diff --git a/.gitea/workflows/ci.yml b/.gitea/workflows/ci.yml
index 19af2119..92e53e93 100644
--- a/.gitea/workflows/ci.yml
+++ b/.gitea/workflows/ci.yml
@@ -26,6 +26,22 @@ jobs:
           apt-get update
           apt-get install -y --no-install-recommends libgtk-4-dev libadwaita-1-dev libsdl3-dev
 
+      # The committed pf-zerocopy SPIR-V blobs are pulled in with include_bytes! and rebuilt only
+      # by hand — edit a .comp, forget the rebuild, and the OLD kernel ships with no compile error
+      # or failing test. Recompile each shader and diff the disassembly. Filtering OpSourceExtension
+      # (+ --no-header) is exactly what absorbs the shaderc-vs-glslang generator difference; every
+      # instruction, ID and constant must match.
+      - name: Shader SPIR-V drift gate (pf-zerocopy)
+        run: |
+          apt-get install -y --no-install-recommends glslang-tools spirv-tools
+          for s in rgb2nv12_buf cursor_blend; do
+            d=crates/pf-zerocopy/src/imp
+            glslangValidator -V "$d/$s.comp" -o "/tmp/$s.spv" >/dev/null
+            diff <(spirv-dis --no-header "$d/$s.spv"  | grep -v OpSourceExtension) \
+                 <(spirv-dis --no-header "/tmp/$s.spv" | grep -v OpSourceExtension) \
+              || { echo "::error::$d/$s.spv is stale — rebuild it from $s.comp"; exit 1; }
+          done
+
       # Best-effort caches (act_runner's built-in cache server). Keyed on Cargo.lock:
       # registry/git are download caches, target/ the incremental build. The target key
       # carries the rustc version — resolved via `rustc --version` (below) rather than parsed
diff --git a/crates/pf-capture/src/linux/pipewire.rs b/crates/pf-capture/src/linux/pipewire.rs
index 70e10810..4fbd3a53 100644
--- a/crates/pf-capture/src/linux/pipewire.rs
+++ b/crates/pf-capture/src/linux/pipewire.rs
@@ -321,14 +321,15 @@ fn consume_frame(ud: &mut UserData, spa_buf: *mut spa::sys::spa_buffer) {
     // attach no fence. Covers both the GPU import and the CPU mmap read below.
     if datas[0].type_() == pw::spa::buffer::DataType::DmaBuf {
         match pf_zerocopy::dmabuf_fence::wait_read_ready(datas[0].fd(), 100) {
-            Ok(waited) => {
+            Ok(outcome) => {
                 static F1: std::sync::atomic::AtomicBool = std::sync::atomic::AtomicBool::new(true);
                 if F1.swap(false, Ordering::Relaxed) {
                     tracing::info!(
-                        waited,
-                        "dmabuf implicit-fence sync active (waited=true → driver fences \
-                         the render, race closed; false → no implicit fence, zero-copy \
-                         may still show stale frames)"
+                        ?outcome,
+                        "dmabuf implicit-fence sync active (Signaled → driver fences the \
+                         render, race closed; NoFence → no implicit fence, zero-copy may \
+                         still show stale frames; TimedOut → fence pending past 100ms, \
+                         proceeded anyway)"
                     );
                 }
             }
diff --git a/crates/pf-zerocopy/src/dmabuf_fence.rs b/crates/pf-zerocopy/src/dmabuf_fence.rs
index cb5afb24..406d99cc 100644
--- a/crates/pf-zerocopy/src/dmabuf_fence.rs
+++ b/crates/pf-zerocopy/src/dmabuf_fence.rs
@@ -11,12 +11,14 @@
 //! fd we can `poll()` — readable once the producer's writes complete. This makes zero-copy capture
 //! race-free WITHOUT the producer doing anything, *iff* the driver actually attaches the fence. If it
 //! attaches none, the export yields an already-signaled sync_file (poll returns immediately) — no
-//! wait, no harm, and `waited=false` tells us the driver doesn't fence (so zero-copy would still race).
+//! wait, no harm, and [`WaitOutcome::NoFence`] tells us the driver doesn't fence (so zero-copy
+//! would still race).
 
 // Every `unsafe` block in this file carries a `// SAFETY:` proof; enforce it (unsafe-proof program).
 #![deny(clippy::undocumented_unsafe_blocks)]
 
 use std::os::fd::RawFd;
+use std::time::{Duration, Instant};
 
 // linux/dma-buf.h ioctls on the DMA_BUF_BASE ('b' = 0x62) magic. _IOWR = dir(3)<<30 | size<<16 | base<<8 | nr.
 const DMA_BUF_BASE: u64 = 0x62;
@@ -34,11 +36,25 @@ const DMA_BUF_IOCTL_EXPORT_SYNC_FILE: u64 = iowr(2, std::mem::size_of::<DmaBufEx
 /// We will READ the buffer → export the fence(s) we must wait for before reading (the producer's writes).
 const DMA_BUF_SYNC_READ: u32 = 1 << 0;
 
-/// Wait until the producer's writes to `dmabuf_fd` complete (or `timeout_ms` elapses). Returns:
-/// - `Ok(true)`  — a render was still in flight and we waited on its fence (the race was real, now closed).
-/// - `Ok(false)` — no fence / already signaled (the driver attaches no implicit fence; zero-copy can race).
-/// - `Err`       — the ioctl failed (e.g. the kernel/driver lacks `EXPORT_SYNC_FILE`).
-pub fn wait_read_ready(dmabuf_fd: RawFd, timeout_ms: i32) -> std::io::Result<bool> {
+/// What the implicit-fence wait actually observed — the operator-facing diagnostic for "does
+/// implicit fencing work on this box" must not conflate these (a timeout or an interrupted wait
+/// proceeds with a possibly mid-render buffer; a signaled fence closed the race for real).
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+pub enum WaitOutcome {
+    /// No sync_file / already signaled — the driver attaches no implicit fence (or the render
+    /// finished before we looked); zero-copy can still race.
+    NoFence,
+    /// A render was in flight and we waited until its fence signaled — the race was real, now closed.
+    Signaled,
+    /// A render was in flight and `timeout_ms` elapsed first — we proceed (fail-open: blocking
+    /// longer would stall capture), possibly encoding a mid-render buffer.
+    TimedOut,
+}
+
+/// Wait until the producer's writes to `dmabuf_fd` complete (or `timeout_ms` elapses; negative =
+/// no timeout). `Err` means the ioctl or the poll itself failed (e.g. the kernel/driver lacks
+/// `EXPORT_SYNC_FILE`); see [`WaitOutcome`] for the success cases.
+pub fn wait_read_ready(dmabuf_fd: RawFd, timeout_ms: i32) -> std::io::Result<WaitOutcome> {
     let mut req = DmaBufExportSyncFile {
         flags: DMA_BUF_SYNC_READ,
         fd: -1,
@@ -54,31 +70,90 @@ pub fn wait_read_ready(dmabuf_fd: RawFd, timeout_ms: i32) -> std::io::Result<boo
     }
     let sync_fd = req.fd;
     if sync_fd < 0 {
-        return Ok(false); // no sync_file exported
-    }
-    let mut pfd = libc::pollfd {
-        fd: sync_fd,
-        events: libc::POLLIN,
-        revents: 0,
-    };
-    // Non-blocking probe: not-yet-signaled (poll==0) means the producer is still rendering.
-    // SAFETY: `&mut pfd` points at a single live `libc::pollfd` and `nfds == 1` matches that one
-    // element; `pfd.fd` is `sync_fd`, the sync_file fd just exported (already checked `>= 0`).
-    // `poll` reads `fd`/`events` and writes `revents` for this non-blocking (timeout 0) probe, then
-    // returns — `pfd` outlives the call and aliases nothing.
-    let pending = unsafe { libc::poll(&mut pfd, 1, 0) } == 0;
-    if pending {
-        pfd.revents = 0;
-        // SAFETY: same live single-element `pfd` (its `revents` reset to 0 just above), `nfds == 1`,
-        // and `sync_fd` still open. This blocking `poll` (up to `timeout_ms`) waits for the render
-        // fence to signal; it reads `fd`/`events`, writes `revents`, and returns before `pfd` ends.
-        unsafe { libc::poll(&mut pfd, 1, timeout_ms) }; // block until the render fence signals
+        return Ok(WaitOutcome::NoFence); // no sync_file exported
     }
+    let outcome = poll_readable(sync_fd, timeout_ms);
     // SAFETY: `sync_fd` is the sync_file fd the EXPORT_SYNC_FILE ioctl created and handed us to own;
     // this point is reached only when `sync_fd >= 0`, this `close` runs exactly once on it, and it is
     // never used afterward — no double-close or use-after-close.
     unsafe { libc::close(sync_fd) };
-    Ok(pending)
+    outcome
+}
+
+/// Poll `fd` for `POLLIN`: a non-blocking probe first (already-readable ⇒ [`WaitOutcome::NoFence`]
+/// — the fence was signaled before we looked), then a blocking wait up to `timeout_ms` (negative =
+/// no timeout). `EINTR` retries with the remaining budget instead of silently skipping the wait —
+/// the host spawns subprocesses, so a `SIGCHLD` mid-poll is a real occurrence, and skipping here
+/// would reopen the exact stale-frame race this file exists to close.
+fn poll_readable(fd: RawFd, timeout_ms: i32) -> std::io::Result<WaitOutcome> {
+    let mut pfd = libc::pollfd {
+        fd,
+        events: libc::POLLIN,
+        revents: 0,
+    };
+    // Non-blocking probe: not-yet-readable (poll == 0) means the producer is still rendering.
+    let probed = loop {
+        // SAFETY: `&mut pfd` points at a single live `libc::pollfd` and `nfds == 1` matches that
+        // one element; `fd` is the caller's live sync_file fd. `poll` reads `fd`/`events` and
+        // writes `revents` for this non-blocking (timeout 0) probe, then returns — `pfd` outlives
+        // the call and aliases nothing.
+        let r = unsafe { libc::poll(&mut pfd, 1, 0) };
+        if r >= 0 {
+            break r;
+        }
+        let e = std::io::Error::last_os_error();
+        if e.raw_os_error() != Some(libc::EINTR) {
+            return Err(e);
+        }
+    };
+    if probed > 0 {
+        if pfd.revents & libc::POLLIN != 0 {
+            return Ok(WaitOutcome::NoFence); // signaled before we looked
+        }
+        // POLLERR/POLLNVAL without POLLIN — the fd is broken, not signaled.
+        return Err(std::io::Error::other(format!(
+            "poll(sync_file) revents {:#x} without POLLIN",
+            pfd.revents
+        )));
+    }
+    let deadline =
+        (timeout_ms >= 0).then(|| Instant::now() + Duration::from_millis(timeout_ms as u64));
+    loop {
+        let remaining = match deadline {
+            None => -1, // poll's "no timeout"
+            Some(d) => match d.checked_duration_since(Instant::now()) {
+                None => return Ok(WaitOutcome::TimedOut),
+                // +1: round up so a sub-millisecond remainder still waits instead of busy-polling.
+                Some(rem) => (rem.as_millis() as i32).saturating_add(1),
+            },
+        };
+        pfd.revents = 0;
+        // SAFETY: same live single-element `pfd` (its `revents` reset to 0 just above), `nfds == 1`,
+        // and `fd` still open (closed by the caller only after this function returns). This blocking
+        // `poll` (up to `remaining` ms) waits for the fence to signal; it reads `fd`/`events`,
+        // writes `revents`, and returns before `pfd` ends.
+        let r = unsafe { libc::poll(&mut pfd, 1, remaining) };
+        match r {
+            0 => return Ok(WaitOutcome::TimedOut),
+            r if r > 0 => {
+                if pfd.revents & libc::POLLIN != 0 {
+                    return Ok(WaitOutcome::Signaled);
+                }
+                // POLLERR/POLLNVAL without POLLIN — the fd is broken, not signaled.
+                return Err(std::io::Error::other(format!(
+                    "poll(sync_file) revents {:#x} without POLLIN",
+                    pfd.revents
+                )));
+            }
+            _ => {
+                let e = std::io::Error::last_os_error();
+                if e.raw_os_error() != Some(libc::EINTR) {
+                    return Err(e);
+                }
+                // Interrupted — recompute the remaining budget and keep waiting.
+            }
+        }
+    }
 }
 
 #[cfg(test)]
@@ -90,4 +165,24 @@ mod tests {
     fn ioctl_number_matches_dma_buf_h() {
         assert_eq!(DMA_BUF_IOCTL_EXPORT_SYNC_FILE, 0xC008_6202);
     }
+
+    /// The poll state machine, driven by a pipe (readable ⇔ signaled): a not-yet-readable fd that
+    /// stays quiet is a truthful `TimedOut` (not the old `Ok(true)`), and one already readable at
+    /// the probe is `NoFence`.
+    #[test]
+    fn poll_readable_reports_the_truth() {
+        use std::io::Write;
+        use std::os::fd::AsRawFd;
+
+        let (r, mut w) = std::io::pipe().unwrap();
+        assert_eq!(
+            poll_readable(r.as_raw_fd(), 10).unwrap(),
+            WaitOutcome::TimedOut
+        );
+        w.write_all(b"x").unwrap();
+        assert_eq!(
+            poll_readable(r.as_raw_fd(), 10).unwrap(),
+            WaitOutcome::NoFence
+        );
+    }
 }
diff --git a/crates/pf-zerocopy/src/imp/client.rs b/crates/pf-zerocopy/src/imp/client.rs
index e260dc6a..303ef213 100644
--- a/crates/pf-zerocopy/src/imp/client.rs
+++ b/crates/pf-zerocopy/src/imp/client.rs
@@ -1,5 +1,5 @@
 //! Host side of the isolated zero-copy GPU import (design:
-//! [`design/zerocopy-worker-isolation.md`]): spawns the `zerocopy-worker` subprocess, mirrors the
+//! `design/zerocopy-worker-isolation.md`): spawns the `zerocopy-worker` subprocess, mirrors the
 //! [`super::egl::EglImporter`] entry points over the [`super::proto`] socket, and materializes
 //! the worker's pooled CUDA buffers in this process via CUDA IPC (each buffer's handles are
 //! opened exactly once and reused as the pool recycles). A worker death — the whole point of the
@@ -151,7 +151,7 @@ pub struct RemoteImporter {
 
 impl RemoteImporter {
     /// Spawn the worker from this host binary and complete the readiness handshake. The worker
-    /// is exec'd through the pinned [`SELF_EXE`] fd, so it is always the exact image this
+    /// is exec'd through the pinned `SELF_EXE` fd, so it is always the exact image this
     /// process runs — even after the installed binary was replaced mid-flight. An `Err` here
     /// means "no isolated zero-copy available" — callers fall back to the CPU path, exactly like
     /// an in-process `EglImporter::new()` failure.
diff --git a/crates/pf-zerocopy/src/imp/cuda.rs b/crates/pf-zerocopy/src/imp/cuda.rs
index 7a5a98f1..0526e558 100644
--- a/crates/pf-zerocopy/src/imp/cuda.rs
+++ b/crates/pf-zerocopy/src/imp/cuda.rs
@@ -1,4 +1,4 @@
-//! CUDA driver-side state for the zero-copy path, layered over the raw driver-API FFI in [`ffi`]
+//! CUDA driver-side state for the zero-copy path, layered over the raw driver-API FFI in `ffi`
 //! (the `dlopen`'d `libcuda.so.1` symbol table — hand-rolled because no Rust crate exposes the
 //! GL-interop calls, and runtime-loaded so one binary runs on NVIDIA *and* on AMD/Intel where
 //! `libcuda` is absent). This facade owns the higher-level pieces on top of that layer:
@@ -356,7 +356,10 @@ fn alloc_pitched_nv12(
     // SAFETY: two independent `cuMemAllocPitch_v2` calls (wrapper → live table). `&mut y_ptr`/
     // `&mut y_pitch` and `&mut uv_ptr`/`&mut uv_pitch` are live, distinct stack out-params the
     // driver writes each plane's pointer and pitch into; all outlive their synchronous calls. The
-    // dimension/element-size args are by-value ints. No aliasing — four separate locals.
+    // dimension/element-size args are by-value ints. No aliasing — four separate locals. If the UV
+    // allocation fails, the just-created Y allocation is freed before the error propagates — this
+    // runs per frame under VRAM pressure (`BufferPool::get`'s pool-miss fallback), so a leak here
+    // would compound exactly when memory is already scarce.
     unsafe {
         ck(
             cuMemAllocPitch_v2(
@@ -369,7 +372,7 @@ fn alloc_pitched_nv12(
             "cuMemAllocPitch_v2(Y)",
         )?;
         // Chroma is W/2 samples wide at 2 bytes each = W bytes; H/2 rows.
-        ck(
+        if let Err(e) = ck(
             cuMemAllocPitch_v2(
                 &mut uv_ptr,
                 &mut uv_pitch,
@@ -378,7 +381,10 @@ fn alloc_pitched_nv12(
                 16,
             ),
             "cuMemAllocPitch_v2(UV)",
-        )?;
+        ) {
+            let _ = cuMemFree_v2(y_ptr);
+            return Err(e);
+        }
     }
     Ok(((y_ptr, y_pitch), (uv_ptr, uv_pitch)))
 }
@@ -432,7 +438,7 @@ impl InputSurface {
     }
 
     /// Planar YUV444 (8-bit 4:4:4): one allocation, Y|U|V full-res planes stacked (see
-    /// [`alloc_pitched_yuv444`]).
+    /// `alloc_pitched_yuv444`).
     pub fn alloc_yuv444(width: u32, height: u32) -> Result<InputSurface> {
         let (ptr, pitch) = alloc_pitched_yuv444(width, height)?;
         Ok(InputSurface { ptr, pitch, height })
@@ -509,7 +515,7 @@ pub struct BufferPool {
     pitch: usize,
     /// NV12 pools carry a second (chroma) pitch; `Some` ⇒ buffers from this pool have a UV plane.
     uv_pitch: Option<usize>,
-    /// YUV444 pools: one allocation of 3·`height` stacked 1-byte planes (see [`alloc_pitched_yuv444`]).
+    /// YUV444 pools: one allocation of 3·`height` stacked 1-byte planes (see `alloc_pitched_yuv444`).
     yuv444: bool,
 }
 
@@ -629,10 +635,10 @@ pub struct DeviceBuffer {
     pub pitch: usize,
     pub width: u32,
     pub height: u32,
-    /// NV12 only: the interleaved chroma plane `(ptr, pitch)` paired with the Y plane in [`ptr`].
-    /// `None` for the default 4-byte RGB/BGRx path. When `Some`, [`ptr`] is the Y plane (1 byte/px).
+    /// NV12 only: the interleaved chroma plane `(ptr, pitch)` paired with the Y plane in [`ptr`](Self::ptr).
+    /// `None` for the default 4-byte RGB/BGRx path. When `Some`, [`ptr`](Self::ptr) is the Y plane (1 byte/px).
     pub uv: Option<(CUdeviceptr, usize)>,
-    /// Planar YUV444: [`ptr`] is ONE allocation of 3·[`height`](Self::height) rows at
+    /// Planar YUV444: [`ptr`](Self::ptr) is ONE allocation of 3·[`height`](Self::height) rows at
     /// [`pitch`](Self::pitch) — the full-res 1-byte Y, U, V planes stacked in that order
     /// (`uv` stays `None`; the single-plane wire/IPC path carries it unchanged).
     pub yuv444: bool,
@@ -1005,8 +1011,15 @@ pub fn copy_nv12_to_device(
     // `sync: false` shifts the source-lifetime obligation to the caller (documented above).
     // Wrappers → live table.
     unsafe {
-        copy_async(&y, "cuMemcpy2DAsync_v2(nv12 Y dev->dev)")?;
-        copy_async(&uv, "cuMemcpy2DAsync_v2(nv12 UV dev->dev)")?;
+        // On a failed enqueue, drain the stream before propagating: the caller drops `src` on
+        // `Err`, which recycles it into its pool — a copy still in flight would then race the
+        // next frame written into the same allocation.
+        let r = copy_async(&y, "cuMemcpy2DAsync_v2(nv12 Y dev->dev)")
+            .and_then(|()| copy_async(&uv, "cuMemcpy2DAsync_v2(nv12 UV dev->dev)"));
+        if r.is_err() {
+            let _ = sync_copy_stream();
+            return r;
+        }
         if sync {
             sync_copy_stream()?;
         }
@@ -1045,8 +1058,15 @@ pub fn copy_yuv444_to_device(
         // `src.ptr + pitch·h·i` stays within the live 3·H-row stacked allocation (`yuv444`
         // checked above), `dst_ptr`/`dst_pitch` is the caller's live NVENC plane; `w`×`h` fits
         // both. Completion is the trailing stream sync below (`sync`) or the caller's
-        // stream-ordering obligation (`sync: false`, documented above). Wrapper → live table.
-        unsafe { copy_async(&copy, "cuMemcpy2DAsync_v2(yuv444 plane dev->dev)")? };
+        // stream-ordering obligation (`sync: false`, documented above). A failed enqueue drains
+        // the stream first — earlier planes are already queued, and the caller drops (recycles)
+        // `src` on `Err`. Wrapper → live table.
+        unsafe {
+            if let Err(e) = copy_async(&copy, "cuMemcpy2DAsync_v2(yuv444 plane dev->dev)") {
+                let _ = sync_copy_stream();
+                return Err(e);
+            }
+        }
     }
     if sync {
         // SAFETY: one stream sync after the last enqueue covers all three planes (FIFO) — the
diff --git a/crates/pf-zerocopy/src/imp/cursor_blend.comp b/crates/pf-zerocopy/src/imp/cursor_blend.comp
index 0d940361..41aa8479 100644
--- a/crates/pf-zerocopy/src/imp/cursor_blend.comp
+++ b/crates/pf-zerocopy/src/imp/cursor_blend.comp
@@ -17,7 +17,8 @@
 // rows for NV12, whose 2 chroma bytes-pairs land in one exclusive word). Spans are aligned to the
 // SURFACE, not the cursor, so neighbouring invocations never share a word even at odd `ox`.
 //
-// Rebuild: glslc cursor_blend.comp -o cursor_blend.spv   (vendored beside this file)
+// Rebuild: glslangValidator -V cursor_blend.comp -o cursor_blend.spv   (vendored beside this
+// file; or glslc — CI diffs the disassembly against this source)
 
 layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
diff --git a/crates/pf-zerocopy/src/imp/egl.rs b/crates/pf-zerocopy/src/imp/egl.rs
index 4f4c482c..41f95733 100644
--- a/crates/pf-zerocopy/src/imp/egl.rs
+++ b/crates/pf-zerocopy/src/imp/egl.rs
@@ -8,7 +8,7 @@
 //! format is opaque). So we follow OBS/Sunshine: bind the `EGLImage` to a GL texture
 //! (`glEGLImageTargetTexture2DOES`), render it through a fullscreen-triangle shader into a plain
 //! immutable `GL_RGBA8` texture (de-tiling and swizzling to the BGRx the encoder wants), then
-//! register *that* texture with CUDA ([`MappedTexture`]) and copy it device-to-device into an
+//! register *that* texture with CUDA (`cuda::RegisteredTexture`) and copy it device-to-device into an
 //! owned [`DeviceBuffer`] so the dmabuf can be returned to the compositor immediately.
 
 #![allow(non_upper_case_globals)]
diff --git a/crates/pf-zerocopy/src/imp/mod.rs b/crates/pf-zerocopy/src/imp/mod.rs
index e51d1278..2e2893ff 100644
--- a/crates/pf-zerocopy/src/imp/mod.rs
+++ b/crates/pf-zerocopy/src/imp/mod.rs
@@ -24,11 +24,33 @@ use std::sync::atomic::{AtomicBool, AtomicU32, Ordering};
 pub use cuda::DeviceBuffer;
 pub use egl::{DmabufPlane, EglImporter};
 
-/// Whether a `PUNKTFUNK_*` flag is truthy (`1`/`true`/`yes`/`on`), or `None` when unset.
+/// Whether a `PUNKTFUNK_*` flag is truthy (`1`/`true`/`yes`/`on`, any case), falsy (`0`/`false`/
+/// `no`/`off`, any case), or `None` when unset — and, crucially, `None` for an *unrecognised*
+/// spelling too. These flags default ON (`PUNKTFUNK_ZEROCOPY`, `PUNKTFUNK_NV12`), so treating a
+/// typo'd `TRUE` as "off" silently inverted the operator's intent host-wide (a systemd drop-in or
+/// Nix module is exactly where such spellings come from). Unrecognised values are logged once so
+/// they are visible instead of silent.
 fn flag_opt(name: &str) -> Option<bool> {
-    std::env::var(name)
-        .ok()
-        .map(|v| matches!(v.trim(), "1" | "true" | "yes" | "on"))
+    let v = std::env::var(name).ok()?;
+    match v.trim().to_ascii_lowercase().as_str() {
+        "1" | "true" | "yes" | "on" => Some(true),
+        "0" | "false" | "no" | "off" | "" => Some(false),
+        _ => {
+            use std::collections::HashSet;
+            use std::sync::Mutex;
+            static WARNED: Mutex<Option<HashSet<String>>> = Mutex::new(None);
+            let mut g = WARNED.lock().unwrap();
+            if g.get_or_insert_with(HashSet::new).insert(name.to_string()) {
+                tracing::warn!(
+                    flag = name,
+                    value = %v,
+                    "unrecognised boolean value for this PUNKTFUNK_* flag — expected \
+                     1/true/yes/on or 0/false/no/off (any case); using the flag's default"
+                );
+            }
+            None
+        }
+    }
 }
 
 /// Whether a `PUNKTFUNK_*` flag is truthy (`1`/`true`/`yes`/`on`); unset ⇒ false.
@@ -49,7 +71,7 @@ pub fn vaapi_dmabuf_forced() -> bool {
 /// (`design/zerocopy-worker-isolation.md`), so a driver fault on a producer-invalidated dmabuf kills
 /// the worker and the host degrades to its capture-loss rebuild instead of dying — the reason the
 /// NVENC path stayed opt-in is gone. Fallbacks stay in place: VAAPI has a one-shot CPU downgrade if
-/// the LINEAR-dmabuf offer never negotiates ([`note_vaapi_dmabuf_failed`]); NVENC falls back per
+/// the LINEAR-dmabuf offer never negotiates ([`note_raw_dmabuf_negotiation_failed`]); NVENC falls back per
 /// capture when no importer/importable modifier is available and latches the import off after
 /// repeated worker deaths. `PUNKTFUNK_ZEROCOPY=0` opts out; `PUNKTFUNK_FORCE_SHM` forces the
 /// race-free SHM path.
@@ -204,7 +226,7 @@ static GPU_IMPORT_DISABLED: AtomicBool = AtomicBool::new(false);
 const GPU_IMPORT_DEATH_LATCH: u32 = 3;
 
 /// Record a worker death (transport-level failure). Latches the process-wide disable after
-/// [`GPU_IMPORT_DEATH_LATCH`] consecutive deaths.
+/// `GPU_IMPORT_DEATH_LATCH` consecutive deaths.
 pub fn note_gpu_import_death() {
     let streak = GPU_IMPORT_DEATH_STREAK.fetch_add(1, Ordering::Relaxed) + 1;
     if streak >= GPU_IMPORT_DEATH_LATCH && !GPU_IMPORT_DISABLED.swap(true, Ordering::Relaxed) {
@@ -243,7 +265,7 @@ static RAW_DMABUF_DISABLED: AtomicBool = AtomicBool::new(false);
 const RAW_DMABUF_FAILURE_LATCH: u32 = 3;
 
 /// Record an encoder-side raw-dmabuf import failure. Latches the process-wide disable after
-/// [`RAW_DMABUF_FAILURE_LATCH`] consecutive failures.
+/// `RAW_DMABUF_FAILURE_LATCH` consecutive failures.
 pub fn note_raw_dmabuf_import_failure(reason: &str) {
     let streak = RAW_DMABUF_FAILURE_STREAK.fetch_add(1, Ordering::Relaxed) + 1;
     if streak >= RAW_DMABUF_FAILURE_LATCH && !RAW_DMABUF_DISABLED.swap(true, Ordering::Relaxed) {
@@ -465,6 +487,38 @@ pub fn nv12_selftest() -> anyhow::Result<()> {
 mod tests {
     use super::*;
 
+    /// [`bt709_limited`] is the sole oracle for the GPU colour self-test — a coefficient typo here
+    /// would fail the self-test on a correct GPU (operator blames the driver), or, mirrored into
+    /// the shaders, pass it on genuinely wrong output. Pin it to externally-known BT.709
+    /// limited-range anchors instead of trusting it to check itself.
+    #[test]
+    fn bt709_limited_reference_matches_known_anchors() {
+        let close = |a: f64, b: f64| (a - b).abs() < 1e-9;
+
+        let (y, u, v) = bt709_limited(0, 0, 0); // black
+        assert!(
+            close(y, 16.0) && close(u, 128.0) && close(v, 128.0),
+            "black → ({y}, {u}, {v})"
+        );
+
+        let (y, u, v) = bt709_limited(255, 255, 255); // white
+        assert!(
+            close(y, 235.0) && close(u, 128.0) && close(v, 128.0),
+            "white → ({y}, {u}, {v})"
+        );
+
+        // Pure red saturates V (Kr row sums to exactly +0.5), pure blue saturates U.
+        let (_, _, v) = bt709_limited(255, 0, 0);
+        assert!(close(v, 240.0), "red V → {v}");
+        let (_, u, _) = bt709_limited(0, 0, 255);
+        assert!(close(u, 240.0), "blue U → {u}");
+
+        // One mid-scale anchor so a swapped Kr/Kb pair can't cancel out: BT.709 Y of pure green
+        // is 16 + 219·0.7152.
+        let (y, _, _) = bt709_limited(0, 255, 0);
+        assert!(close(y, 16.0 + 219.0 * 0.7152), "green Y → {y}");
+    }
+
     /// Single test owning the process-global latch statics (they are never reset by design).
     #[test]
     fn gpu_import_death_latch() {
diff --git a/crates/pf-zerocopy/src/imp/proto.rs b/crates/pf-zerocopy/src/imp/proto.rs
index 01635347..ccc774db 100644
--- a/crates/pf-zerocopy/src/imp/proto.rs
+++ b/crates/pf-zerocopy/src/imp/proto.rs
@@ -1,6 +1,6 @@
 //! Wire protocol between the PipeWire capture thread and the isolated zero-copy GPU-import
 //! worker process (`punktfunk-host zerocopy-worker`; design:
-//! [`design/zerocopy-worker-isolation.md`]). Transport is a `SOCK_SEQPACKET` unix socketpair —
+//! `design/zerocopy-worker-isolation.md`). Transport is a `SOCK_SEQPACKET` unix socketpair —
 //! reliable, ordered, message-framed (one `sendmsg` = one message) — with dmabuf fds riding as
 //! `SCM_RIGHTS` control data. Bodies are small serde_json blobs (~200 B/frame); pixels never
 //! cross the socket (they move GPU-side via CUDA IPC, see [`super::cuda::ipc_export`]).
diff --git a/crates/pf-zerocopy/src/imp/rgb2nv12_buf.comp b/crates/pf-zerocopy/src/imp/rgb2nv12_buf.comp
index fe809206..8e25cbf3 100644
--- a/crates/pf-zerocopy/src/imp/rgb2nv12_buf.comp
+++ b/crates/pf-zerocopy/src/imp/rgb2nv12_buf.comp
@@ -11,7 +11,8 @@
 // touch the same word (an 8-bit-storage-free way to write byte planes race-free). All pitches
 // and the UV offset are in WORDS and must be word-aligned (the Rust side sizes them so).
 //
-// Rebuild: glslc rgb2nv12_buf.comp -o rgb2nv12_buf.spv
+// Rebuild: glslangValidator -V rgb2nv12_buf.comp -o rgb2nv12_buf.spv   (or glslc; CI diffs the
+// disassembly against this source, tolerating only the generator header)
 layout(local_size_x = 8, local_size_y = 8) in;
 
 layout(std430, binding = 0) readonly buffer Src {
diff --git a/crates/pf-zerocopy/src/imp/vkslot.rs b/crates/pf-zerocopy/src/imp/vkslot.rs
index 1a3dec0e..a61426a9 100644
--- a/crates/pf-zerocopy/src/imp/vkslot.rs
+++ b/crates/pf-zerocopy/src/imp/vkslot.rs
@@ -45,7 +45,7 @@ use ash::vk;
 pub const CURSOR_MAX: u32 = cuda::CURSOR_MAX;
 
 /// The vendored SPIR-V for `cursor_blend.comp` (beside this file; rebuild with
-/// `glslc cursor_blend.comp -o cursor_blend.spv`).
+/// `glslangValidator -V cursor_blend.comp -o cursor_blend.spv`; CI gates drift).
 const CURSOR_SPV: &[u8] = include_bytes!("cursor_blend.spv");
 
 /// NVENC input-surface layout — selects the spec-constant `MODE` pipeline and the allocation
@@ -84,9 +84,10 @@ impl SlotFormat {
 }
 
 /// What the encoder holds per ring slot: the CUDA view it registers with NVENC plus the id it
-/// hands back to [`VkSlotBlend::blend`]. The backing Vulkan objects + CUDA mapping live in the
-/// [`VkSlotBlend`] (freed by [`free_slots`](VkSlotBlend::free_slots) / drop), so this is Copy —
-/// the encoder's ring keeps its existing shape.
+/// hands back to [`VkSlotBlend::blend_ref`] / [`VkSlotBlend::blend_ref_ordered`]. The backing
+/// Vulkan objects + CUDA mapping live in the [`VkSlotBlend`] (freed by
+/// [`free_slots`](VkSlotBlend::free_slots) / drop), so this is Copy — the encoder's ring keeps
+/// its existing shape.
 #[derive(Clone, Copy)]
 pub struct VkSlotRef {
     /// Device pointer NVENC registers (CUDA's mapping of the Vulkan memory).
@@ -654,7 +655,7 @@ impl VkSlotBlend {
     }
 
     /// Free every allocated slot (encoder teardown, alongside its ring clear). CUDA mappings drop
-    /// first (field order in [`SlotAlloc`] frees `cuda` via its own `Drop` before we free the VK
+    /// first (field order in `SlotAlloc` frees `cuda` via its own `Drop` before we free the VK
     /// objects explicitly here).
     pub fn free_slots(&mut self) {
         // Ordered blends return with their work still on the queue — quiesce before freeing the
@@ -1062,3 +1063,62 @@ impl Drop for VkSlotBlend {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn slot() -> VkSlotRef {
+        VkSlotRef {
+            ptr: 0,
+            pitch: 2048,
+            height: 1080,
+            id: 0,
+        }
+    }
+
+    fn geo(fmt: SlotFormat, cw: u32, ch: u32, ox: i32, oy: i32) -> Option<(Push, u32, u32)> {
+        VkSlotBlend::blend_geometry(&slot(), fmt, 1920, cw, ch, ox, oy)
+    }
+
+    /// An empty (clamped-away) cursor rect dispatches nothing.
+    #[test]
+    fn empty_rect_is_none() {
+        assert!(geo(SlotFormat::Argb, 0, 32, 10, 10).is_none());
+        assert!(geo(SlotFormat::Nv12, 32, 0, 10, 10).is_none());
+    }
+
+    /// Oversized bitmaps clamp to `CURSOR_MAX` — the push constants must agree with the staging
+    /// buffer's capacity, or the shader reads past the uploaded bitmap.
+    #[test]
+    fn cursor_dims_clamp_to_max() {
+        let (push, _, _) = geo(SlotFormat::Argb, CURSOR_MAX + 100, CURSOR_MAX + 1, 0, 0).unwrap();
+        assert_eq!(push.cur_w, CURSOR_MAX);
+        assert_eq!(push.cur_h, CURSOR_MAX);
+    }
+
+    /// ARGB dispatches per cursor pixel; NV12/YUV444 per word-aligned 4-px span, NV12 walking
+    /// 2-row blocks and YUV444 single rows. A 32×32 cursor at ox=13: spans cover the aligned
+    /// x∈[12,48) → 9 spans → 2 groups of 8.
+    #[test]
+    fn group_counts_per_format() {
+        let (_, gx, gy) = geo(SlotFormat::Argb, 32, 32, 13, 0).unwrap();
+        assert_eq!((gx, gy), (4, 4)); // 32/8 in both axes
+
+        let (_, gx, gy) = geo(SlotFormat::Nv12, 32, 32, 13, 0).unwrap();
+        assert_eq!((gx, gy), (2, 2)); // 9 spans → 2 groups; 16 2-row blocks → 2 groups
+
+        let (_, gx, gy) = geo(SlotFormat::Yuv444, 32, 32, 13, 0).unwrap();
+        assert_eq!((gx, gy), (2, 4)); // 9 spans → 2 groups; 32 rows → 4 groups
+    }
+
+    /// Negative `ox` must anchor spans with FLOOR alignment (`>>` on the signed value), not
+    /// truncating division: at ox=-5 the aligned start is -8, giving 9 spans over a 32-px cursor
+    /// (truncation would start at -4 and dispatch only 8 — silently dropping the right edge).
+    #[test]
+    fn negative_ox_floor_aligns_the_span_origin() {
+        let (push, gx, _) = geo(SlotFormat::Nv12, 32, 32, -5, 0).unwrap();
+        assert_eq!(push.ox, -5, "push constants carry the true origin");
+        assert_eq!(gx, 2, "9 spans from the floor-aligned start → 2 groups");
+    }
+}
diff --git a/crates/pf-zerocopy/src/imp/vulkan.rs b/crates/pf-zerocopy/src/imp/vulkan.rs
index cd8806fa..b0a9c846 100644
--- a/crates/pf-zerocopy/src/imp/vulkan.rs
+++ b/crates/pf-zerocopy/src/imp/vulkan.rs
@@ -51,7 +51,7 @@ struct Csc {
 }
 
 /// The buffer-to-buffer RGB→NV12 compute shader (see `rgb2nv12_buf.comp` beside this file;
-/// rebuild with `glslc rgb2nv12_buf.comp -o rgb2nv12_buf.spv`).
+/// rebuild with `glslangValidator -V rgb2nv12_buf.comp -o rgb2nv12_buf.spv`; CI gates drift).
 const CSC_SPV: &[u8] = include_bytes!("rgb2nv12_buf.spv");
 
 pub struct VkBridge {
diff --git a/crates/pf-zerocopy/src/imp/worker.rs b/crates/pf-zerocopy/src/imp/worker.rs
index d6ce0a3a..8fcf3304 100644
--- a/crates/pf-zerocopy/src/imp/worker.rs
+++ b/crates/pf-zerocopy/src/imp/worker.rs
@@ -1,5 +1,5 @@
 //! The isolated zero-copy GPU-import worker (`punktfunk-host zerocopy-worker`; design:
-//! [`design/zerocopy-worker-isolation.md`]). It owns the fragile driver stack — the headless
+//! `design/zerocopy-worker-isolation.md`). It owns the fragile driver stack — the headless
 //! EGLDisplay + GL context, the CUDA context, and the Vulkan bridge — so that a driver fault on a
 //! producer-invalidated dmabuf (the `cuGraphicsMapResources` SIGSEGV the F44 Game→Desktop switch
 //! reproduced) kills THIS process, not the streaming host. The host observes the dead socket,
@@ -43,9 +43,24 @@ pub fn run_from_args(args: &[String]) -> Result<()> {
         .transpose()
         .context("parse --fd")?
         .unwrap_or(3);
+    // Refuse to adopt anything that cannot be the spawning host's socket: a negative fd is
+    // outright UB inside `OwnedFd` (its niche), and 0–2 would make this worker close one of its
+    // own stdio streams on exit. Then confirm the number actually holds an open socket — the
+    // subcommand is hidden but runnable by hand, and adopting an arbitrary inherited fd would
+    // close it behind its real owner.
+    anyhow::ensure!(fd >= 3, "--fd must be >= 3 (got {fd})");
+    // SAFETY: `libc::stat` is plain-old-data for which all-zero is a valid value, so
+    // `mem::zeroed()` is a sound initializer; `fstat` writes into the live, correctly-sized
+    // `&mut st` and only reads `fd`. `st_mode` is read only after the return value is checked.
+    let is_socket = unsafe {
+        let mut st: libc::stat = std::mem::zeroed();
+        libc::fstat(fd, &mut st) == 0 && (st.st_mode & libc::S_IFMT) == libc::S_IFSOCK
+    };
+    anyhow::ensure!(is_socket, "--fd {fd} is not an open socket");
     // SAFETY: the spawning host `dup2`'d its socketpair end onto exactly this fd number before
-    // exec (the subcommand's contract) and nothing else in this fresh process owns it, so
-    // `OwnedFd` takes sole ownership and closes it exactly once at exit.
+    // exec (the subcommand's contract, just verified to be an open socket ≥ 3) and nothing else
+    // in this fresh process owns it, so `OwnedFd` takes sole ownership and closes it exactly
+    // once at exit.
     let sock = unsafe { OwnedFd::from_raw_fd(fd) };
     run(sock)
 }