diff --git a/scripts/fleet-health-monitor.sh b/scripts/fleet-health-monitor.sh
new file mode 100644
index 0000000..49eccda
--- /dev/null
+++ b/scripts/fleet-health-monitor.sh
@@ -0,0 +1,27 @@
+#!/bin/bash
+# Fleet crash-loop / dead-worker detector (Erik). Reads v_fleet_health (fed by the
+# Pi heartbeats in fleet_health) and alerts on verdict<>ok. Closes the blind spot
+# where a crash-looping pm2 process reports "online" and a naive liveness check
+# misses it — the 3-Pi worker fleet was dead ~months (2026-07) exactly this way.
+# Cron: */10. Alert = log + optional ntfy push (FLEET_NTFY_URL).
+set -uo pipefail
+LOG=/opt/tip/logs/fleet-health.log
+mkdir -p "$(dirname "$LOG")"
+TS=$(date '+%Y-%m-%d %H:%M:%S')
+
+# shellcheck disable=SC1091
+source /opt/tip/.env 2>/dev/null || true
+Q="select host||' '||pm_name||' -> '||verdict||' (restart_delta='||restart_delta||', uptime='||uptime_s||'s, age='||age_s||'s)' from v_fleet_health where verdict <> 'ok' order by host"
+ALERTS=$(docker exec -e PGPASSWORD="${POSTGRES_PASSWORD:-}" tip-postgres psql -U tip -d transceiver_db -tA -c "$Q" 2>/dev/null)
+
+if [ -n "$ALERTS" ]; then
+  N=$(printf '%s\n' "$ALERTS" | grep -c .)
+  echo "[$TS] FLEET ALERT ($N):" >> "$LOG"
+  printf '%s\n' "$ALERTS" | sed 's/^/[    ] /' >> "$LOG"
+  # best-effort push (set FLEET_NTFY_URL in /opt/tip/.env to enable, e.g. https://ntfy.sh/tip-fleet)
+  if [ -n "${FLEET_NTFY_URL:-}" ]; then
+    printf 'TIP fleet alert (%s):\n%s\n' "$N" "$ALERTS" | curl -s -m 8 -H "Title: TIP fleet crash-loop/dead" -d @- "$FLEET_NTFY_URL" >/dev/null 2>&1 || true
+  fi
+else
+  echo "[$TS] fleet ok" >> "$LOG"
+fi
diff --git a/scripts/fleet-heartbeat.mjs b/scripts/fleet-heartbeat.mjs
new file mode 100644
index 0000000..7639a59
--- /dev/null
+++ b/scripts/fleet-heartbeat.mjs
@@ -0,0 +1,37 @@
+/**
+ * Pi/host fleet heartbeat: self-report local pm2 process health to Postgres so a
+ * central detector on Erik can catch crash-loops (restart_time velocity + near-0
+ * uptime) and stale/dead hosts — which a naive "pm2 status == online" check misses.
+ * Runs on each Pi via cron (every 5 min). DB via the local tunnel (127.0.0.1:5433).
+ */
+import { execFileSync } from "node:child_process";
+import os from "node:os";
+import { config } from "dotenv";
+import pg from "pg";
+
+config({ path: new URL("../.env", import.meta.url) }); // ~/tip/.env
+
+const host = os.hostname();
+let procs = [];
+try { procs = JSON.parse(execFileSync("pm2", ["jlist"], { encoding: "utf8", maxBuffer: 8 * 1024 * 1024 })); }
+catch { process.exit(0); }
+
+const pool = new pg.Pool({
+  host: "127.0.0.1", port: 5433,
+  user: process.env.POSTGRES_USER || "tip",
+  database: process.env.POSTGRES_DB || "transceiver_db",
+  password: process.env.POSTGRES_PASSWORD,
+  max: 1, connectionTimeoutMillis: 8000, statement_timeout: 8000,
+});
+
+try {
+  for (const p of procs) {
+    const e = p.pm2_env || {};
+    const uptimeS = Math.round((Date.now() - (e.pm_uptime || Date.now())) / 1000);
+    await pool.query(
+      "insert into fleet_health(host,pm_name,status,restart_time,uptime_s) values($1,$2,$3,$4,$5)",
+      [host, p.name, e.status || null, e.restart_time ?? 0, uptimeS],
+    );
+  }
+} catch { /* heartbeat is best-effort; never crash the host */ }
+finally { await pool.end().catch(() => {}); }