diff --git a/extras/barnard-memwatch.sh b/extras/barnard-memwatch.sh
new file mode 100755
index 0000000..c975f2d
--- /dev/null
+++ b/extras/barnard-memwatch.sh
@@ -0,0 +1,80 @@
+#!/bin/bash
+# barnard-memwatch.sh
+# Description: Records barnard's memory use alongside its Go heap size, and
+# captures profiles before the kernel's OOM killer can take the evidence away.
+#
+# Start barnard with -profile, then run this alongside it. The ratio between
+# the two numbers is the diagnosis:
+#
+#   RSS large, go_heap small -> the growth is in cgo memory (OpenAL, opus,
+#                               rnnoise). The Go garbage collector cannot see
+#                               it and so applies no back pressure at all.
+#   RSS large, go_heap large -> the growth is Go-side, and the heap profile
+#                               this script dumps names what is holding it.
+#
+# Usage: barnard-memwatch.sh [output-log]
+#
+# Copyright 2026, Storm Dragon, <storm_dragon@linux-a11y.org>
+#
+# This is free software; you can redistribute it and/or modify it under the
+# terms of the GNU General Public License as published by the Free
+# Software Foundation; either version 3, or (at your option) any later
+# version.
+#
+# This software is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+# General Public License for more details.
+
+set -u
+
+# barnard serves its profiles here when started with -profile.
+profile_host="localhost:6060"
+# Dump profiles once resident memory passes this many kilobytes.
+dump_threshold_kb=2000000
+sample_interval=30
+
+output="${1:-$HOME/barnard-memwatch.log}"
+
+pid="$(pgrep -n -x barnard)" || {
+    echo "barnard is not running" >&2
+    exit 1
+}
+
+if ! curl -s -m 2 "http://${profile_host}/debug/pprof/" > /dev/null; then
+    echo "no profile server on ${profile_host}; start barnard with -profile" >&2
+    exit 1
+fi
+
+echo "watching barnard (pid ${pid}), writing to ${output}"
+
+dumped=0
+while kill -0 "$pid" 2> /dev/null; do
+    rss="$(awk '/^VmRSS/{print $2}' "/proc/${pid}/status" 2> /dev/null)"
+    swap="$(awk '/^VmSwap/{print $2}' "/proc/${pid}/status" 2> /dev/null)"
+    heap="$(curl -s -m 2 "http://${profile_host}/debug/pprof/heap?debug=1" \
+        | awk '/^# HeapInuse/{print $4}')"
+    goroutines="$(curl -s -m 2 "http://${profile_host}/debug/pprof/goroutine?debug=1" \
+        | head -1 | grep -o '[0-9]*')"
+    threads="$(ls "/proc/${pid}/task" 2> /dev/null | wc -l)"
+
+    printf '%s rss=%skB swap=%skB go_heap=%sB goroutines=%s threads=%s\n' \
+        "$(date +%T)" "${rss:-?}" "${swap:-0}" "${heap:-?}" \
+        "${goroutines:-?}" "${threads}" >> "$output"
+
+    # Capture the evidence once, while the process is still alive to ask.
+    if [[ ${dumped} -eq 0 && ${rss:-0} -gt ${dump_threshold_kb} ]]; then
+        dumped=1
+        curl -s -m 10 -o "${output}.heap" \
+            "http://${profile_host}/debug/pprof/heap"
+        curl -s -m 10 -o "${output}.goroutine" \
+            "http://${profile_host}/debug/pprof/goroutine?debug=2"
+        cp "/proc/${pid}/smaps_rollup" "${output}.smaps" 2> /dev/null
+        printf '%s *** dumped profiles at rss=%skB ***\n' \
+            "$(date +%T)" "${rss}" >> "$output"
+    fi
+
+    sleep "${sample_interval}"
+done
+
+echo "barnard exited; log is in ${output}"