diff --git a/.env.example b/.env.example
index 9d66eb8fbb..afb47dd356 100644
--- a/.env.example
+++ b/.env.example
@@ -279,6 +279,7 @@ REDIS_URL=redis://redis:6379                # REQUIRED for the self-host review
 # BROWSERLESS_QUEUED=4                       # (--profile visual-review) max requests waiting for a session.
 # BROWSERLESS_TIMEOUT_MS=30000               # (--profile visual-review) per-session timeout.
 # BROWSERLESS_MEM_LIMIT=2g                   # (--profile visual-review) container memory cap.
+# BROWSERLESS_EXPORTER_MEM_LIMIT=64m          # (--profile visual-review) browserless-exporter sidecar cap (#5368)
 # REVIEW_AUDIT_DIR=                          # persist visual-review screenshot PNGs to this filesystem path
 #                                            #   instead of re-rendering on demand. Unset = re-render each time.
 #                                            #   Only relevant when BROWSER_WS_ENDPOINT above is set. Ignored
diff --git a/docker-compose.yml b/docker-compose.yml
index 128af1d23a..8d0c91ad7c 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -553,6 +553,40 @@ services:
       start_period: 15s
       retries: 5
 
+  # browserless's own /metrics?token=<token> endpoint returns JSON (a rolling history of 5-minute-window
+  # snapshots), not Prometheus text -- this sidecar polls it and re-exposes the newest snapshot's fields as
+  # real Prometheus gauges (#5368), same "translate non-Prometheus output" pattern as backup-exporter above.
+  # Gated on the same --profile visual-review as browserless itself; meaningless without it running.
+  browserless-exporter:
+    image: alpine:3.20
+    restart: unless-stopped
+    <<: *default-logging
+    profiles: ["visual-review"]
+    depends_on:
+      browserless:
+        condition: service_healthy
+    environment:
+      BROWSERLESS_METRICS_URL: "http://browserless:3000/metrics"
+      BROWSERLESS_TOKEN: "${BROWSERLESS_TOKEN:-}"
+    volumes:
+      # Directory mount, not a single-file one -- same #reporting-exporter-stale-bind-mount rationale as
+      # the backup-exporter service above.
+      - ./scripts:/scripts:ro
+    expose:
+      - "9102"
+    command:
+      - /bin/sh
+      - -c
+      - "apk add --no-cache jq busybox-extras >/dev/null 2>&1 && sh /scripts/browserless-metrics.sh"
+    healthcheck:
+      test: ["CMD-SHELL", "wget -qO- http://127.0.0.1:9102/metrics | grep -q '^browserless_exporter_last_scrape_success'"]
+      interval: 30s
+      retries: 5
+    deploy:
+      resources:
+        limits:
+          memory: "${BROWSERLESS_EXPORTER_MEM_LIMIT:-64m}"
+
   # ── REES / review-enrichment (--profile rees) ──────────────────────────────
   # Self-hosted alternative to pointing REES_URL at a managed/external instance (see the self-hosting docs).
   # REES runs the heavier PR analysis (dependency CVEs, secret scan, license, EOL runtime, lockfile drift, IaC
diff --git a/grafana/dashboards/browserless-metrics.json b/grafana/dashboards/browserless-metrics.json
new file mode 100644
index 0000000000..0b224780b7
--- /dev/null
+++ b/grafana/dashboards/browserless-metrics.json
@@ -0,0 +1,129 @@
+{
+  "__inputs": [],
+  "__requires": [
+    { "type": "grafana", "id": "grafana", "name": "Grafana", "version": "10.0.0" },
+    { "type": "datasource", "id": "prometheus", "name": "Prometheus", "version": "1.0.0" }
+  ],
+  "annotations": { "list": [] },
+  "editable": false,
+  "fiscalYearStartMonth": 0,
+  "graphTooltip": 1,
+  "id": null,
+  "links": [],
+  "description": "browserless (headless-Chromium visual-review screenshot capture, --profile visual-review, #5368) queue depth, concurrency, and rejection/error/timeout rate, translated from its own JSON /metrics by the browserless-exporter sidecar. Absent/empty unless the visual-review profile is active.",
+  "panels": [
+    { "collapsed": false, "gridPos": { "h": 1, "w": 24, "x": 0, "y": 0 }, "id": 1, "title": "Exporter health", "type": "row" },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": {
+          "color": { "mode": "thresholds" },
+          "mappings": [
+            { "type": "value", "options": { "0": { "text": "FAILED", "color": "red" }, "1": { "text": "OK", "color": "green" } } }
+          ],
+          "thresholds": { "mode": "absolute", "steps": [{ "color": "red", "value": null }, { "color": "green", "value": 1 }] },
+          "noValue": "no data"
+        }
+      },
+      "gridPos": { "h": 5, "w": 8, "x": 0, "y": 1 },
+      "id": 2,
+      "options": { "colorMode": "background", "graphMode": "none", "justifyMode": "auto", "orientation": "horizontal", "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false }, "textMode": "auto" },
+      "title": "Last scrape of browserless's own /metrics",
+      "description": "0 means the exporter's most recent poll of browserless itself failed (wrong token, browserless down, or network issue) — the panels below are then serving stale data.",
+      "type": "stat",
+      "targets": [{ "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_exporter_last_scrape_success", "legendFormat": "scrape ok", "refId": "A" }]
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": { "defaults": { "color": { "mode": "thresholds" }, "mappings": [], "thresholds": { "mode": "absolute", "steps": [{ "color": "green", "value": null }] }, "unit": "dtdurations" } },
+      "gridPos": { "h": 5, "w": 8, "x": 8, "y": 1 },
+      "id": 3,
+      "options": { "colorMode": "value", "graphMode": "none", "justifyMode": "auto", "orientation": "auto", "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false }, "textMode": "auto" },
+      "title": "Time since browserless's own sample",
+      "description": "browserless_sample_timestamp_seconds is the timestamp browserless itself assigned to its newest 5-minute window. A large, growing value here means browserless has stopped reporting new windows even if the exporter's own scrape is succeeding.",
+      "type": "stat",
+      "targets": [{ "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "time() - browserless_sample_timestamp_seconds", "legendFormat": "age", "refId": "A" }]
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": { "defaults": { "color": { "mode": "thresholds" }, "mappings": [], "thresholds": { "mode": "absolute", "steps": [{ "color": "green", "value": null }] }, "unit": "ms" } },
+      "gridPos": { "h": 5, "w": 8, "x": 16, "y": 1 },
+      "id": 4,
+      "options": { "colorMode": "value", "graphMode": "none", "justifyMode": "auto", "orientation": "auto", "reduceOptions": { "calcs": ["lastNotNull"], "fields": "", "values": false }, "textMode": "auto" },
+      "title": "Mean session time",
+      "type": "stat",
+      "targets": [{ "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_session_mean_time_ms", "legendFormat": "mean", "refId": "A" }]
+    },
+
+    { "collapsed": false, "gridPos": { "h": 1, "w": 24, "x": 0, "y": 6 }, "id": 10, "title": "Queue, concurrency & outcomes", "type": "row" },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": { "defaults": { "color": { "mode": "palette-classic" }, "custom": { "lineWidth": 2, "fillOpacity": 10 } } },
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 7 },
+      "id": 11,
+      "options": { "legend": { "calcs": ["mean", "max"], "displayMode": "table", "placement": "bottom" }, "tooltip": { "mode": "multi", "sort": "desc" } },
+      "title": "Queue depth & concurrency",
+      "description": "queued: sessions waiting over the configured concurrency limit. running: sessions active at the sample point. max_concurrent: peak concurrent sessions in the window.",
+      "type": "timeseries",
+      "targets": [
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_queued", "legendFormat": "queued", "refId": "A" },
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_running", "legendFormat": "running", "refId": "B" },
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_max_concurrent", "legendFormat": "max_concurrent", "refId": "C" }
+      ]
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": { "defaults": { "color": { "mode": "palette-classic" }, "custom": { "lineWidth": 2, "fillOpacity": 10 } } },
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 7 },
+      "id": 12,
+      "options": { "legend": { "calcs": ["mean", "max"], "displayMode": "table", "placement": "bottom" }, "tooltip": { "mode": "multi", "sort": "desc" } },
+      "title": "Session outcomes by window",
+      "description": "Per-5-minute-window counts from browserless's own rolling buffer — a non-zero, sustained rejected/error/timeout/unhealthy count means capture is degrading, not just occasionally flaky.",
+      "type": "timeseries",
+      "targets": [
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_successful", "legendFormat": "successful", "refId": "A" },
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_rejected", "legendFormat": "rejected", "refId": "B" },
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_errors", "legendFormat": "error", "refId": "C" },
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_timedout", "legendFormat": "timedout", "refId": "D" },
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_unauthorized", "legendFormat": "unauthorized", "refId": "E" },
+        { "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_unhealthy", "legendFormat": "unhealthy", "refId": "F" }
+      ]
+    },
+
+    { "collapsed": false, "gridPos": { "h": 1, "w": 24, "x": 0, "y": 15 }, "id": 20, "title": "Resource utilization", "type": "row" },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": { "defaults": { "color": { "mode": "palette-classic" }, "custom": { "lineWidth": 2, "fillOpacity": 10 }, "unit": "percentunit", "max": 1, "min": 0 } },
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 16 },
+      "id": 21,
+      "options": { "legend": { "calcs": ["mean", "max"], "displayMode": "table", "placement": "bottom" }, "tooltip": { "mode": "multi", "sort": "desc" } },
+      "title": "CPU utilization",
+      "type": "timeseries",
+      "targets": [{ "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_cpu_ratio", "legendFormat": "cpu", "refId": "A" }]
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": { "defaults": { "color": { "mode": "palette-classic" }, "custom": { "lineWidth": 2, "fillOpacity": 10 }, "unit": "percentunit", "max": 1, "min": 0 } },
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 16 },
+      "id": 22,
+      "options": { "legend": { "calcs": ["mean", "max"], "displayMode": "table", "placement": "bottom" }, "tooltip": { "mode": "multi", "sort": "desc" } },
+      "title": "Memory utilization",
+      "type": "timeseries",
+      "targets": [{ "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" }, "expr": "browserless_memory_ratio", "legendFormat": "memory", "refId": "A" }]
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 38,
+  "tags": ["gittensory", "browserless", "observability"],
+  "templating": {
+    "list": [
+      { "current": {}, "hide": 0, "includeAll": false, "name": "DS_PROMETHEUS", "options": [], "query": "prometheus", "refresh": 1, "type": "datasource" }
+    ]
+  },
+  "time": { "from": "now-6h", "to": "now" },
+  "timepicker": {},
+  "timezone": "browser",
+  "title": "Gittensory — Browserless (visual review)",
+  "uid": "gittensory-browserless",
+  "version": 1
+}
diff --git a/grafana/dashboards/resource-hub.json b/grafana/dashboards/resource-hub.json
index c20c3eb4a8..c90e895c83 100644
--- a/grafana/dashboards/resource-hub.json
+++ b/grafana/dashboards/resource-hub.json
@@ -66,7 +66,7 @@
       "gridPos": { "h": 13, "w": 12, "x": 12, "y": 0 },
       "options": {
         "mode": "markdown",
-        "content": "## 📊 Dashboards\n- **[Upstream PRs & issues (GitHub)](/d/gittensory-github)** — live, accurate census + open-PR triage (GitHub API).\n- **[Reviews & PRs (maintainer)](/d/gittensory-maintainer)** — gittensory's own review activity + reviewed-PR log.\n- **[AI usage](/d/gittensory-ai-usage)** — durable cross-provider ai_usage_events (filterable by provider/feature/model), live Prometheus counters, and Claude Code's own OTEL session telemetry, all in one place.\n- **[Gittensory (infra)](/d/gittensory)** — queue, jobs, HTTP, GitHub API cache/rate limits.\n- **[GPU metrics](/d/gittensory-gpu)** — utilization/VRAM for a self-hosted Ollama GPU box.\n- **[Infra health](/d/gittensory-infra-health)** — host CPU/mem/disk/network (node-exporter), per-container resource usage (cAdvisor), Redis, Qdrant, and whether the observability stack itself is up.\n- **[REES (review-enrichment)](/d/gittensory-rees)** — request outcomes/latency and per-analyzer run/timeout/degrade rates for the optional `--profile rees` service.\n\n## 📈 Metrics & logs\n- **Prometheus** — [targets](http://localhost:9090/targets) · [graph](http://localhost:9090)\n- **Alertmanager** — [alerts](http://localhost:9093)\n- **Loki** — query in [Explore](/explore) (pick the *Loki* datasource), e.g. `{compose_service=\"gittensory\"}`\n- **Sentry** — release/source-map enriched errors. Edit the dashboard link if your project URL differs.\n\n## 🩺 Quick health checks\n| What | Where |\n|---|---|\n| App serving | `GET /ready` → 200 |\n| AI wired | boot log `selfhost_ai_provider` |\n| Embeds wired | boot log `selfhost_embed_provider` |\n| Vectors wired | boot log `selfhost_vectorize` |\n| Token spend | **[AI usage](/d/gittensory-ai-usage)** dashboard |\n\n## 📚 Docs\n- [Maintainer self-hosting](https://gittensory.aethereal.dev/docs/maintainer-self-hosting) — setup, configuration, AI, REES, RAG, operations, and troubleshooting."
+        "content": "## 📊 Dashboards\n- **[Upstream PRs & issues (GitHub)](/d/gittensory-github)** — live, accurate census + open-PR triage (GitHub API).\n- **[Reviews & PRs (maintainer)](/d/gittensory-maintainer)** — gittensory's own review activity + reviewed-PR log.\n- **[AI usage](/d/gittensory-ai-usage)** — durable cross-provider ai_usage_events (filterable by provider/feature/model), live Prometheus counters, and Claude Code's own OTEL session telemetry, all in one place.\n- **[Gittensory (infra)](/d/gittensory)** — queue, jobs, HTTP, GitHub API cache/rate limits.\n- **[GPU metrics](/d/gittensory-gpu)** — utilization/VRAM for a self-hosted Ollama GPU box.\n- **[Infra health](/d/gittensory-infra-health)** — host CPU/mem/disk/network (node-exporter), per-container resource usage (cAdvisor), Redis, Qdrant, and whether the observability stack itself is up.\n- **[REES (review-enrichment)](/d/gittensory-rees)** — request outcomes/latency and per-analyzer run/timeout/degrade rates for the optional `--profile rees` service.\n- **[Browserless (visual review)](/d/gittensory-browserless)** — queue depth, concurrency, and rejection/error/timeout rate for the optional `--profile visual-review` screenshot service.\n\n## 📈 Metrics & logs\n- **Prometheus** — [targets](http://localhost:9090/targets) · [graph](http://localhost:9090)\n- **Alertmanager** — [alerts](http://localhost:9093)\n- **Loki** — query in [Explore](/explore) (pick the *Loki* datasource), e.g. `{compose_service=\"gittensory\"}`\n- **Sentry** — release/source-map enriched errors. Edit the dashboard link if your project URL differs.\n\n## 🩺 Quick health checks\n| What | Where |\n|---|---|\n| App serving | `GET /ready` → 200 |\n| AI wired | boot log `selfhost_ai_provider` |\n| Embeds wired | boot log `selfhost_embed_provider` |\n| Vectors wired | boot log `selfhost_vectorize` |\n| Token spend | **[AI usage](/d/gittensory-ai-usage)** dashboard |\n\n## 📚 Docs\n- [Maintainer self-hosting](https://gittensory.aethereal.dev/docs/maintainer-self-hosting) — setup, configuration, AI, REES, RAG, operations, and troubleshooting."
       }
     }
   ]
diff --git a/prometheus/prometheus.yml b/prometheus/prometheus.yml
index a664e2af09..99e9565e08 100644
--- a/prometheus/prometheus.yml
+++ b/prometheus/prometheus.yml
@@ -90,6 +90,15 @@ scrape_configs:
       - targets: ["rees:8080"]
     scrape_interval: 15s
 
+  # browserless queue/concurrency/rejection/error metrics, translated from its own JSON /metrics by the
+  # browserless-exporter sidecar (--profile visual-review, #5368). Same BusyBox-httpd-wrapper shape as
+  # gittensory-backup above, so it needs the same explicit scrape protocol fallback.
+  - job_name: browserless
+    fallback_scrape_protocol: PrometheusText0.0.4
+    static_configs:
+      - targets: ["browserless-exporter:9102"]
+    scrape_interval: 30s
+
   # Self-monitoring: the observability stack watching itself (--profile observability, #5366). Every
   # component here ships its own native Prometheus /metrics — previously scraped by nothing, so a stack
   # component silently degrading (e.g. Loki dropping ingested lines, Alertmanager failing to notify) had
diff --git a/scripts/browserless-metrics.sh b/scripts/browserless-metrics.sh
new file mode 100644
index 0000000000..06ad20217a
--- /dev/null
+++ b/scripts/browserless-metrics.sh
@@ -0,0 +1,110 @@
+#!/bin/sh
+# Translate browserless's JSON /metrics endpoint into Prometheus text format (#5368). browserless's own
+# /metrics?token=<token> returns a rolling history of 5-minute-window snapshots (an array, oldest first) --
+# this exporter polls it, takes only the NEWEST snapshot (the array's last element), and re-exposes its
+# fields as Prometheus gauges. Gauges, not counters: each field is a per-window sample from browserless's own
+# rolling buffer, not a monotonic total, so it can legitimately go up or down between scrapes.
+# Intended for the browserless-exporter compose sidecar; it never touches the app's own state.
+set -eu
+
+URL=${BROWSERLESS_METRICS_URL:-http://browserless:3000/metrics}
+TOKEN=${BROWSERLESS_TOKEN:-}
+OUT=${BROWSERLESS_METRICS_DIR:-/metrics}
+FILE=${BROWSERLESS_METRICS_FILE:-$OUT/metrics}
+INTERVAL=${BROWSERLESS_METRICS_INTERVAL_SECONDS:-30}
+PORT=${BROWSERLESS_METRICS_PORT:-9102}
+
+case "$INTERVAL" in
+  ''|*[!0-9]*) INTERVAL=30 ;;
+esac
+
+write_metrics() {
+  mkdir -p "$OUT"
+  tmp="$FILE.tmp"
+  now=$(date +%s)
+  raw=$(wget -qO- "$URL?token=$TOKEN" 2>/dev/null || true)
+  latest=$(printf '%s' "$raw" | jq -c '.[-1] // empty' 2>/dev/null || true)
+
+  {
+    echo "# HELP browserless_exporter_last_scrape_success Whether the most recent poll of browserless's own /metrics succeeded (1) or failed (0)."
+    echo "# TYPE browserless_exporter_last_scrape_success gauge"
+    echo "# HELP browserless_exporter_last_scrape_timestamp_seconds Unix timestamp of the most recent poll attempt, successful or not."
+    echo "# TYPE browserless_exporter_last_scrape_timestamp_seconds gauge"
+
+    if [ -n "$latest" ]; then
+      echo "browserless_exporter_last_scrape_success 1"
+      echo "browserless_exporter_last_scrape_timestamp_seconds $now"
+
+      echo "# HELP browserless_queued Sessions queued (over browserless's concurrency limit) in the most recent 5-minute window."
+      echo "# TYPE browserless_queued gauge"
+      echo "browserless_queued $(printf '%s' "$latest" | jq '.queued // 0')"
+
+      echo "# HELP browserless_running Sessions currently running at the time of the most recent window sample."
+      echo "# TYPE browserless_running gauge"
+      echo "browserless_running $(printf '%s' "$latest" | jq '.running // 0')"
+
+      echo "# HELP browserless_max_concurrent Peak concurrent sessions observed in the most recent 5-minute window."
+      echo "# TYPE browserless_max_concurrent gauge"
+      echo "browserless_max_concurrent $(printf '%s' "$latest" | jq '.maxConcurrent // 0')"
+
+      echo "# HELP browserless_rejected Sessions rejected (over capacity) in the most recent 5-minute window."
+      echo "# TYPE browserless_rejected gauge"
+      echo "browserless_rejected $(printf '%s' "$latest" | jq '.rejected // 0')"
+
+      echo "# HELP browserless_errors Sessions that errored in the most recent 5-minute window."
+      echo "# TYPE browserless_errors gauge"
+      echo "browserless_errors $(printf '%s' "$latest" | jq '.error // 0')"
+
+      echo "# HELP browserless_timedout Sessions that timed out in the most recent 5-minute window."
+      echo "# TYPE browserless_timedout gauge"
+      echo "browserless_timedout $(printf '%s' "$latest" | jq '.timedout // 0')"
+
+      echo "# HELP browserless_unauthorized Unauthorized session requests (bad/missing token) in the most recent 5-minute window."
+      echo "# TYPE browserless_unauthorized gauge"
+      echo "browserless_unauthorized $(printf '%s' "$latest" | jq '.unauthorized // 0')"
+
+      echo "# HELP browserless_unhealthy Sessions marked unhealthy in the most recent 5-minute window."
+      echo "# TYPE browserless_unhealthy gauge"
+      echo "browserless_unhealthy $(printf '%s' "$latest" | jq '.unhealthy // 0')"
+
+      echo "# HELP browserless_successful Sessions that completed successfully in the most recent 5-minute window."
+      echo "# TYPE browserless_successful gauge"
+      echo "browserless_successful $(printf '%s' "$latest" | jq '.successful // 0')"
+
+      echo "# HELP browserless_session_mean_time_ms Mean session duration in milliseconds over the most recent 5-minute window."
+      echo "# TYPE browserless_session_mean_time_ms gauge"
+      echo "browserless_session_mean_time_ms $(printf '%s' "$latest" | jq '.meanTime // 0')"
+
+      echo "# HELP browserless_cpu_ratio Host CPU utilization ratio (0-1) sampled at the most recent window."
+      echo "# TYPE browserless_cpu_ratio gauge"
+      echo "browserless_cpu_ratio $(printf '%s' "$latest" | jq '.cpu // 0')"
+
+      echo "# HELP browserless_memory_ratio Host memory utilization ratio (0-1) sampled at the most recent window."
+      echo "# TYPE browserless_memory_ratio gauge"
+      echo "browserless_memory_ratio $(printf '%s' "$latest" | jq '.memory // 0')"
+
+      echo "# HELP browserless_sample_timestamp_seconds Unix timestamp browserless itself assigned to the most recent window sample."
+      echo "# TYPE browserless_sample_timestamp_seconds gauge"
+      echo "browserless_sample_timestamp_seconds $(printf '%s' "$latest" | jq '(.date // 0) / 1000')"
+    else
+      echo "browserless_exporter_last_scrape_success 0"
+      echo "browserless_exporter_last_scrape_timestamp_seconds $now"
+    fi
+  } > "$tmp"
+  mv "$tmp" "$FILE"
+}
+
+if [ "${BROWSERLESS_METRICS_ONCE:-}" = "1" ]; then
+  write_metrics
+  exit 0
+fi
+
+write_metrics
+httpd -f -p "$PORT" -h "$OUT" &
+server=$!
+trap 'kill "$server" 2>/dev/null || true' INT TERM EXIT
+
+while true; do
+  sleep "$INTERVAL"
+  write_metrics
+done
diff --git a/test/unit/selfhost-compose-resource-limits.test.ts b/test/unit/selfhost-compose-resource-limits.test.ts
index 3e58b169d7..501a1e5f8d 100644
--- a/test/unit/selfhost-compose-resource-limits.test.ts
+++ b/test/unit/selfhost-compose-resource-limits.test.ts
@@ -32,6 +32,7 @@ describe("docker-compose.yml — per-service memory limits (#1828, #2495, #3893)
     "node-exporter": "${NODE_EXPORTER_MEM_LIMIT:-128m}",
     cadvisor: "${CADVISOR_MEM_LIMIT:-256m}",
     "redis-exporter": "${REDIS_EXPORTER_MEM_LIMIT:-64m}",
+    "browserless-exporter": "${BROWSERLESS_EXPORTER_MEM_LIMIT:-64m}",
   };
 
   it("caps the core app and every heavyweight optional service with an operator-overridable memory limit", () => {
@@ -67,6 +68,7 @@ describe("docker-compose.yml — per-service memory limits (#1828, #2495, #3893)
       "NODE_EXPORTER_MEM_LIMIT",
       "CADVISOR_MEM_LIMIT",
       "REDIS_EXPORTER_MEM_LIMIT",
+      "BROWSERLESS_EXPORTER_MEM_LIMIT",
     ]) {
       expect(env, key).toContain(key);
     }
diff --git a/test/unit/selfhost-grafana-browserless-dashboard.test.ts b/test/unit/selfhost-grafana-browserless-dashboard.test.ts
new file mode 100644
index 0000000000..aaa865b9c4
--- /dev/null
+++ b/test/unit/selfhost-grafana-browserless-dashboard.test.ts
@@ -0,0 +1,79 @@
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+import { describe, expect, it } from "vitest";
+
+type DashboardTarget = { expr?: string; legendFormat?: string };
+type DashboardPanel = {
+  id?: number;
+  title?: string;
+  type?: string;
+  description?: string;
+  datasource?: { type?: string; uid?: string };
+  targets?: DashboardTarget[];
+};
+type Dashboard = {
+  uid: string;
+  title: string;
+  tags: string[];
+  panels: DashboardPanel[];
+};
+
+const dashboardPath = join(process.cwd(), "grafana/dashboards/browserless-metrics.json");
+
+function readDashboard(): Dashboard {
+  return JSON.parse(readFileSync(dashboardPath, "utf8")) as Dashboard;
+}
+
+function allTargets(dashboard = readDashboard()): DashboardTarget[] {
+  return dashboard.panels.flatMap((panel) => panel.targets ?? []);
+}
+
+describe("Gittensory — Browserless (visual review) dashboard (#5368)", () => {
+  it("declares the expected uid/title/tags", () => {
+    const dashboard = readDashboard();
+    expect(dashboard.uid).toBe("gittensory-browserless");
+    expect(dashboard.title).toBe("Gittensory — Browserless (visual review)");
+    expect(dashboard.tags).toEqual(["gittensory", "browserless", "observability"]);
+  });
+
+  it("every panel target uses the Prometheus datasource variable, never a hardcoded uid", () => {
+    const dashboard = readDashboard();
+    for (const panel of dashboard.panels) {
+      if (panel.type === "row") continue;
+      expect(panel.datasource?.type, panel.title).toBe("prometheus");
+      expect(panel.datasource?.uid, panel.title).toBe("${DS_PROMETHEUS}");
+      for (const target of panel.targets ?? []) {
+        expect(target.expr, panel.title).toBeTruthy();
+      }
+    }
+  });
+
+  it("uses the real metric names browserless-metrics.sh emits", () => {
+    const targets = allTargets();
+    const expectedMetrics = [
+      "browserless_exporter_last_scrape_success",
+      "browserless_sample_timestamp_seconds",
+      "browserless_queued",
+      "browserless_running",
+      "browserless_max_concurrent",
+      "browserless_rejected",
+      "browserless_errors",
+      "browserless_timedout",
+      "browserless_unauthorized",
+      "browserless_unhealthy",
+      "browserless_successful",
+      "browserless_session_mean_time_ms",
+      "browserless_cpu_ratio",
+      "browserless_memory_ratio",
+    ];
+    for (const metric of expectedMetrics) {
+      expect(targets.some((t) => t.expr?.includes(metric)), metric).toBe(true);
+    }
+  });
+
+  it("surfaces exporter-scrape health separately from browserless's own sample freshness", () => {
+    const targets = allTargets();
+    expect(targets.some((t) => t.expr === "browserless_exporter_last_scrape_success")).toBe(true);
+    expect(targets.some((t) => t.expr === "time() - browserless_sample_timestamp_seconds")).toBe(true);
+  });
+});
diff --git a/test/unit/selfhost-observability-config.test.ts b/test/unit/selfhost-observability-config.test.ts
index d4d578fd43..09108994c9 100644
--- a/test/unit/selfhost-observability-config.test.ts
+++ b/test/unit/selfhost-observability-config.test.ts
@@ -271,4 +271,52 @@ describe("self-host observability trace config", () => {
       ]),
     );
   });
+
+  it("translates browserless's JSON /metrics into Prometheus text via a sidecar, gated on --profile visual-review (#5368)", () => {
+    const compose = record(readYaml("docker-compose.yml"));
+    const services = record(compose.services);
+    const browserlessExporter = record(services["browserless-exporter"]);
+    const prometheus = record(readYaml("prometheus/prometheus.yml"));
+    const scrapeConfigs = prometheus.scrape_configs as Array<Record<string, any>>;
+
+    expect(browserlessExporter.image).toBe("alpine:3.20");
+    expect(browserlessExporter.profiles).toEqual(["visual-review"]);
+    expect(browserlessExporter.depends_on?.browserless).toEqual({ condition: "service_healthy" });
+    expect(browserlessExporter.environment).toMatchObject({
+      BROWSERLESS_METRICS_URL: "http://browserless:3000/metrics",
+      BROWSERLESS_TOKEN: "${BROWSERLESS_TOKEN:-}",
+    });
+    expect(browserlessExporter.expose).toEqual(["9102"]);
+    expect(browserlessExporter.command).toEqual([
+      "/bin/sh",
+      "-c",
+      "apk add --no-cache jq busybox-extras >/dev/null 2>&1 && sh /scripts/browserless-metrics.sh",
+    ]);
+    expect(browserlessExporter.healthcheck?.test).toEqual([
+      "CMD-SHELL",
+      "wget -qO- http://127.0.0.1:9102/metrics | grep -q '^browserless_exporter_last_scrape_success'",
+    ]);
+
+    expect(scrapeConfigs).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          job_name: "browserless",
+          fallback_scrape_protocol: "PrometheusText0.0.4",
+          static_configs: [{ targets: ["browserless-exporter:9102"] }],
+        }),
+      ]),
+    );
+  });
+
+  it("ships the browserless-metrics.sh translator with the newest-window-wins + scrape-failure-safe shape it claims", () => {
+    const script = readFileSync(join(process.cwd(), "scripts/browserless-metrics.sh"), "utf8");
+
+    // Picks the LAST array element (browserless's rolling history is oldest-first), not the first or a merge.
+    expect(script).toContain("jq -c '.[-1] // empty'");
+    // A failed poll must not corrupt/blank the previously-served file: written atomically via tmp+mv, and the
+    // failure branch still emits a valid, parseable metrics document (just success=0) rather than an empty body.
+    expect(script).toContain('mv "$tmp" "$FILE"');
+    expect(script).toContain("browserless_exporter_last_scrape_success 0");
+    expect(script).toContain("browserless_exporter_last_scrape_success 1");
+  });
 });