loss-aware sibling of
+/// [`RequestKeyframe`]. The client detected a `frame_index` gap and reports the range `[first_frame,
+/// last_frame]` of access units it can no longer trust (from the first missing index through the
+/// newest received). Instead of a full IDR (a 20-40× spike that deepens the loss it recovers), a host
+/// whose encoder supports RFI re-references a known-good picture *before* `first_frame` — an AMD LTR
+/// force-reference or an NVENC `nvEncInvalidateRefFrames` — emitting a single clean P-frame it tags
+/// [`crate::packet::USER_FLAG_RECOVERY_ANCHOR`] so the client lifts its freeze on it. A host that
+/// can't RFI (no valid reference / libavcodec backend) forces an IDR instead, exactly as for a bare
+/// [`RequestKeyframe`]; a host that predates this ignores the unknown message and the client's
+/// keyframe backstop still recovers. Fire-and-forget — the recovered frame is the only ack.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct RfiRequest {
+    /// First access-unit `frame_index` the client can no longer trust (the gap start).
+    pub first_frame: u32,
+    /// Newest received `frame_index` at the time of the report (the invalidation range end).
+    pub last_frame: u32,
+}
+
 /// `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
@@ -467,6 +485,8 @@ pub const MSG_LOSS_REPORT: u8 = 0x04;
 pub const MSG_SET_BITRATE: u8 = 0x05;
 /// Type byte of [`BitrateChanged`].
 pub const MSG_BITRATE_CHANGED: u8 = 0x06;
+/// Type byte of [`RfiRequest`].
+pub const MSG_RFI_REQUEST: u8 = 0x07;
 /// Type byte of [`ProbeRequest`].
 pub const MSG_PROBE_REQUEST: u8 = 0x20;
 /// Type byte of [`ProbeResult`].
@@ -1032,6 +1052,28 @@ impl RequestKeyframe {
     }
 }
 
+impl RfiRequest {
+    pub fn encode(&self) -> Vec<u8> {
+        // magic[0..4] type[4] first_frame[5..9] last_frame[9..13]
+        let mut b = Vec::with_capacity(13);
+        b.extend_from_slice(CTL_MAGIC);
+        b.push(MSG_RFI_REQUEST);
+        b.extend_from_slice(&self.first_frame.to_le_bytes());
+        b.extend_from_slice(&self.last_frame.to_le_bytes());
+        b
+    }
+
+    pub fn decode(b: &[u8]) -> Result<RfiRequest> {
+        if b.len() != 13 || &b[0..4] != CTL_MAGIC || b[4] != MSG_RFI_REQUEST {
+            return Err(PunktfunkError::InvalidArg("bad RfiRequest"));
+        }
+        Ok(RfiRequest {
+            first_frame: u32::from_le_bytes(b[5..9].try_into().unwrap()),
+            last_frame: u32::from_le_bytes(b[9..13].try_into().unwrap()),
+        })
+    }
+}
+
 impl LossReport {
     pub fn encode(&self) -> Vec<u8> {
         // magic[0..4] type[4] loss_ppm[5..9]

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

@@ -633,6 +633,35 @@ fn request_keyframe_roundtrip() {
     assert!(RequestKeyframe::decode(&[bytes.as_slice(), &[0]].concat()).is_err());
 }
 
+#[test]
+fn rfi_request_roundtrip() {
+    for (first_frame, last_frame) in [(0u32, 0u32), (40, 47), (5, 5), (1_000_000, u32::MAX)] {
+        let r = RfiRequest {
+            first_frame,
+            last_frame,
+        };
+        assert_eq!(RfiRequest::decode(&r.encode()).unwrap(), r);
+    }
+    // Disjoint from the bare keyframe request (its loss-unaware sibling) and others: type byte + length.
+    assert!(RfiRequest::decode(&RequestKeyframe.encode()).is_err());
+    assert!(RequestKeyframe::decode(
+        &RfiRequest {
+            first_frame: 1,
+            last_frame: 2
+        }
+        .encode()
+    )
+    .is_err());
+    // Exact length — no trailing bytes.
+    let bytes = RfiRequest {
+        first_frame: 3,
+        last_frame: 9,
+    }
+    .encode();
+    assert!(RfiRequest::decode(&[bytes.as_slice(), &[0]].concat()).is_err());
+    assert!(RfiRequest::decode(&bytes[..bytes.len() - 1]).is_err());
+}
+
 #[test]
 fn loss_report_roundtrip() {
     for loss_ppm in [0u32, 1, 12_345, 50_000, 1_000_000] {

crates/punktfunk-host/src/capture.rs

@@ -466,5 +466,8 @@ pub mod dxgi;
 #[cfg(target_os = "windows")]
 #[path = "capture/windows/idd_push.rs"]
 pub mod idd_push;
+#[cfg(target_os = "windows")]
+#[path = "capture/windows/synthetic_nv12.rs"]
+pub mod synthetic_nv12;
 #[cfg(target_os = "linux")]
 mod linux;

crates/punktfunk-host/src/capture/windows/synthetic_nv12.rs

@@ -0,0 +1,181 @@
+//! A headless synthetic **NV12 D3D11** capture source for exercising the GPU encoders on Windows
+//! without a real capture session.
+//!
+//! The native AMF path (and the D3D11 zero-copy NVENC/QSV paths) require an NV12 texture that lives
+//! on the GPU — the CPU-Bgrx [`SyntheticCapturer`](crate::capture::SyntheticCapturer) can't provide
+//! one, and DXGI Desktop Duplication can't create one under an ssh session-0 (E_ACCESSDENIED). This
+//! source builds an NV12 texture on the selected render adapter and fills it with a **moving** luma
+//! ramp each frame, so the encoder sees genuine motion (P-frame residuals + the intra-refresh wave
+//! under content change) — exactly what an intra-refresh recovery validation needs. Driven by
+//! `spike --source synthetic-nv12`.
+
+use crate::capture::dxgi::{make_device, D3d11Frame};
+use crate::capture::{CapturedFrame, Capturer, FramePayload, PixelFormat};
+use anyhow::{Context, Result};
+use windows::Win32::Graphics::Direct3D11::{
+    ID3D11Device, ID3D11DeviceContext, ID3D11Texture2D, D3D11_BIND_SHADER_RESOURCE,
+    D3D11_CPU_ACCESS_WRITE, D3D11_MAPPED_SUBRESOURCE, D3D11_MAP_WRITE, D3D11_TEXTURE2D_DESC,
+    D3D11_USAGE, D3D11_USAGE_DEFAULT, D3D11_USAGE_STAGING,
+};
+use windows::Win32::Graphics::Dxgi::Common::{DXGI_FORMAT_NV12, DXGI_SAMPLE_DESC};
+use windows::Win32::Graphics::Dxgi::{CreateDXGIFactory1, IDXGIAdapter1, IDXGIFactory4};
+
+/// Synthetic NV12 frames on the GPU. Owns its own D3D11 device + immediate context and two NV12
+/// textures: a CPU-writable STAGING scratch it fills each frame, and a DEFAULT texture it copies
+/// into and hands to the encoder. The encoder copies out of the DEFAULT texture synchronously
+/// (spike drives capture→submit→poll on one thread), so reusing one DEFAULT texture is sound.
+pub struct SyntheticNv12Capturer {
+    device: ID3D11Device,
+    context: ID3D11DeviceContext,
+    default_tex: ID3D11Texture2D,
+    staging: ID3D11Texture2D,
+    width: u32,
+    height: u32,
+    fps: u32,
+    frame_idx: u64,
+}
+
+// SAFETY: mirrors `D3d11Frame`'s reasoning — the device is created free-threaded (`make_device`
+// passes no `SINGLETHREADED` flag) and D3D11 uses interlocked COM refcounting, so moving the whole
+// capturer (device + immediate context + textures) to its owning thread and using it only there is
+// sound. The value is moved, never aliased (no `Sync`), so the single-threaded immediate context is
+// never touched concurrently.
+unsafe impl Send for SyntheticNv12Capturer {}
+
+impl SyntheticNv12Capturer {
+    pub fn new(width: u32, height: u32, fps: u32) -> Result<Self> {
+        // NV12 is 4:2:0 — both dimensions must be even (the chroma plane is width/2 × height/2).
+        let width = (width & !1).max(2);
+        let height = (height & !1).max(2);
+        // SAFETY: a self-contained builder owning every handle it creates; each COM call is checked
+        // and the returned owners drop with their wrappers.
+        unsafe {
+            let adapter = resolve_render_adapter().context("resolve render adapter for NV12 source")?;
+            let (device, context) = make_device(&adapter).context("create D3D11 device")?;
+            let default_tex = create_nv12(
+                &device,
+                width,
+                height,
+                D3D11_USAGE_DEFAULT,
+                0,
+                D3D11_BIND_SHADER_RESOURCE.0 as u32,
+            )
+            .context("create NV12 default texture")?;
+            let staging = create_nv12(
+                &device,
+                width,
+                height,
+                D3D11_USAGE_STAGING,
+                D3D11_CPU_ACCESS_WRITE.0 as u32,
+                0,
+            )
+            .context("create NV12 staging texture")?;
+            Ok(SyntheticNv12Capturer {
+                device,
+                context,
+                default_tex,
+                staging,
+                width,
+                height,
+                fps,
+                frame_idx: 0,
+            })
+        }
+    }
+}
+
+impl Capturer for SyntheticNv12Capturer {
+    fn next_frame(&mut self) -> Result<CapturedFrame> {
+        let pts_ns = self.frame_idx * 1_000_000_000 / self.fps.max(1) as u64;
+        // SAFETY: Map/Unmap/CopyResource on this capturer's own single-threaded immediate context;
+        // all writes stay within the mapped NV12 surface (Y: H rows of RowPitch; UV: H/2 rows of
+        // RowPitch beginning at RowPitch*H — the standard NV12 plane layout).
+        unsafe {
+            let mut map = D3D11_MAPPED_SUBRESOURCE::default();
+            self.context
+                .Map(&self.staging, 0, D3D11_MAP_WRITE, 0, Some(&mut map))
+                .context("Map(NV12 staging)")?;
+            let pitch = map.RowPitch as usize;
+            let base = map.pData as *mut u8;
+            // A diagonal luma ramp that shifts 4 codes/frame — strong, deterministic motion.
+            let shift = (self.frame_idx as u32).wrapping_mul(4);
+            for y in 0..self.height {
+                let row = base.add(y as usize * pitch);
+                for x in 0..self.width {
+                    *row.add(x as usize) = x.wrapping_add(y).wrapping_add(shift) as u8;
+                }
+            }
+            // UV plane (neutral gray = 128) at offset RowPitch*H: H/2 rows, `width` bytes each
+            // (width/2 interleaved Cb,Cr pairs).
+            let uv = base.add(pitch * self.height as usize);
+            for r in 0..(self.height / 2) {
+                let row = uv.add(r as usize * pitch);
+                for c in 0..self.width {
+                    *row.add(c as usize) = 128;
+                }
+            }
+            self.context.Unmap(&self.staging, 0);
+            self.context.CopyResource(&self.default_tex, &self.staging);
+        }
+        self.frame_idx += 1;
+        Ok(CapturedFrame {
+            width: self.width,
+            height: self.height,
+            pts_ns,
+            format: PixelFormat::Nv12,
+            payload: FramePayload::D3d11(D3d11Frame {
+                texture: self.default_tex.clone(),
+                device: self.device.clone(),
+            }),
+        })
+    }
+}
+
+/// Resolve the same render adapter the encoder will pick (`PUNKTFUNK_RENDER_ADAPTER` / preference /
+/// max-VRAM LUID), falling back to adapter 0.
+///
+/// # Safety
+/// Calls DXGI factory/adapter enumeration; returns owned COM objects or an error.
+unsafe fn resolve_render_adapter() -> Result<IDXGIAdapter1> {
+    let factory: IDXGIFactory4 = CreateDXGIFactory1().context("CreateDXGIFactory1")?;
+    if let Some(luid) = crate::win_adapter::resolve_render_adapter_luid() {
+        if let Ok(a) = factory.EnumAdapterByLuid::<IDXGIAdapter1>(luid) {
+            return Ok(a);
+        }
+    }
+    factory.EnumAdapters1(0).context("EnumAdapters1(0)")
+}
+
+/// Create an NV12 `Texture2D` with the given usage/CPU-access/bind flags.
+///
+/// # Safety
+/// `device` must be a live D3D11 device; the returned texture is owned by the caller.
+unsafe fn create_nv12(
+    device: &ID3D11Device,
+    width: u32,
+    height: u32,
+    usage: D3D11_USAGE,
+    cpu_access: u32,
+    bind: u32,
+) -> Result<ID3D11Texture2D> {
+    let desc = D3D11_TEXTURE2D_DESC {
+        Width: width,
+        Height: height,
+        MipLevels: 1,
+        ArraySize: 1,
+        Format: DXGI_FORMAT_NV12,
+        SampleDesc: DXGI_SAMPLE_DESC {
+            Count: 1,
+            Quality: 0,
+        },
+        Usage: usage,
+        BindFlags: bind,
+        CPUAccessFlags: cpu_access,
+        ..Default::default()
+    };
+    let mut tex: Option<ID3D11Texture2D> = None;
+    device
+        .CreateTexture2D(&desc, None, Some(&mut tex))
+        .context("CreateTexture2D(NV12)")?;
+    tex.context("CreateTexture2D returned a null NV12 texture")
+}

crates/punktfunk-host/src/encode.rs

@@ -19,6 +19,13 @@ pub struct EncodedFrame {
     pub pts_ns: u64,
     /// True for IDR/keyframes (sets the SOF/keyframe wire flags).
     pub keyframe: bool,
+    /// True when this AU is a **reference-frame-invalidation recovery frame** — a clean P-frame the
+    /// encoder coded against a known-good reference in response to
+    /// [`invalidate_ref_frames`](Encoder::invalidate_ref_frames) (AMD LTR force-reference). The pump
+    /// tags it [`punktfunk_core::packet::USER_FLAG_RECOVERY_ANCHOR`] so the client lifts its post-loss
+    /// freeze on it without an IDR. Only the native-AMF LTR path sets it; every other backend leaves
+    /// it `false` (their RFI, when present, re-references transparently with no distinct clean-point AU).
+    pub recovery_anchor: bool,
 }
 
 /// Codec selection negotiated with the client.
@@ -208,12 +215,28 @@ pub struct EncoderCaps {
     /// the encoder's real chroma disagrees with what was negotiated (the in-band SPS is authoritative
     /// for the decoder either way).
     pub chroma_444: bool,
-    /// The encoder runs a periodic **intra-refresh wave** (a moving band of intra blocks +
-    /// recovery-point SEI, no periodic IDR): FEC-unrecoverable loss self-heals within one wave, so
-    /// the session glue rate-limits client keyframe requests instead of answering each with a full
-    /// IDR (the 20-40× frame-size spike that cascades under loss). Linux NVENC sets it when
-    /// `PUNKTFUNK_INTRA_REFRESH` opened the encoder in that mode; VAAPI/software never do.
+    /// The encoder runs a periodic **intra-refresh wave** — a moving band of intra blocks that
+    /// re-codes the whole picture over ~0.5 s, no periodic IDR. FEC-unrecoverable loss self-heals as
+    /// the band sweeps, so the session glue rate-limits client keyframe requests instead of answering
+    /// each with a full IDR (the 20-40× frame-size spike that cascades under loss). Linux NVENC / AMF
+    /// set it when `PUNKTFUNK_INTRA_REFRESH` opened the encoder in that mode; VAAPI/QSV/software never
+    /// do. NOTE — the wave carries NO decoder-visible clean-point: FFmpeg never sets `AV_FRAME_FLAG_KEY`
+    /// at a recovery point (H.264 flags key only when `recovery_frame_cnt == 0`; HEVC only on IRAP),
+    /// and AMF emits no recovery-point SEI at all. So this cap ALONE does not let the client lift its
+    /// post-loss freeze without an IDR — that needs [`intra_refresh_recovery`](Self::intra_refresh_recovery).
     pub intra_refresh: bool,
+    /// The intra-refresh wave is a *validated constrained GDR* — verified on real hardware to fully
+    /// heal a lost picture within one wave period with no residual artifacts. Only then does the host
+    /// tag each wave-boundary AU with [`USER_FLAG_RECOVERY_POINT`](punktfunk_core::packet::USER_FLAG_RECOVERY_POINT),
+    /// so the client can lift its freeze on the second mark (a proven clean re-anchor) instead of
+    /// waiting out its backstop and forcing a full IDR. Default `false` on every backend until on-glass
+    /// validation flips it — an un-validated encoder keeps the IDR recovery path, so this is inert and
+    /// cannot regress. Meaningless unless [`intra_refresh`](Self::intra_refresh) is also set.
+    pub intra_refresh_recovery: bool,
+    /// Length of the intra-refresh wave in frames — the boundary period the host marks on (it sets
+    /// `USER_FLAG_RECOVERY_POINT` on every Nth emitted AU, re-phased at each IDR). 0 when intra-refresh
+    /// is off. Only consulted when [`intra_refresh_recovery`](Self::intra_refresh_recovery) is set.
+    pub intra_refresh_period: u32,
 }
 
 /// A hardware encoder. One per session; runs on the encode thread.

crates/punktfunk-host/src/encode/linux/mod.rs

@@ -177,6 +177,10 @@ pub struct NvencEncoder {
     /// Opened in intra-refresh mode (surfaced via [`caps`](Encoder::caps) so the session glue
     /// rate-limits forced IDRs — the wave heals loss without them).
     intra_refresh: bool,
+    /// Resolved wave length in frames when [`intra_refresh`](Self::intra_refresh), else 0. Cached at
+    /// open so the pump's per-AU `caps()` doesn't re-read `PUNKTFUNK_IR_PERIOD_FRAMES`; the pump marks
+    /// every Nth AU with `USER_FLAG_RECOVERY_POINT` for the client's clean re-anchor.
+    intra_refresh_period: u32,
 }
 
 // `CudaHw` holds raw `AVBufferRef`s and `sws_444` a raw `SwsContext`; the encoder lives on a single
@@ -525,6 +529,11 @@ impl NvencEncoder {
             frame_idx: 0,
             force_kf: false,
             intra_refresh,
+            intra_refresh_period: if intra_refresh {
+                intra_refresh_period(fps).max(1) as u32
+            } else {
+                0
+            },
         })
     }
 }
@@ -536,6 +545,12 @@ impl Encoder for NvencEncoder {
             // convert. RFI/HDR-SEI stay unsupported on libavcodec NVENC (the trait defaults).
             chroma_444: self.want_444,
             intra_refresh: self.intra_refresh,
+            // NVENC intra-refresh is purpose-built GDR loss recovery (moving band + recovery-point
+            // SEI): the wave heals a lost picture within one period, so mark the boundary AUs and let
+            // the client re-anchor on them instead of forcing a full IDR. Tied to `intra_refresh`
+            // (already the `PUNKTFUNK_INTRA_REFRESH` opt-in), unlike AMF/QSV which stay unvalidated.
+            intra_refresh_recovery: self.intra_refresh,
+            intra_refresh_period: self.intra_refresh_period,
             ..super::EncoderCaps::default()
         }
     }
@@ -578,6 +593,7 @@ impl Encoder for NvencEncoder {
                     data,
                     pts_ns,
                     keyframe: pkt.is_key(),
+                    recovery_anchor: false,
                 }))
             }
             // No packet ready yet (need another input frame).

crates/punktfunk-host/src/encode/linux/vaapi.rs

@@ -294,6 +294,7 @@ fn poll_encoder(enc: &mut encoder::video::Encoder, fps: u32) -> Result<Option<En
                 data,
                 pts_ns: pts * 1_000_000_000 / fps as u64,
                 keyframe: pkt.is_key(),
+                recovery_anchor: false,
             }))
         }
         Err(ffmpeg::Error::Other { errno })

crates/punktfunk-host/src/encode/sw.rs

@@ -211,6 +211,7 @@ impl Encoder for OpenH264Encoder {
                 data,
                 pts_ns,
                 keyframe,
+                recovery_anchor: false,
             });
         }
         self.frame_idx += 1;

crates/punktfunk-host/src/encode/windows/amf.rs

@@ -644,6 +644,26 @@ struct CodecProps {
     /// HEVC 64-px CTBs. `None` on AV1 (v1.4.36 exposes only a mode enum, no slot-size control —
     /// loss recovery stays IDR there).
     intra_refresh: Option<(PCWSTR, u32)>,
+    /// LTR-RFI recovery property names (design: the AMD twin of NVENC intra-refresh recovery).
+    /// `None` on AV1 — its reference management uses a frame-marking OBU mechanism this path does
+    /// not drive, so LTR recovery is AVC/HEVC-only.
+    ltr: Option<LtrProps>,
+}
+
+/// The four AMF LTR (long-term-reference) property names, codec-prefixed (AVC bare, HEVC `Hevc*`).
+/// Two are static (`max_*`, set once at open); two are per-frame (`mark`/`force`, set on the input
+/// surface each `submit`). Together they let a loss re-reference a known-good older frame — a clean
+/// P-frame instead of a 20–40× IDR spike.
+struct LtrProps {
+    /// `MaxOfLTRFrames` — number of user LTR slots (we request [`NUM_LTR_SLOTS`]).
+    max_ltr_frames: PCWSTR,
+    /// `MaxNumRefFrames` — reference-picture budget; must exceed 1 for LTR to engage.
+    max_num_ref_frames: PCWSTR,
+    /// `MarkCurrentWithLTRIndex` (per-frame) — tag the current frame as long-term reference slot N.
+    mark_ltr_index: PCWSTR,
+    /// `ForceLTRReferenceBitfield` (per-frame) — force the current frame to reference only the LTR
+    /// slots in the bitfield (`1<<N`), breaking the corrupted short-term chain after a loss.
+    force_ltr_bitfield: PCWSTR,
 }
 
 /// The two payload shapes `lowlatency` takes across codecs.
@@ -689,6 +709,12 @@ fn codec_props(codec: Codec) -> CodecProps {
             out_primaries: w!("OutColorPrimaries"),
             hdr_metadata: None,
             intra_refresh: Some((w!("IntraRefreshMBsNumberPerSlot"), 16)),
+            ltr: Some(LtrProps {
+                max_ltr_frames: w!("MaxOfLTRFrames"),
+                max_num_ref_frames: w!("MaxNumRefFrames"),
+                mark_ltr_index: w!("MarkCurrentWithLTRIndex"),
+                force_ltr_bitfield: w!("ForceLTRReferenceBitfield"),
+            }),
         },
         Codec::H265 => CodecProps {
             component: w!("AMFVideoEncoderHW_HEVC"),
@@ -716,6 +742,12 @@ fn codec_props(codec: Codec) -> CodecProps {
             out_primaries: w!("HevcOutColorPrimaries"),
             hdr_metadata: Some(w!("HevcInHDRMetadata")),
             intra_refresh: Some((w!("HevcIntraRefreshCTBsNumberPerSlot"), 64)),
+            ltr: Some(LtrProps {
+                max_ltr_frames: w!("HevcMaxOfLTRFrames"),
+                max_num_ref_frames: w!("HevcMaxNumRefFrames"),
+                mark_ltr_index: w!("HevcMarkCurrentWithLTRIndex"),
+                force_ltr_bitfield: w!("HevcForceLTRReferenceBitfield"),
+            }),
         },
         Codec::Av1 => CodecProps {
             component: w!("AMFVideoEncoderHW_AV1"),
@@ -743,6 +775,7 @@ fn codec_props(codec: Codec) -> CodecProps {
             out_primaries: w!("Av1OutputColorPrimaries"),
             hdr_metadata: Some(w!("Av1InHDRMetadata")),
             intra_refresh: None,
+            ltr: None,
         },
     }
 }
@@ -797,6 +830,45 @@ fn intra_refresh_period(fps: u32) -> u32 {
         .unwrap_or_else(|| (fps.max(16) / 2).max(2))
 }
 
+/// Number of user-controlled LTR slots. AMD exposes up to 2; two rotating slots hold a sliding pair
+/// of recent long-term references, so a loss can re-reference the newest one *before* the loss point.
+const NUM_LTR_SLOTS: usize = 2;
+
+/// AMD's real clean loss-recovery path (the NVENC-RFI twin): the encoder marks frames as long-term
+/// references, and on loss forces a later frame to re-reference a known-good one — a clean P-frame,
+/// not a 20-40× IDR spike. On by default when the driver supports it (AMF intra-refresh cannot heal —
+/// no constrained-intra-prediction property exists in the API, header-confirmed + PSNR-proven — and
+/// LTR is mutually exclusive with it, so LTR wins). `PUNKTFUNK_NO_AMF_LTR=1` forces the old full-IDR
+/// recovery for debugging.
+fn ltr_disabled() -> bool {
+    std::env::var("PUNKTFUNK_NO_AMF_LTR")
+        .map(|v| matches!(v.trim(), "1" | "true" | "yes" | "on"))
+        .unwrap_or(false)
+}
+
+/// Cadence (frames) between LTR marks — a fresh long-term reference roughly every half second by
+/// default (`PUNKTFUNK_LTR_INTERVAL_FRAMES` overrides). With [`NUM_LTR_SLOTS`] slots this keeps ~one
+/// second of recent references, so a loss up to ~1 s old still has a known-good frame to force; a
+/// smaller interval means the forced reference is more recent (a smaller recovery-frame residual).
+fn ltr_mark_interval(fps: u32) -> i64 {
+    std::env::var("PUNKTFUNK_LTR_INTERVAL_FRAMES")
+        .ok()
+        .and_then(|s| s.parse::<i64>().ok())
+        .filter(|v| *v >= 1)
+        .unwrap_or_else(|| (fps.max(2) / 2).max(1) as i64)
+}
+
+/// Validation hook (`PUNKTFUNK_LTR_FORCE_AT=N`, spike-only): at `frame_idx == N` the encoder
+/// self-triggers its real [`invalidate_ref_frames`](Encoder::invalidate_ref_frames) path, so a
+/// headless spike run can exercise LTR recovery end-to-end (mark → force → recovery-anchor tag)
+/// without a live client sending an [`RfiRequest`](punktfunk_core::quic::RfiRequest). `None` normally.
+fn ltr_test_force_at() -> Option<i64> {
+    std::env::var("PUNKTFUNK_LTR_FORCE_AT")
+        .ok()
+        .and_then(|s| s.parse::<i64>().ok())
+        .filter(|v| *v > 0)
+}
+
 // ---------------------------------------------------------------------------------------------
 // Owned-pointer guards (release exactly once; Terminate before Release for context/component,
 // mirroring amfenc.c's teardown order).
@@ -930,11 +1002,12 @@ struct Inner {
     dctx: ID3D11DeviceContext,
     ring: Vec<ID3D11Texture2D>,
     next: usize,
-    /// (pts_ns, forced-IDR) per submitted-but-unretrieved frame, FIFO — the AMF encoder emits
-    /// AUs in submit order (B-frames are never enabled), pairing with `QueryOutput`. Its length is
-    /// the count of input surfaces AMF still holds, so `submit` bounds it below [`RING`] to keep
-    /// the input ring from being overwritten under it.
-    pending: VecDeque<(u64, bool)>,
+    /// (pts_ns, forced-IDR, recovery-anchor) per submitted-but-unretrieved frame, FIFO — the AMF
+    /// encoder emits AUs in submit order (B-frames are never enabled), pairing with `QueryOutput`.
+    /// The third field tags the LTR-RFI re-anchor frame so the AU carries `recovery_anchor` for the
+    /// client's freeze-lift. Its length is the count of input surfaces AMF still holds, so `submit`
+    /// bounds it below [`RING`] to keep the input ring from being overwritten under it.
+    pending: VecDeque<(u64, bool, bool)>,
     /// AUs already pulled by `submit`'s backpressure drain, waiting to be handed out by `poll`
     /// (FIFO, strictly older than anything still in `pending`). Empty in the steady state — only
     /// fills when the encoder falls behind and `submit` drains to free an input slot.
@@ -988,6 +1061,26 @@ pub struct AmfEncoder {
     /// gates [`EncoderCaps::intra_refresh`] so keyframe-request rate-limiting only happens when
     /// the wave really runs.
     ir_active: bool,
+    // --- Long-Term-Reference reference-frame-invalidation recovery (the AMD RFI path) ---
+    /// The driver accepted the LTR properties at open — gates [`EncoderCaps::supports_rfi`] and all
+    /// the per-frame LTR marking/forcing below. When true, intra-refresh is NOT set (mutually
+    /// exclusive) and loss recovery re-references a known-good LTR instead of forcing a full IDR.
+    ltr_active: bool,
+    /// The `frame_idx` currently stored in each of the two LTR slots (`None` = never marked). On loss
+    /// the newest slot with an index *before* the loss is the known-good reference to force.
+    ltr_slots: [Option<i64>; NUM_LTR_SLOTS],
+    /// The slot the next LTR mark writes (round-robins `0,1,0,1,…` so the two slots hold a sliding
+    /// pair of recent references).
+    next_ltr_slot: usize,
+    /// Cadence (frames) between LTR marks — a fresh long-term reference roughly this often.
+    ltr_mark_interval: i64,
+    /// Set by [`invalidate_ref_frames`](Encoder::invalidate_ref_frames): the LTR slot the *next*
+    /// submitted frame must force-reference (`ForceLTRReferenceBitfield`). Consumed on that submit.
+    pending_force: Option<usize>,
+    /// Validation hook (`PUNKTFUNK_LTR_FORCE_AT=N`, spike-only): at `frame_idx == N`, self-trigger the
+    /// real [`invalidate_ref_frames`](Encoder::invalidate_ref_frames) path so a headless spike run can
+    /// exercise LTR recovery end-to-end without a live client. `None` in normal operation.
+    ltr_test_force_at: Option<i64>,
     /// Consecutive [`reset`](Self::reset)s that have NOT been followed by a produced AU (cleared in
     /// `poll` on any output). An in-place `Terminate`+re-`Init` heals a transient component stall,
     /// but it re-inits the SAME context — so if the fault is the context / VCN session itself (the
@@ -1084,17 +1177,33 @@ impl AmfEncoder {
             force_kf: false,
             hdr_meta: None,
             ir_active: false,
+            ltr_active: false,
+            ltr_slots: [None; NUM_LTR_SLOTS],
+            next_ltr_slot: 0,
+            ltr_mark_interval: ltr_mark_interval(fps),
+            pending_force: None,
+            ltr_test_force_at: ltr_test_force_at(),
             resets_without_output: 0,
         })
     }
 
+    /// Whether this encoder should *attempt* the LTR-RFI recovery path (design: the AMD twin of
+    /// NVENC intra-refresh recovery). Gated to AVC/HEVC — AMF exposes user LTR only for those two
+    /// codecs — and defeatable via `PUNKTFUNK_NO_AMF_LTR`. Whether the driver actually *accepts* the
+    /// properties is a separate question answered by [`apply_static_props`], which sets `ltr_active`.
+    fn ltr_wanted(&self) -> bool {
+        !ltr_disabled() && matches!(self.codec, Codec::H264 | Codec::H265)
+    }
+
     /// Apply the static encoder configuration (design §3.4 — the native mirror of the ffmpeg
     /// opts block in `open_win_encoder`). Called before `Init`, and again on a `reset()`
     /// re-`Init` (Terminate does not guarantee property retention across every driver).
-    /// Returns whether the intra-refresh wave was requested AND accepted by this driver — the
-    /// caller stores it so [`Encoder::caps`] only rate-limits keyframe requests when the wave
-    /// really runs.
-    unsafe fn apply_static_props(&self, comp: *mut sys::AmfComponent) -> Result<bool> {
+    /// Returns `(ir_active, ltr_active)`: whether the intra-refresh wave / the LTR-RFI slots were
+    /// requested AND accepted by this driver. The two are mutually exclusive (LTR wins when both are
+    /// wanted). The caller stores both — `ir_active` so [`Encoder::caps`] only rate-limits keyframe
+    /// requests when a wave runs, `ltr_active` so [`Encoder::caps`] advertises `supports_rfi` and the
+    /// per-frame mark/force logic in `submit` only fires when the slots exist.
+    unsafe fn apply_static_props(&self, comp: *mut sys::AmfComponent) -> Result<(bool, bool)> {
         let p = &self.props;
         // Usage first: it "fully configures parameter set" — everything after is an override.
         set_prop(
@@ -1145,7 +1254,39 @@ impl AmfEncoder {
         // whole picture refreshes every `period` frames — per-slot units = ceil(total blocks /
         // period). Optional by VCN generation; the return value gates `caps().intra_refresh`.
         let mut ir_active = false;
-        if let Some((name, block)) = p.intra_refresh {
+        let mut ltr_active = false;
+        if let Some(ltr) = p.ltr.as_ref().filter(|_| self.ltr_wanted()) {
+            // LTR-RFI recovery (design: the AMD twin of NVENC intra-refresh recovery). Request
+            // NUM_LTR_SLOTS user-controlled long-term references. LTR needs >1 reference frames and
+            // is MUTUALLY EXCLUSIVE with intra-refresh (AMF disables one if both are set), so the
+            // intra-refresh block below is skipped whenever LTR engages.
+            let ref_ok = set_prop(
+                comp,
+                ltr.max_num_ref_frames,
+                AmfVariant::from_i64(NUM_LTR_SLOTS as i64),
+                false,
+            )?;
+            let ltr_ok = set_prop(
+                comp,
+                ltr.max_ltr_frames,
+                AmfVariant::from_i64(NUM_LTR_SLOTS as i64),
+                false,
+            )?;
+            ltr_active = ref_ok && ltr_ok;
+            if ltr_active {
+                tracing::info!(
+                    slots = NUM_LTR_SLOTS,
+                    mark_interval = self.ltr_mark_interval,
+                    "AMF LTR-RFI recovery enabled (loss recovery re-references a known-good LTR, not a full IDR)"
+                );
+            } else {
+                tracing::warn!(
+                    ref_ok,
+                    ltr_ok,
+                    "this VCN/driver rejected an LTR property — loss recovery stays full-IDR"
+                );
+            }
+        } else if let Some((name, block)) = p.intra_refresh {
             if intra_refresh_requested() {
                 let period = intra_refresh_period(self.fps);
                 let blocks = self.width.div_ceil(block) * self.height.div_ceil(block);
@@ -1273,7 +1414,7 @@ impl AmfEncoder {
             AmfVariant::from_i64(primaries),
             self.ten_bit,
         )?;
-        Ok(ir_active)
+        Ok((ir_active, ltr_active))
     }
 
     /// Build (or rebuild, on a capture-device change) the AMF context + encoder component on the
@@ -1323,7 +1464,7 @@ impl AmfEncoder {
                 bail!("AMF CreateComponent returned null");
             }
             let comp = Component(comp);
-            let ir_active = self.apply_static_props(comp.0)?;
+            let (ir_active, ltr_active) = self.apply_static_props(comp.0)?;
             let fmt = if self.ten_bit {
                 sys::AMF_SURFACE_P010
             } else {
@@ -1334,6 +1475,14 @@ impl AmfEncoder {
                 "AMF encoder Init",
             )?;
             self.ir_active = ir_active;
+            // A rebuilt component starts with fresh (empty) LTR slots — a new context has no
+            // reference history, so any prior marks are void and the first frame re-IDRs anyway.
+            self.ltr_active = ltr_active;
+            if ltr_active {
+                self.ltr_slots = [None; NUM_LTR_SLOTS];
+                self.next_ltr_slot = 0;
+                self.pending_force = None;
+            }
 
             // Owned input ring on the capturer's device (design §3.2): RENDER_TARGET |
             // SHADER_RESOURCE, the same bind flags the validated ffmpeg zero-copy pool uses.
@@ -1594,7 +1743,7 @@ enum DrainOutcome {
 /// single encode thread with no other AMF call to this component in flight.
 unsafe fn drain_one_output(
     comp: *mut sys::AmfComponent,
-    pending: &mut VecDeque<(u64, bool)>,
+    pending: &mut VecDeque<(u64, bool, bool)>,
     output_data_type: PCWSTR,
     output_key_max: i64,
 ) -> Result<DrainOutcome> {
@@ -1641,11 +1790,12 @@ unsafe fn drain_one_output(
         bail!("AMF output buffer is empty");
     }
     let au = std::slice::from_raw_parts(native as *const u8, size).to_vec();
-    let (pts_ns, forced) = pending.pop_front().unwrap_or((0, false));
+    let (pts_ns, forced, recovery_anchor) = pending.pop_front().unwrap_or((0, false, false));
     Ok(DrainOutcome::Frame(EncodedFrame {
         data: au,
         pts_ns,
         keyframe: key_prop || forced,
+        recovery_anchor,
     }))
 }
 
@@ -1689,9 +1839,57 @@ impl Encoder for AmfEncoder {
             expected
         );
         self.ensure_inner(&frame.device)?;
+        let cur_idx = self.frame_idx;
         let forced = std::mem::take(&mut self.force_kf) || self.frame_idx == 0;
         let pts_100ns = self.frame_idx * 10_000_000 / self.fps.max(1) as i64;
         self.frame_idx += 1;
+        // --- LTR-RFI per-frame decisions (design: the AMD twin of NVENC intra-refresh recovery) ---
+        // Decided here, before borrowing `inner`, because the test hook re-enters `&mut self`
+        // (`invalidate_ref_frames`) and the mark cadence mutates the slot bookkeeping. The two
+        // per-frame property names are copied out (PCWSTR is Copy) so the unsafe surface block can
+        // set them without re-borrowing `self.props` under the live `inner` borrow.
+        let ltr_names = self
+            .props
+            .ltr
+            .as_ref()
+            .map(|l| (l.mark_ltr_index, l.force_ltr_bitfield));
+        let mut mark_slot: Option<usize> = None;
+        let mut force_slot: Option<usize> = None;
+        let mut recovery_anchor = false;
+        if self.ltr_active {
+            if forced {
+                // An IDR resets the decoder's reference buffers — every prior LTR mark is void.
+                // Re-anchor from scratch: drop the stale slots (the mark cadence below tags the IDR
+                // as the first fresh long-term reference) and cancel any force queued against them.
+                self.ltr_slots = [None; NUM_LTR_SLOTS];
+                self.next_ltr_slot = 0;
+                self.pending_force = None;
+            } else if self.ltr_test_force_at == Some(cur_idx) {
+                // Spike-only validation hook: self-trigger the real invalidate path so a headless
+                // run exercises mark → force → recovery-anchor without a live client's RfiRequest.
+                let triggered = self.invalidate_ref_frames(cur_idx, cur_idx);
+                tracing::info!(
+                    frame = cur_idx,
+                    triggered,
+                    "AMF LTR test hook fired invalidate_ref_frames"
+                );
+            }
+            // Apply a queued force (from invalidate_ref_frames / the test hook) to THIS frame: it
+            // becomes the clean re-anchor P-frame the client lifts its post-loss freeze on.
+            if let Some(slot) = self.pending_force.take() {
+                force_slot = Some(slot);
+                recovery_anchor = true;
+            }
+            // Mark cadence: refresh a long-term reference on every IDR and every `ltr_mark_interval`
+            // frames — but never on the recovery frame itself (marking rotates `next_ltr_slot` and
+            // could overwrite the very slot being forced; the next cadence mark re-establishes it).
+            if force_slot.is_none() && (forced || cur_idx % self.ltr_mark_interval == 0) {
+                let slot = self.next_ltr_slot;
+                self.ltr_slots[slot] = Some(cur_idx);
+                self.next_ltr_slot = (self.next_ltr_slot + 1) % NUM_LTR_SLOTS;
+                mark_slot = Some(slot);
+            }
+        }
         let inner = self.inner.as_mut().expect("ensure_inner succeeded");
         // Push the HDR mastering metadata when it changed (or a rebuilt component lost it) — a
         // dynamic property, so mid-stream regrades take effect on the next IDR. Best-effort: a
@@ -1831,6 +2029,47 @@ impl Encoder for AmfEncoder {
                     Codec::Av1 => {}
                 }
             }
+            // LTR-RFI per-frame properties (design: the AMD twin of NVENC intra-refresh recovery).
+            // `mark_slot`/`force_slot` were decided above. Marking tags the current frame as a
+            // long-term reference; forcing makes it re-reference a known-good LTR — a clean P-frame
+            // that breaks the corrupted short-term chain after a loss, no 20-40× IDR. Best-effort:
+            // a rejecting driver just leaves the client on its keyframe-request fallback.
+            if let Some((mark_name, force_name)) = ltr_names {
+                if let Some(slot) = mark_slot {
+                    let r = ((*(*surf.0).vtbl).set_property)(
+                        surf.0,
+                        mark_name.0,
+                        AmfVariant::from_i64(slot as i64),
+                    );
+                    if r != sys::AMF_OK {
+                        tracing::warn!(
+                            slot,
+                            result = %format!("{} ({r})", result_name(r)),
+                            "AMF LTR mark rejected"
+                        );
+                    }
+                }
+                if let Some(slot) = force_slot {
+                    let r = ((*(*surf.0).vtbl).set_property)(
+                        surf.0,
+                        force_name.0,
+                        AmfVariant::from_i64(1_i64 << slot),
+                    );
+                    if r == sys::AMF_OK {
+                        tracing::info!(
+                            slot,
+                            frame = cur_idx,
+                            "AMF LTR-RFI: re-referencing known-good LTR (clean recovery, no IDR)"
+                        );
+                    } else {
+                        tracing::warn!(
+                            slot,
+                            result = %format!("{} ({r})", result_name(r)),
+                            "AMF LTR force-reference rejected — client stays frozen until its IDR fallback"
+                        );
+                    }
+                }
+            }
             let mut r = ((*(*inner.comp.0).vtbl).submit_input)(inner.comp.0, surf.0);
             // Backstop back-pressure: the in-flight bound above already keeps a slot free, but if
             // AMF's own input queue is momentarily full, AMF_INPUT_FULL is "busy, drain me and
@@ -1873,7 +2112,7 @@ impl Encoder for AmfEncoder {
                 }
             }
         }
-        inner.pending.push_back((captured.pts_ns, forced));
+        inner.pending.push_back((captured.pts_ns, forced, recovery_anchor));
         Ok(())
     }
 
@@ -1887,11 +2126,63 @@ impl Encoder for AmfEncoder {
         self.hdr_meta = meta;
     }
 
+    /// LTR-RFI recovery (the AMD twin of the Windows NVENC `nvEncInvalidateRefFrames` path): a loss
+    /// of client frames `[first, last]` is answered by forcing the *next* submitted frame to
+    /// re-reference the newest long-term reference marked *before* the loss — a clean P-frame the
+    /// client can decode against a picture it still holds, instead of a 20-40× IDR spike.
+    ///
+    /// Returns `true` when a usable pre-loss LTR exists (so the caller must NOT also force an IDR);
+    /// `false` when the loss predates every live LTR — then the only correct recovery is a keyframe,
+    /// and the caller falls back to [`request_keyframe`](Self::request_keyframe). Runs on the encode
+    /// thread (like submit/poll); the force is applied on the next `submit`.
+    fn invalidate_ref_frames(&mut self, first: i64, last: i64) -> bool {
+        // No live LTR session (driver declined the slots, or AV1 which has no user-LTR path) or a
+        // nonsense range → caller forces a full IDR.
+        if !self.ltr_active || first < 0 || first > last {
+            return false;
+        }
+        // Pick the newest LTR strictly OLDER than the loss: the most recent known-good reference the
+        // client still holds, so re-referencing it costs the least (smallest recovery-frame residual).
+        // Frame numbers are 1:1 with the client's (both count submissions in order — see the NVENC
+        // path), so `ltr_slots` (which store `frame_idx`) compare directly against `first`.
+        let mut best: Option<(usize, i64)> = None;
+        for (slot, marked) in self.ltr_slots.iter().enumerate() {
+            if let Some(idx) = *marked {
+                if idx < first && best.is_none_or(|(_, b)| idx > b) {
+                    best = Some((slot, idx));
+                }
+            }
+        }
+        match best {
+            Some((slot, ltr_frame)) => {
+                // Queue the force for the next submit; that frame ships tagged `recovery_anchor`.
+                self.pending_force = Some(slot);
+                tracing::info!(
+                    first,
+                    last,
+                    slot,
+                    ltr_frame,
+                    "AMF LTR-RFI: forcing the next frame to re-reference a known-good LTR (no IDR)"
+                );
+                true
+            }
+            None => {
+                tracing::info!(
+                    first,
+                    last,
+                    "AMF LTR-RFI: no live LTR older than the loss — falling back to IDR recovery"
+                );
+                false
+            }
+        }
+    }
+
     fn caps(&self) -> EncoderCaps {
         EncoderCaps {
-            // AMF has no NVENC-style reference invalidation — the intra-refresh wave is the
-            // loss-recovery substitute; without it every unrecoverable loss costs an IDR.
-            supports_rfi: false,
+            // LTR-RFI: AMD's reference invalidation is the user long-term-reference path (mark a
+            // frame, force a later one to re-reference it). True only when the live driver accepted
+            // the LTR slots at open — otherwise loss recovery falls back to a full IDR.
+            supports_rfi: self.ltr_active,
             // In-band mastering/CLL via `*InHDRMetadata` (HEVC SEI / AV1 metadata OBU); AVC has
             // no such property (and no HDR sessions negotiate H.264).
             supports_hdr_metadata: self.ten_bit && self.props.hdr_metadata.is_some(),
@@ -1901,6 +2192,11 @@ impl Encoder for AmfEncoder {
             // accepted the property (queried per loss event, so the post-first-frame value is
             // what the session glue's IDR rate-limiting sees).
             intra_refresh: self.ir_active,
+            // Not yet: the AMD VCN wave heals in principle, but its constrained-GDR
+            // heal-within-a-period is unvalidated on-glass and AMF emits no recovery-point SEI, so
+            // the host keeps the IDR recovery path. Flip both once verified on real hardware.
+            intra_refresh_recovery: false,
+            intra_refresh_period: 0,
         }
     }
 
@@ -1992,6 +2288,7 @@ impl Encoder for AmfEncoder {
             self.inner = None;
             self.bound_device = 0;
             self.ir_active = false;
+            self.ltr_active = false;
             return true;
         }
         let inner = self
@@ -2016,8 +2313,14 @@ impl Encoder for AmfEncoder {
                 sys::AMF_SURFACE_NV12
             };
             match self.apply_static_props(comp) {
-                Ok(ir) => {
+                Ok((ir, ltr)) => {
                     self.ir_active = ir;
+                    // Re-Init voids the reference history: the rebuilt stream restarts at IDR with
+                    // empty LTR slots, so any prior marks are stale and must be dropped.
+                    self.ltr_active = ltr;
+                    self.ltr_slots = [None; NUM_LTR_SLOTS];
+                    self.next_ltr_slot = 0;
+                    self.pending_force = None;
                     ((*(*comp).vtbl).init)(comp, fmt, self.width as i32, self.height as i32)
                         == sys::AMF_OK
                 }
@@ -2030,6 +2333,7 @@ impl Encoder for AmfEncoder {
             );
         } else {
             self.ir_active = false;
+            self.ltr_active = false;
             // Full teardown; the next submit reopens context + component on the current device.
             tracing::warn!("AMF in-place re-Init failed — full context teardown, reopening lazily");
             self.inner = None;

crates/punktfunk-host/src/encode/windows/ffmpeg_win.rs

@@ -339,6 +339,7 @@ fn poll_encoder(enc: &mut encoder::video::Encoder, fps: u32) -> Result<PollOutco
                 data,
                 pts_ns: pts * 1_000_000_000 / fps as u64,
                 keyframe: pkt.is_key(),
+                recovery_anchor: false,
             }))
         }
         Err(ffmpeg::Error::Other { errno })

crates/punktfunk-host/src/encode/windows/nvenc.rs

@@ -1157,6 +1157,7 @@ impl NvencD3d11Encoder {
                 data,
                 pts_ns,
                 keyframe,
+                recovery_anchor: false,
             });
         Ok(())
     }
@@ -1424,6 +1425,8 @@ impl Encoder for NvencD3d11Encoder {
             // The direct-NVENC path recovers via real RFI (or a forced IDR), not the Linux
             // libavcodec intra-refresh mode.
             intra_refresh: false,
+            intra_refresh_recovery: false,
+            intra_refresh_period: 0,
         }
     }
 
@@ -1542,6 +1545,7 @@ impl Encoder for NvencD3d11Encoder {
                 data,
                 pts_ns,
                 keyframe,
+                recovery_anchor: false,
             }))
         }
     }

crates/punktfunk-host/src/main.rs

@@ -696,10 +696,14 @@ fn parse_spike(args: &[String]) -> Result<Options> {
             "--source" => {
                 source = match next()?.as_str() {
                     "synthetic" => Source::Synthetic,
+                    "synthetic-nv12" => Source::SyntheticNv12,
                     "portal" => Source::Portal,
                     "kwin-virtual" => Source::KwinVirtual,
                     other => {
-                        bail!("unknown --source '{other}' (synthetic|portal|kwin-virtual)")
+                        bail!(
+                            "unknown --source '{other}' \
+                             (synthetic|synthetic-nv12|portal|kwin-virtual)"
+                        )
                     }
                 }
             }

crates/punktfunk-host/src/pipeline.rs

@@ -26,12 +26,16 @@ pub fn pump_once(
         data,
         pts_ns,
         keyframe,
+        recovery_anchor,
     }) = encoder.poll()?
     {
         let mut flags = FLAG_PIC as u32;
         if keyframe {
             flags |= FLAG_SOF as u32;
         }
+        if recovery_anchor {
+            flags |= punktfunk_core::packet::USER_FLAG_RECOVERY_ANCHOR;
+        }
         // core does FEC + packetize + pace + send.
         session.submit_frame(&data, pts_ns, flags)?;
     }

crates/punktfunk-host/src/punktfunk1.rs

@@ -34,7 +34,7 @@ use punktfunk_core::packet::{FLAG_PIC, FLAG_PROBE, FLAG_SOF};
 use punktfunk_core::quic::{
     endpoint, io, BitrateChanged, ClockEcho, ClockProbe, ColorInfo, Hello, LossReport,
     PairChallenge, PairProof, PairRequest, PairResult, ProbeRequest, ProbeResult, Reconfigure,
-    Reconfigured, RequestKeyframe, SetBitrate, Start, Welcome,
+    Reconfigured, RequestKeyframe, RfiRequest, SetBitrate, Start, Welcome,
 };
 use punktfunk_core::transport::UdpTransport;
 use punktfunk_core::Session;
@@ -1124,6 +1124,10 @@ async fn serve_session(
     // (inbound requests, outbound probe results) are multiplexed with `select!`.
     let (reconfig_tx, reconfig_rx) = std::sync::mpsc::channel::<punktfunk_core::Mode>();
     let (keyframe_tx, keyframe_rx) = std::sync::mpsc::channel::<()>();
+    // Client LTR-RFI recovery: the control task forwards each `RfiRequest`'s lost-frame range here;
+    // the encode loop prefers `Encoder::invalidate_ref_frames` (a clean re-anchor P-frame) over a
+    // full IDR when the encoder supports it (native-AMF LTR / Windows NVENC).
+    let (rfi_tx, rfi_rx) = std::sync::mpsc::channel::<(u32, u32)>();
     let (bitrate_tx, bitrate_rx) = std::sync::mpsc::channel::<u32>();
     let (probe_tx, probe_rx) = std::sync::mpsc::channel::<ProbeRequest>();
     let (probe_result_tx, mut probe_result_rx) =
@@ -1199,6 +1203,19 @@ async fn serve_session(
                         if keyframe_tx.send(()).is_err() {
                             break; // data plane gone
                         }
+                    } else if let Ok(req) = RfiRequest::decode(&msg) {
+                        // Client LTR-RFI recovery: it lost the frame range `[first, last]` and asks
+                        // the encoder to re-reference a known-good older frame instead of paying for
+                        // a full IDR. The encode loop attempts `invalidate_ref_frames`, falling back
+                        // to a coalesced keyframe when the encoder can't (range too old / no RFI).
+                        tracing::debug!(
+                            first = req.first_frame,
+                            last = req.last_frame,
+                            "client requested reference-frame invalidation (loss recovery)"
+                        );
+                        if rfi_tx.send((req.first_frame, req.last_frame)).is_err() {
+                            break; // data plane gone
+                        }
                     } else if let Ok(rep) = LossReport::decode(&msg) {
                         // Adaptive FEC: size recovery to the loss the client is seeing. The data-plane
                         // send loop reads `fec_target_ctl` and applies it per frame. Ignored when FEC
@@ -1590,6 +1607,7 @@ async fn serve_session(
                         quit: quit_stream,
                         reconfig: reconfig_rx,
                         keyframe: keyframe_rx,
+                        rfi: rfi_rx,
                         bitrate_rx,
                         compositor,
                         bitrate_kbps,
@@ -2396,6 +2414,29 @@ fn audio_thread(
     tracing::warn!("punktfunk/1 audio requires Linux or Windows — session continues without it");
 }
 
+/// Advance the intra-refresh wave position and decide whether this emitted AU is a wave boundary
+/// that should carry [`USER_FLAG_RECOVERY_POINT`](punktfunk_core::packet::USER_FLAG_RECOVERY_POINT).
+///
+/// `ir_wave_pos` counts frames since the last IDR/wave start; a real IDR re-phases it to 0 (an IDR
+/// restarts the encoder's wave AND is itself a clean anchor, so it is never additionally marked).
+/// Every `period`-th non-IDR AU is a boundary — the client lifts its post-loss freeze on the SECOND
+/// such mark. Pure so the marking cadence is unit-tested without a GPU (see the pump's use in the
+/// encode-poll loop).
+fn mark_recovery_boundary(ir_wave_pos: &mut u32, is_keyframe: bool, period: u32) -> bool {
+    if is_keyframe {
+        *ir_wave_pos = 0;
+        false
+    } else {
+        *ir_wave_pos += 1;
+        if *ir_wave_pos >= period {
+            *ir_wave_pos = 0;
+            true
+        } else {
+            false
+        }
+    }
+}
+
 fn synthetic_stream(
     session: &mut Session,
     frames: u32,
@@ -3396,6 +3437,9 @@ struct SessionContext {
     reconfig: std::sync::mpsc::Receiver<punktfunk_core::Mode>,
     /// Client decode-recovery keyframe requests.
     keyframe: std::sync::mpsc::Receiver<()>,
+    /// Client LTR-RFI recovery requests — the lost-frame range `(first, last)`. The encode loop
+    /// prefers `Encoder::invalidate_ref_frames` over a full IDR when the encoder supports it.
+    rfi: std::sync::mpsc::Receiver<(u32, u32)>,
     /// 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.
     bitrate_rx: std::sync::mpsc::Receiver<u32>,
@@ -3467,6 +3511,7 @@ fn virtual_stream(ctx: SessionContext) -> Result<()> {
         quit,
         reconfig,
         keyframe,
+        rfi,
         bitrate_rx,
         compositor,
         mut bitrate_kbps,
@@ -3684,6 +3729,11 @@ fn virtual_stream(ctx: SessionContext) -> Result<()> {
     // Self-diagnosis for the periodic-stutter class: warns when the served recovery IDRs settle
     // into a stable multi-second rhythm (see [`crate::metronome::Metronome`]).
     let mut recovery_cadence = crate::metronome::Metronome::new();
+    // Position within the current intra-refresh wave (frames since the last IDR/wave start). Only
+    // meaningful on a `caps().intra_refresh_recovery` encoder; the pump tags every wave-boundary AU
+    // with `USER_FLAG_RECOVERY_POINT` so the client can lift its post-loss freeze on a clean
+    // re-anchor without a full IDR. Re-phased to 0 at each emitted IDR (which restarts the wave).
+    let mut ir_wave_pos: u32 = 0;
     // Per-stage latency breakdown (PUNKTFUNK_PERF): per-call µs for the GPU-bound stages so we see
     // exactly where the capture→encoded latency goes — cap=try_latest (ring read + colour convert),
     // submit=encode_picture launch, wait=lock_bitstream (the scheduling wait + ASIC encode, the one
@@ -3900,6 +3950,33 @@ fn virtual_stream(ctx: SessionContext) -> Result<()> {
         while keyframe.try_recv().is_ok() {
             want_kf = true;
         }
+        // Client LTR-RFI recovery: prefer re-referencing a known-good older frame (a clean recovery
+        // P-frame — no 20-40× IDR spike) over a full keyframe when the encoder supports it (native
+        // AMF LTR / Windows NVENC). Drain the backlog (the client re-requests until the recovery
+        // frame lands) coalesced to the widest lost range. Attempt the invalidate only when a full
+        // IDR isn't already queued — an explicit keyframe request means a fully wedged decoder that
+        // needs the IDR, which supersedes an RFI recovery. A failure (range older than the encoder's
+        // live references, or no RFI backend) falls through to the coalesced keyframe path below.
+        let mut rfi_range: Option<(u32, u32)> = None;
+        while let Ok((first, last)) = rfi.try_recv() {
+            rfi_range = Some(match rfi_range {
+                Some((pf, pl)) => (pf.min(first), pl.max(last)),
+                None => (first, last),
+            });
+        }
+        if !want_kf {
+            if let Some((first, last)) = rfi_range {
+                if enc.caps().supports_rfi && enc.invalidate_ref_frames(first as i64, last as i64) {
+                    // The RFI recovered the loss with a clean re-anchor P-frame (no IDR). Anchor the
+                    // keyframe cooldown so the client's echo of the SAME loss — its frames_dropped-
+                    // driven keyframe request, arriving ~one loss-window later — is coalesced away
+                    // instead of emitting a redundant full IDR right after the cheap recovery.
+                    last_forced_idr = Some(std::time::Instant::now());
+                } else {
+                    want_kf = true; // range too old / no RFI backend → coalesced keyframe below
+                }
+            }
+        }
         if want_kf {
             // Clients request a keyframe on EVERY FEC-unrecoverable frame (`frames_dropped` polling)
             // and keep asking until the IDR actually arrives + decodes — a full round-trip on a link
@@ -4225,11 +4302,28 @@ fn virtual_stream(ctx: SessionContext) -> Result<()> {
             last_au_at = std::time::Instant::now();
             encoder_resets = 0;
             let (cap_ns, sub_ns, deadline) = inflight.pop_front().expect("inflight non-empty");
-            let flags = if au.keyframe {
+            let mut flags = if au.keyframe {
                 (FLAG_PIC | FLAG_SOF) as u32
             } else {
                 FLAG_PIC as u32
             };
+            // Intra-refresh recovery marking (inert unless the backend validated its constrained GDR
+            // via `intra_refresh_recovery`): tag every wave-boundary AU with USER_FLAG_RECOVERY_POINT
+            // so the client lifts its post-loss freeze on the second mark — a proven clean re-anchor —
+            // instead of forcing a full IDR. See [`mark_recovery_boundary`] for the cadence.
+            let caps = enc.caps();
+            if caps.intra_refresh_recovery
+                && caps.intra_refresh_period > 0
+                && mark_recovery_boundary(&mut ir_wave_pos, au.keyframe, caps.intra_refresh_period)
+            {
+                flags |= punktfunk_core::packet::USER_FLAG_RECOVERY_POINT;
+            }
+            // Reference-frame-invalidation recovery frame (AMD LTR force-reference): a clean P-frame
+            // off a known-good reference. Tag it so the client lifts its post-loss freeze on this one
+            // AU without an IDR — the definitive single-frame re-anchor (see USER_FLAG_RECOVERY_ANCHOR).
+            if au.recovery_anchor {
+                flags |= punktfunk_core::packet::USER_FLAG_RECOVERY_ANCHOR;
+            }
             // Re-send the HDR mastering metadata (0xCE) on each keyframe (a decoder-resync point) and
             // whenever it changed, so a client that dropped the best-effort datagram re-converges.
             if let Some(m) = last_hdr_meta {
@@ -4654,6 +4748,32 @@ mod tests {
         assert!(reconfig_allowed(None, false));
     }
 
+    #[test]
+    fn recovery_marks_land_every_period_and_rephase_at_idr() {
+        let period = 4;
+        let mut pos = 0u32;
+        // Frames 1..=3 are mid-wave (no mark), frame 4 is the boundary; then it repeats.
+        let marks: Vec<bool> = (0..10)
+            .map(|_| mark_recovery_boundary(&mut pos, false, period))
+            .collect();
+        assert_eq!(
+            marks,
+            vec![false, false, false, true, false, false, false, true, false, false]
+        );
+
+        // An IDR mid-wave re-phases: the counter restarts, so the next boundary is a full period
+        // later (an IDR is itself a clean anchor, so it is not additionally marked).
+        let mut pos = 0u32;
+        assert!(!mark_recovery_boundary(&mut pos, false, period)); // pos 1
+        assert!(!mark_recovery_boundary(&mut pos, false, period)); // pos 2
+        assert!(!mark_recovery_boundary(&mut pos, true, period)); // IDR → pos 0, no mark
+        // Now a fresh full period is needed, not just the 2 remaining frames.
+        assert!(!mark_recovery_boundary(&mut pos, false, period)); // pos 1
+        assert!(!mark_recovery_boundary(&mut pos, false, period)); // pos 2
+        assert!(!mark_recovery_boundary(&mut pos, false, period)); // pos 3
+        assert!(mark_recovery_boundary(&mut pos, false, period)); // pos 4 → mark
+    }
+
     #[test]
     fn pad_snapshot_replaces_state_and_seq_gates() {
         use punktfunk_core::input::{gamepad, GamepadSnapshot};

crates/punktfunk-host/src/spike.rs

@@ -22,6 +22,11 @@ use std::time::Instant;
 pub enum Source {
     /// Deterministic moving BGRx test pattern — no capture session required.
     Synthetic,
+    /// Deterministic moving NV12 texture on the GPU (Windows only) — no capture session required.
+    /// Feeds the native AMF / D3D11 zero-copy encoders, which demand an NV12 GPU texture the CPU
+    /// `Synthetic` source can't give them. Used to validate GPU-encoder behaviour (e.g. AMF
+    /// intra-refresh) headlessly.
+    SyntheticNv12,
     /// Live monitor via the xdg ScreenCast portal + PipeWire.
     Portal,
     /// KWin virtual output created at `width`x`height` (zkde_screencast). Lets us validate
@@ -56,6 +61,27 @@ pub fn run(opts: Options) -> Result<()> {
             );
             Box::new(SyntheticCapturer::new(opts.width, opts.height, opts.fps))
         }
+        Source::SyntheticNv12 => {
+            #[cfg(target_os = "windows")]
+            {
+                tracing::info!(
+                    width = opts.width,
+                    height = opts.height,
+                    fps = opts.fps,
+                    "spike source: synthetic NV12 GPU texture (moving luma ramp)"
+                );
+                Box::new(
+                    capture::synthetic_nv12::SyntheticNv12Capturer::new(
+                        opts.width, opts.height, opts.fps,
+                    )
+                    .context("open synthetic NV12 capturer")?,
+                )
+            }
+            #[cfg(not(target_os = "windows"))]
+            {
+                anyhow::bail!("--source synthetic-nv12 is Windows-only (native AMF / D3D11 encoders)");
+            }
+        }
         Source::Portal => {
             tracing::info!("spike source: xdg ScreenCast portal (live monitor)");
             capture::open_portal_monitor().context("open portal capturer")?

include/punktfunk_core.h

@@ -255,6 +255,25 @@
 // feeding them to the decoder. Punktfunk/1 only (GameStream never sets it).
 #define FLAG_PROBE 8
 
+// Application `user_flags` bit (the u32 [`PacketHeader::user_flags`] word, surfaced to the client
+// as [`crate::session::Frame::flags`]) — NOT a transport packet flag. Marks the access unit that
+// **completes an intra-refresh wave**: the picture is loss-free from here even though the frame is
+// a coded `P` (no IDR, so the decoder never sets `AV_FRAME_FLAG_KEY`). The client lifts its
+// post-loss display freeze on this bit as well as on a real keyframe — the only bitstream-invisible
+// clean point it can honor without forcing a full IDR. Lives above the low nibble because the host
+// reuses `FLAG_PIC`/`FLAG_SOF`/`FLAG_PROBE` bit values inside `user_flags`; `0x10` clears all four.
+#define USER_FLAG_RECOVERY_POINT 16
+
+// Application `user_flags` bit — a **definitive single-frame clean re-anchor**. Unlike
+// [`USER_FLAG_RECOVERY_POINT`] (an intra-refresh wave boundary, where the first boundary after a loss
+// is only half-healed so the client waits for the second), this marks an access unit the host coded
+// to reference a **known-good** picture on purpose — an AMD **LTR reference-frame-invalidation**
+// recovery frame (`ForceLTRReferenceBitfield`): a clean P-frame off a long-term reference the client
+// already has, not an IDR. The picture is loss-free the instant this AU decodes, so the client lifts
+// its post-loss freeze on the **first** such mark. Coded `P` (no IDR), so the decoder never sets
+// `AV_FRAME_FLAG_KEY` — this host flag is the only signal.
+#define USER_FLAG_RECOVERY_ANCHOR 32
+
 // Largest UDP datagram the core will send or accept. `Config::validate` bounds
 // `shard_payload` so `HEADER_LEN + shard_payload + CRYPTO_OVERHEAD ≤ MAX_DATAGRAM_BYTES`.
 #define MAX_DATAGRAM_BYTES 2048
@@ -462,6 +481,11 @@
 #define MSG_BITRATE_CHANGED 6
 #endif
 
+#if defined(PUNKTFUNK_FEATURE_QUIC)
+// Type byte of [`RfiRequest`].
+#define MSG_RFI_REQUEST 7
+#endif
+
 #if defined(PUNKTFUNK_FEATURE_QUIC)
 // Type byte of [`ProbeRequest`].
 #define MSG_PROBE_REQUEST 32

scripts/ci/ensure-windows-toolchain.ps1

@@ -16,6 +16,34 @@ trap {
   exit 1
 }
 
+# --- reclaim disk before building -----------------------------------------------------------------
+# The windows-amd64 runner's system volume is intentionally small (100 GB) and a full Windows CI pass
+# writes ~50 GB of cargo target output into C:\t (x64) / C:\t-a64 (arm64). Left to accumulate across
+# runs that overflows the disk and the build dies with "no space on device" (os error 112) - exactly
+# what took the Windows host build down. The runner bakes in a reclaimer + a scheduled task that keeps
+# an idle box lean (unom/infra's setup-gitea-runner-base.ps1 ->
+# C:\Users\Public\act-runner\clean-runner-disk.ps1); call it here too so THIS job starts with headroom
+# regardless of when that task last ran. Threshold mode (no -Force): it only prunes when actually low,
+# so incremental-compile caches survive when there's room. Best-effort - a cleanup hiccup must never
+# fail the build.
+$reclaimer = 'C:\Users\Public\act-runner\clean-runner-disk.ps1'
+try {
+  if (Test-Path $reclaimer) {
+    & powershell.exe -NoProfile -ExecutionPolicy Bypass -File $reclaimer
+  }
+  else {
+    # Fallback for a runner not yet re-baked with the infra reclaimer: prune the big target dirs when low.
+    $freeGb = [math]::Round((Get-PSDrive C).Free / 1GB, 1)
+    Write-Host "[ensure-toolchain] clean-runner-disk.ps1 absent; C: free ${freeGb} GB"
+    if ($freeGb -lt 35) {
+      foreach ($d in 'C:\t', 'C:\t-a64') {
+        if (Test-Path $d) { Write-Host "  reclaiming $d"; Remove-Item $d -Recurse -Force -ErrorAction SilentlyContinue }
+      }
+    }
+  }
+}
+catch { Write-Warning "disk reclaim step failed (non-fatal): $_" }
+
 $ciDir = $PSScriptRoot
 & "$ciDir\provision-windows-wdk.ps1"
 & "$ciDir\provision-windows-punktfunk-extras.ps1"