diff --git a/docs/RUNBOOK.md b/docs/RUNBOOK.md
index 8f22923..680028d 100644
--- a/docs/RUNBOOK.md
+++ b/docs/RUNBOOK.md
@@ -133,8 +133,10 @@ Optional flags: `--port`, `--peer HOST[:PORT]`, `--bitrate KBPS`,
 
 Notes: the encoder runs a VBV that caps keyframe bursts to roughly two
 frame periods of bytes, and the receiver requests a 4MB UDP receive
-buffer (clamped by `net.core.rmem_max`). For very high `--bitrate`
-values, raise `net.core.rmem_max` on the receiver machine.
+buffer (clamped by `net.core.rmem_max`; the install script raises it to
+4 MB and persists the change). The remaining bottleneck on Wi-Fi-only
+boards is the wireless link itself — lower `--bitrate` until the stream
+is stable.
 
 The headless equivalent runs as part of `meson test` (`udp loopback` test):
 synthetic frames → encode → packetize → localhost UDP → depacketize →
diff --git a/scripts/install-receiver.sh b/scripts/install-receiver.sh
index 0dd9b63..287a7fd 100755
--- a/scripts/install-receiver.sh
+++ b/scripts/install-receiver.sh
@@ -145,6 +145,16 @@ install -m 755 "${BUILD_DIR}/src/app/screencast" "${INSTALL_DIR}/screencast"
 
 systemctl enable --now avahi-daemon 2>/dev/null || true
 
+# UDP receive buffers clamp to net.core.rmem_max (stock Debian: ~208 KB).
+# A keyframe burst at higher bitrates is larger than that, and the packet
+# loss it causes makes every keyframe undecodable (black screen). Applied
+# idempotently; buffers are allocated lazily, so this costs no RAM at rest.
+if [ "$(cat /proc/sys/net/core/rmem_max)" -lt 4194304 ]; then
+    log "raising net.core.rmem_max to 4 MB for RTP burst absorption"
+    sysctl -w net.core.rmem_max=4194304 >/dev/null
+    printf 'net.core.rmem_max=4194304\n' > /etc/sysctl.d/99-screencast-rmem.conf
+fi
+
 # --- systemd autostart --------------------------------------------------------
 if [ "${ENABLE_SERVICE}" = "1" ] && command -v systemctl >/dev/null 2>&1; then
     log "installing the systemd service (autostart on boot)"