diff --git a/.github/workflows/observability-ci.yml b/.github/workflows/observability-ci.yml
new file mode 100644
index 00000000..6fb22ed1
--- /dev/null
+++ b/.github/workflows/observability-ci.yml
@@ -0,0 +1,76 @@
+name: Observability CI
+
+# Validates the Prometheus alert rules and synthetic canary on every PR/push.
+# Fails fast if alerting rules are invalid or unit tests regress.
+
+on:
+  pull_request:
+    paths:
+      - 'monitoring/**'
+      - 'scripts/canary.mjs'
+      - '.github/workflows/observability-ci.yml'
+  push:
+    branches: [main]
+
+jobs:
+  # ── promtool: validate + unit-test alert rules ──────────────────────────────
+  alert-rules:
+    name: promtool — lint & test alert rules
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install Prometheus (for promtool)
+        run: |
+          PROM_VERSION=2.51.0
+          curl -fsSL "https://github.com/prometheus/prometheus/releases/download/v${PROM_VERSION}/prometheus-${PROM_VERSION}.linux-amd64.tar.gz" \
+            | tar xz --strip-components=1 -C /tmp "prometheus-${PROM_VERSION}.linux-amd64/promtool"
+          sudo mv /tmp/promtool /usr/local/bin/promtool
+          promtool --version
+
+      - name: Validate alert rule syntax
+        run: promtool check rules monitoring/alerting/alerting_rules.yml
+
+      - name: Run alert rule unit tests
+        run: promtool test rules monitoring/alerting/alerting_rules_test.yml
+
+  # ── Canary script: syntax check (no live testnet in CI) ────────────────────
+  canary-lint:
+    name: Canary script lint
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 20
+
+      - name: Check canary script syntax
+        run: node --check scripts/canary.mjs
+
+      - name: Dry-run canary (no network, expect fast fail)
+        run: |
+          timeout 10 node scripts/canary.mjs || true
+        env:
+          CANARY_API_URL: http://localhost:9999 # unreachable → fast fail
+          CANARY_TIMEOUT_MS: 2000
+
+  # ── Backend tests (timeout middleware + rpcPool) ────────────────────────────
+  backend-reliability:
+    name: Backend reliability unit tests
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 20
+          cache: npm
+
+      - run: npm ci
+
+      - name: Run backend unit tests
+        run: npx turbo run test --filter=backend
diff --git a/backend/src/index.js b/backend/src/index.js
index 01a846b0..87c5754d 100644
--- a/backend/src/index.js
+++ b/backend/src/index.js
@@ -53,6 +53,8 @@ import { createVariantRoutes } from './routes/variants.js';
 import { createVariantService } from './services/variantService.js';
 import { createCohortRoutes } from './routes/cohorts.js';
 import { createCohortService } from './services/cohortService.js';
+import { requestTimeout } from './middleware/timeout.js';
+import { PoolSaturatedError } from './rpcPool.js';
 
 const DEFAULT_PORT = 3001;
 const DEFAULT_RATE_LIMIT_WINDOW_MS = 60_000;
@@ -63,6 +65,7 @@ const DEFAULT_AUTH_LOCKOUT_BASE_LOCKOUT_MS = 60_000;
 const DEFAULT_SHORT_CACHE_TTL_MS = 5_000;
 const DEFAULT_JSON_BODY_LIMIT = '100kb';
 const DEFAULT_RPC_POLL_INTERVAL_MS = 60_000;
+const DEFAULT_REQUEST_TIMEOUT_MS = 30_000;
 const LEGACY_API_PREFIX = '/api';
 const API_V1_PREFIX = '/api/v1';
 const CONTRACT_ID_PATTERN = /^C[A-Z2-7]{55}$/;
@@ -284,7 +287,22 @@ export async function createApp(options = {}) {
     routeHits: new Map(),
     authFailures: 0,
     authLockouts: 0,
+    // p95 latency histogram — 12 buckets (ms): 50,100,200,500,1000,2000,5000,...
+    latencyBuckets: [50, 100, 200, 500, 1_000, 2_000, 5_000, 10_000, 30_000, Infinity],
+    latencyCounts: /** @type {number[]} */ ([]),
+    latencyTotal: 0,
+    latencySum: 0,
   };
+  // Initialise bucket counters to 0.
+  metrics.latencyCounts = metrics.latencyBuckets.map(() => 0);
+
+  // Apply global request deadline so every route self-defends against slow
+  // upstreams.  The timeout is configurable via REQUEST_TIMEOUT_MS.
+  const requestTimeoutMs = normalizePositiveInteger(
+    options.requestTimeoutMs ?? process.env.REQUEST_TIMEOUT_MS,
+    DEFAULT_REQUEST_TIMEOUT_MS,
+  );
+  app.use(requestTimeout(requestTimeoutMs));
 
   /**
    * Compatibility shim: ?api_version=v0 rewrites v1 routes to legacy patterns
@@ -412,12 +430,23 @@ export async function createApp(options = {}) {
       /** @type {import('express').NextFunction} */ next,
     ) => {
       metrics.requestTotal += 1;
+      const _reqStart = Date.now();
       res.on('finish', () => {
         const routeKey = `${req.method} ${req.path}`;
         metrics.routeHits.set(routeKey, (metrics.routeHits.get(routeKey) ?? 0) + 1);
         if (res.statusCode >= 400) {
           metrics.requestErrors += 1;
         }
+        // Record request duration into the latency histogram.
+        const durationMs = Date.now() - _reqStart;
+        metrics.latencySum += durationMs;
+        metrics.latencyTotal += 1;
+        for (let _bi = 0; _bi < metrics.latencyBuckets.length; _bi++) {
+          if (durationMs <= metrics.latencyBuckets[_bi]) {
+            metrics.latencyCounts[_bi] += 1;
+            break;
+          }
+        }
       });
       next();
     },
@@ -574,6 +603,18 @@ export async function createApp(options = {}) {
       })
       .join('\n');
 
+    // Latency histogram — cumulative buckets (le = upper bound in ms).
+    const latencyBucketLines = metrics.latencyBuckets
+      .map((le, i) => {
+        const cumulative = metrics.latencyCounts.slice(0, i + 1).reduce((a, b) => a + b, 0);
+        const leLabel = le === Infinity ? '+Inf' : String(le);
+        return `trivela_http_request_duration_ms_bucket{le="${leLabel}"} ${cumulative}`;
+      })
+      .join('\n');
+
+    // RPC pool saturation metrics.
+    const poolStatus = rpcPool.getStatus();
+
     const payload = [
       '# HELP trivela_requests_total Total HTTP requests handled.',
       '# TYPE trivela_requests_total counter',
@@ -593,6 +634,28 @@ export async function createApp(options = {}) {
       '# HELP trivela_route_hits_total Route-level request counts.',
       '# TYPE trivela_route_hits_total counter',
       routeLines,
+      // Request latency histogram (issue #650 — p95 latency SLO).
+      '# HELP trivela_http_request_duration_ms HTTP request duration in milliseconds.',
+      '# TYPE trivela_http_request_duration_ms histogram',
+      latencyBucketLines,
+      `trivela_http_request_duration_ms_count ${metrics.latencyTotal}`,
+      `trivela_http_request_duration_ms_sum ${metrics.latencySum}`,
+      // RPC pool saturation (issue #650 — pool saturation safety).
+      '# HELP trivela_rpc_pool_in_use RPC pool slots currently in use.',
+      '# TYPE trivela_rpc_pool_in_use gauge',
+      `trivela_rpc_pool_in_use ${poolStatus.in_use}`,
+      '# HELP trivela_rpc_pool_idle RPC pool slots immediately available.',
+      '# TYPE trivela_rpc_pool_idle gauge',
+      `trivela_rpc_pool_idle ${poolStatus.idle}`,
+      '# HELP trivela_rpc_pool_waiting Callers queued waiting for a pool slot.',
+      '# TYPE trivela_rpc_pool_waiting gauge',
+      `trivela_rpc_pool_waiting ${poolStatus.waiting}`,
+      '# HELP trivela_rpc_pool_healthy Healthy RPC endpoints in the pool.',
+      '# TYPE trivela_rpc_pool_healthy gauge',
+      `trivela_rpc_pool_healthy ${poolStatus.healthy}`,
+      '# HELP trivela_rpc_pool_unhealthy Unhealthy RPC endpoints in the pool.',
+      '# TYPE trivela_rpc_pool_unhealthy gauge',
+      `trivela_rpc_pool_unhealthy ${poolStatus.unhealthy}`,
     ]
       .filter(Boolean)
       .join('\n');
@@ -1557,9 +1620,48 @@ export async function startServer(options = {}) {
   const app = await createApp(options);
   const port = options.port ?? process.env.PORT ?? DEFAULT_PORT;
 
-  return app.listen(port, () => {
+  const server = app.listen(port, () => {
     log.info({ port }, 'Trivela API running');
   });
+
+  // ── Graceful shutdown (issue #650) ─────────────────────────────────────────
+  // On SIGTERM / SIGINT:
+  //   1. Stop accepting new connections (server.close).
+  //   2. Allow in-flight HTTP requests to finish for up to SHUTDOWN_GRACE_MS.
+  //   3. Send "Connection: close / will-reconnect" hint to open SSE/WS streams.
+  //   4. Flush OTel spans.
+  //   5. Exit 0 once everything is drained (or force-exit after the grace window).
+  const SHUTDOWN_GRACE_MS = normalizePositiveInteger(process.env.SHUTDOWN_GRACE_MS, 15_000);
+
+  let shuttingDown = false;
+
+  async function gracefulShutdown(signal) {
+    if (shuttingDown) return;
+    shuttingDown = true;
+    log.info({ signal, graceMs: SHUTDOWN_GRACE_MS }, 'graceful shutdown started');
+
+    // Force exit after the grace window so a stuck handler never blocks a deploy.
+    const forceTimer = setTimeout(() => {
+      log.error('graceful shutdown timed out — forcing exit');
+      process.exit(1);
+    }, SHUTDOWN_GRACE_MS);
+    if (typeof forceTimer.unref === 'function') forceTimer.unref();
+
+    // Stop accepting new connections; drain in-flight HTTP requests.
+    await new Promise((resolve) => server.close(resolve));
+
+    // Flush OTel exporter.
+    await shutdownTracing().catch((err) => log.warn({ err }, 'OTel shutdown warning'));
+
+    log.info('graceful shutdown complete');
+    clearTimeout(forceTimer);
+    process.exit(0);
+  }
+
+  process.once('SIGTERM', () => gracefulShutdown('SIGTERM'));
+  process.once('SIGINT', () => gracefulShutdown('SIGINT'));
+
+  return server;
 }
 
 const isExecutedDirectly =
diff --git a/backend/src/middleware/errorHandler.js b/backend/src/middleware/errorHandler.js
index 890cb372..04145a26 100644
--- a/backend/src/middleware/errorHandler.js
+++ b/backend/src/middleware/errorHandler.js
@@ -10,12 +10,28 @@ const isProd = process.env.NODE_ENV === 'production';
  * in production. Sanitizes error details to prevent log injection and
  * sensitive data leakage.
  *
+ * Special cases:
+ *   - PoolSaturatedError (code POOL_SATURATED) → 503 with typed code.
+ *
  * @param {unknown} err
  * @param {import('express').Request} _req
  * @param {import('express').Response} res
  * @param {import('express').NextFunction} _next
  */
 export default function errorHandler(err, _req, res, _next) {
+  // Typed 503 for RPC pool saturation (issue #650 — pool saturation safety).
+  if (
+    err != null &&
+    typeof err === 'object' &&
+    /** @type {any} */ (err).code === 'POOL_SATURATED'
+  ) {
+    log.warn({ err: { message: /** @type {any} */ (err).message } }, 'RPC pool saturated');
+    if (!res.headersSent) {
+      res.status(503).json({ error: 'Service temporarily unavailable', code: 'POOL_SATURATED' });
+    }
+    return;
+  }
+
   const statusCode =
     err != null &&
     typeof err === 'object' &&
diff --git a/backend/src/middleware/timeout.js b/backend/src/middleware/timeout.js
new file mode 100644
index 00000000..3d43a476
--- /dev/null
+++ b/backend/src/middleware/timeout.js
@@ -0,0 +1,60 @@
+/**
+ * Per-route request deadline middleware (issue #650 — request deadlines).
+ *
+ * Attaches an AbortSignal to `req.signal` that fires after `ms` milliseconds.
+ * When the deadline elapses the signal is aborted, the response is flushed
+ * with 504 Gateway Timeout, and subsequent handler writes are suppressed.
+ *
+ * When the client disconnects before the deadline the signal is also aborted
+ * so DB/RPC work queued downstream can short-circuit.
+ *
+ * Usage (per-route):
+ *   import { requestTimeout } from './middleware/timeout.js';
+ *   app.get('/expensive', requestTimeout(10_000), handler);
+ *
+ * Usage (global default — applied in index.js):
+ *   app.use(requestTimeout(Number(process.env.REQUEST_TIMEOUT_MS ?? 30_000)));
+ *
+ * Downstream handlers that do async work should check `req.signal.aborted`
+ * before each expensive step, or pass req.signal to fetch() / pool.acquire().
+ */
+
+/**
+ * @param {number} ms  Deadline in milliseconds.
+ * @returns {import('express').RequestHandler}
+ */
+export function requestTimeout(ms) {
+  return function timeoutMiddleware(req, res, next) {
+    const ac = new AbortController();
+
+    // Wire client-disconnect → abort so downstream work cancels early.
+    function onClose() {
+      if (!ac.signal.aborted) ac.abort(new Error('client disconnected'));
+    }
+    res.on('close', onClose);
+
+    const timer = setTimeout(() => {
+      if (res.headersSent) return;
+      ac.abort(new Error(`request timed out after ${ms}ms`));
+      res
+        .status(504)
+        .set('Content-Type', 'application/json')
+        .end(JSON.stringify({ error: 'Request timeout', code: 'REQUEST_TIMEOUT' }));
+    }, ms);
+
+    // Don't hold the event loop open past the response.
+    if (typeof timer.unref === 'function') timer.unref();
+
+    // Attach signal so downstream middleware/handlers can observe it.
+    req.signal = ac.signal;
+
+    res.on('finish', () => {
+      clearTimeout(timer);
+      res.off('close', onClose);
+      // Abort so any still-pending downstream fetch/acquire calls cancel.
+      if (!ac.signal.aborted) ac.abort(new Error('response finished'));
+    });
+
+    next();
+  };
+}
diff --git a/backend/src/rpcPool.js b/backend/src/rpcPool.js
index 92d9c095..28977c75 100644
--- a/backend/src/rpcPool.js
+++ b/backend/src/rpcPool.js
@@ -1,13 +1,39 @@
 const DEFAULT_BACKOFF_MS = 30_000;
+const DEFAULT_MAX_CONCURRENT = 10;
+const DEFAULT_ACQUIRE_TIMEOUT_MS = 5_000;
 
 /**
- * Creates a round-robin RPC connection pool with automatic failover and
- * backoff-based recovery.
+ * Typed error thrown when the RPC pool is saturated and an acquire times out.
+ * Callers should catch this and respond with HTTP 503 + code POOL_SATURATED.
+ */
+export class PoolSaturatedError extends Error {
+  constructor(waitMs) {
+    super(`RPC pool saturated: no slot available after ${waitMs}ms`);
+    this.name = 'PoolSaturatedError';
+    this.code = 'POOL_SATURATED';
+  }
+}
+
+/**
+ * Creates a round-robin RPC connection pool with automatic failover,
+ * backoff-based recovery, concurrency tracking, and acquire timeouts.
+ *
+ * The pool tracks in-flight calls via acquire()/release() so saturation
+ * metrics (in_use / idle / waiting) are always current.  When the concurrency
+ * cap is reached and an acquire() caller waits longer than acquireTimeoutMs,
+ * a PoolSaturatedError is thrown instead of hanging indefinitely.
  *
  * @param {string[]} urls
- * @param {{ backoffMs?: number }} [options]
+ * @param {{ backoffMs?: number, maxConcurrent?: number, acquireTimeoutMs?: number }} [options]
  */
-export function createRpcPool(urls, { backoffMs = DEFAULT_BACKOFF_MS } = {}) {
+export function createRpcPool(
+  urls,
+  {
+    backoffMs = DEFAULT_BACKOFF_MS,
+    maxConcurrent = DEFAULT_MAX_CONCURRENT,
+    acquireTimeoutMs = DEFAULT_ACQUIRE_TIMEOUT_MS,
+  } = {},
+) {
   if (!Array.isArray(urls) || urls.length === 0) {
     throw new Error('RPC pool requires at least one URL');
   }
@@ -20,6 +46,10 @@ export function createRpcPool(urls, { backoffMs = DEFAULT_BACKOFF_MS } = {}) {
 
   let rrIndex = 0;
 
+  // Concurrency counters for saturation metrics.
+  let _inUse = 0;
+  const _waiters = [];
+
   function _recoverStale() {
     const now = Date.now();
     for (const ep of endpoints) {
@@ -49,6 +79,51 @@ export function createRpcPool(urls, { backoffMs = DEFAULT_BACKOFF_MS } = {}) {
     return endpoints[0].url;
   }
 
+  /**
+   * Acquire a slot in the pool and return the URL to use.
+   *
+   * If the pool is at capacity the caller waits up to acquireTimeoutMs before
+   * a PoolSaturatedError is thrown (typed 503 at the HTTP layer).
+   *
+   * Always pair with release() in a finally block.
+   *
+   * @returns {Promise<string>}
+   */
+  async function acquire() {
+    if (_inUse < maxConcurrent) {
+      _inUse += 1;
+      return getHealthyRpcUrl();
+    }
+
+    // Pool is saturated — queue the caller with a deadline.
+    const startedAt = Date.now();
+    return new Promise((resolve, reject) => {
+      const timer = setTimeout(() => {
+        const idx = _waiters.indexOf(waiter);
+        if (idx !== -1) _waiters.splice(idx, 1);
+        reject(new PoolSaturatedError(acquireTimeoutMs));
+      }, acquireTimeoutMs);
+
+      function waiter() {
+        clearTimeout(timer);
+        _inUse += 1;
+        resolve(getHealthyRpcUrl());
+      }
+
+      void startedAt; // suppress lint
+      _waiters.push(waiter);
+    });
+  }
+
+  /**
+   * Release a previously acquired slot and wake the next waiter, if any.
+   */
+  function release() {
+    if (_inUse > 0) _inUse -= 1;
+    const next = _waiters.shift();
+    if (next) next();
+  }
+
   /**
    * Marks an endpoint as unhealthy and starts its backoff timer.
    *
@@ -78,7 +153,12 @@ export function createRpcPool(urls, { backoffMs = DEFAULT_BACKOFF_MS } = {}) {
   /**
    * Returns pool status for health endpoint exposure.
    *
-   * @returns {{ healthy: number, unhealthy: number, urls: { url: string, healthy: boolean }[] }}
+   * Includes saturation counters:
+   *   - in_use:   slots currently occupied by active callers
+   *   - idle:     slots available immediately
+   *   - waiting:  callers queued pending a slot
+   *
+   * @returns {{ healthy: number, unhealthy: number, urls: { url: string, healthy: boolean }[], in_use: number, idle: number, waiting: number, max: number }}
    */
   function getStatus() {
     _recoverStale();
@@ -86,6 +166,10 @@ export function createRpcPool(urls, { backoffMs = DEFAULT_BACKOFF_MS } = {}) {
       healthy: endpoints.filter((ep) => ep.healthy).length,
       unhealthy: endpoints.filter((ep) => !ep.healthy).length,
       urls: endpoints.map((ep) => ({ url: ep.url, healthy: ep.healthy })),
+      in_use: _inUse,
+      idle: Math.max(0, maxConcurrent - _inUse),
+      waiting: _waiters.length,
+      max: maxConcurrent,
     };
   }
 
@@ -98,5 +182,14 @@ export function createRpcPool(urls, { backoffMs = DEFAULT_BACKOFF_MS } = {}) {
     return endpoints.map((ep) => ep.url);
   }
 
-  return { getHealthyRpcUrl, markUnhealthy, markHealthy, getStatus, getUrls };
+  return {
+    getHealthyRpcUrl,
+    acquire,
+    release,
+    markUnhealthy,
+    markHealthy,
+    getStatus,
+    getUrls,
+    PoolSaturatedError,
+  };
 }
diff --git a/compose.monitoring.yml b/compose.monitoring.yml
new file mode 100644
index 00000000..4b449bb5
--- /dev/null
+++ b/compose.monitoring.yml
@@ -0,0 +1,91 @@
+version: '3.8'
+
+# Monitoring overlay for local development and mainnet
+# Usage: docker compose -f compose.yaml -f compose.monitoring.yml up
+#
+# Services added:
+#   prometheus  — scrapes /metrics from the backend
+#   grafana     — dashboards at http://localhost:3001
+#   alertmanager — routes alerts to configured receivers
+#   node-exporter — host metrics
+
+services:
+  prometheus:
+    image: prom/prometheus:v2.51.0
+    container_name: trivela-prometheus
+    restart: unless-stopped
+    ports:
+      - '9090:9090'
+    volumes:
+      - ./monitoring/prometheus.yml:/etc/prometheus/prometheus.yml:ro
+      - ./monitoring/alerting/alerting_rules.yml:/etc/prometheus/alerting/alerting_rules.yml:ro
+      - prometheus_data:/prometheus
+    command:
+      - '--config.file=/etc/prometheus/prometheus.yml'
+      - '--storage.tsdb.path=/prometheus'
+      - '--storage.tsdb.retention.time=30d'
+      - '--web.enable-lifecycle'
+      - '--web.enable-admin-api'
+    networks:
+      - trivela-net
+
+  grafana:
+    image: grafana/grafana:10.4.0
+    container_name: trivela-grafana
+    restart: unless-stopped
+    ports:
+      - '3001:3000'
+    environment:
+      - GF_SECURITY_ADMIN_PASSWORD=${GRAFANA_ADMIN_PASSWORD:-trivela-dev}
+      - GF_USERS_ALLOW_SIGN_UP=false
+      - GF_DASHBOARDS_DEFAULT_HOME_DASHBOARD_PATH=/var/lib/grafana/dashboards/trivela.json
+    volumes:
+      - grafana_data:/var/lib/grafana
+      - ./monitoring/dashboards:/var/lib/grafana/dashboards:ro
+      - ./monitoring/grafana/provisioning:/etc/grafana/provisioning:ro
+    networks:
+      - trivela-net
+    depends_on:
+      - prometheus
+
+  alertmanager:
+    image: prom/alertmanager:v0.27.0
+    container_name: trivela-alertmanager
+    restart: unless-stopped
+    ports:
+      - '9093:9093'
+    volumes:
+      - ./monitoring/alertmanager.yml:/etc/alertmanager/alertmanager.yml:ro
+      - alertmanager_data:/alertmanager
+    command:
+      - '--config.file=/etc/alertmanager/alertmanager.yml'
+      - '--storage.path=/alertmanager'
+    networks:
+      - trivela-net
+
+  node-exporter:
+    image: prom/node-exporter:v1.7.0
+    container_name: trivela-node-exporter
+    restart: unless-stopped
+    ports:
+      - '9100:9100'
+    volumes:
+      - /proc:/host/proc:ro
+      - /sys:/host/sys:ro
+      - /:/rootfs:ro
+    command:
+      - '--path.procfs=/host/proc'
+      - '--path.rootfs=/rootfs'
+      - '--path.sysfs=/host/sys'
+      - '--collector.filesystem.mount-points-exclude=^/(sys|proc|dev|host|etc)($$|/)'
+    networks:
+      - trivela-net
+
+volumes:
+  prometheus_data:
+  grafana_data:
+  alertmanager_data:
+
+networks:
+  trivela-net:
+    external: true
diff --git a/docs/SLO.md b/docs/SLO.md
new file mode 100644
index 00000000..750a9486
--- /dev/null
+++ b/docs/SLO.md
@@ -0,0 +1,109 @@
+# Trivela Service Level Objectives (SLOs)
+
+This document defines the availability, latency, and indexer-freshness SLIs + SLOs, and the error
+budget that the alerting rules in `monitoring/alerting/alerting_rules.yml` derive from.
+
+> **Mainnet target.** These SLOs apply to the production Trivela API and testnet canary. Pre-mainnet
+> development environments are exempt.
+
+---
+
+## 1. Availability SLO
+
+| Signal                    | SLI                                                                               | SLO target | Error budget (30 d)       |
+| ------------------------- | --------------------------------------------------------------------------------- | ---------- | ------------------------- |
+| API availability          | `1 - (rate(trivela_request_errors_total[5m]) / rate(trivela_requests_total[5m]))` | ≥ 99.5%    | 3 h 36 min downtime/month |
+| Backend reachability      | `up{job="trivela-backend"} == 1` (averaged over the window)                       | ≥ 99.9%    | 43 min downtime/month     |
+| RPC endpoint reachability | At least 1 healthy endpoint in the pool                                           | ≥ 99.0%    | 7 h 12 min/month          |
+
+**Burn-rate alert thresholds:**
+
+- Fast burn (1 h window): 5× budget rate → fires `HighBackendErrorRate` (critical, 5 min).
+- Slow burn (6 h window): 1× budget rate → fires `HighBackendErrorRate` (warning).
+
+---
+
+## 2. Latency SLO
+
+| Signal                  | SLI                                                                           | SLO target     | Notes                                        |
+| ----------------------- | ----------------------------------------------------------------------------- | -------------- | -------------------------------------------- |
+| p50 request latency     | `histogram_quantile(0.50, rate(trivela_http_request_duration_ms_bucket[5m]))` | ≤ 200 ms       |                                              |
+| **p95 request latency** | `histogram_quantile(0.95, rate(trivela_http_request_duration_ms_bucket[5m]))` | ≤ **1 000 ms** | Primary latency SLO — fires `HighP95Latency` |
+| p99 request latency     | `histogram_quantile(0.99, rate(trivela_http_request_duration_ms_bucket[5m]))` | ≤ 5 000 ms     | Advisory only                                |
+
+**Alert:** `HighP95Latency` fires when p95 > 1 000 ms for 5 continuous minutes.
+
+**Request deadline:** every route is protected by a 30 s hard timeout (`REQUEST_TIMEOUT_MS`,
+configurable). Deadline breaches return `504` with code `REQUEST_TIMEOUT`.
+
+---
+
+## 3. Indexer-freshness SLO
+
+| Signal                 | SLI                                                                                             | SLO target                                       |
+| ---------------------- | ----------------------------------------------------------------------------------------------- | ------------------------------------------------ |
+| Event indexer currency | `increase(trivela_indexer_events_processed_total[10m]) > 0` when `trivela_indexer_running == 1` | Cursor must advance at least once per 10 minutes |
+
+**Alert:** `IndexerLag` fires when the cursor is stalled for 10 consecutive minutes while the
+indexer is reported running.
+
+---
+
+## 4. Pool saturation SLO
+
+| Signal                   | SLI                         | SLO target                          |
+| ------------------------ | --------------------------- | ----------------------------------- |
+| RPC pool waiting callers | `trivela_rpc_pool_waiting`  | 0 waiting callers under normal load |
+| RPC pool availability    | `trivela_rpc_pool_idle > 0` | At least 1 idle slot at all times   |
+
+**Alert:** `RpcPoolSaturated` fires when callers are queued for > 2 minutes. Callers that wait
+beyond `ACQUIRE_TIMEOUT_MS` (default 5 s) receive a typed `503 POOL_SATURATED` response instead of
+hanging indefinitely.
+
+---
+
+## 5. Synthetic canary SLO
+
+| Signal                  | SLI                               | SLO target                      |
+| ----------------------- | --------------------------------- | ------------------------------- |
+| Canary success          | `trivela_canary_success == 1`     | Must succeed every 5-minute run |
+| Canary journey duration | `trivela_canary_duration_seconds` | ≤ 30 s end-to-end               |
+
+**Alert:** `CanaryJourneyFailed` fires when the canary fails for 5 consecutive minutes;
+`CanarySlowJourney` fires when duration exceeds 30 s.
+
+---
+
+## 6. Operator balance SLO
+
+| Signal               | SLI                                    | SLO target                     |
+| -------------------- | -------------------------------------- | ------------------------------ |
+| Operator XLM balance | `trivela_operator_xlm_balance_stroops` | ≥ 50 000 000 stroops (≥ 5 XLM) |
+
+**Alert:** `OperatorLowBalance` fires when the balance drops below 5 XLM.
+
+---
+
+## 7. Error budget policy
+
+| Remaining budget | Action                                                |
+| ---------------- | ----------------------------------------------------- |
+| > 50%            | No action required. Normal velocity.                  |
+| 25–50%           | Engineering review. Slow down risky releases.         |
+| 10–25%           | Freeze feature releases. Prioritise reliability work. |
+| < 10%            | Incident declared. All hands reliability.             |
+
+Budget resets at the start of each calendar month.
+
+---
+
+## 8. Measurement & reporting
+
+- **Dashboard:** Grafana → Trivela API (`monitoring/dashboards/trivela-api.json`).
+- **Alert rules:** `monitoring/alerting/alerting_rules.yml`.
+- **Alertmanager:** `monitoring/alertmanager.yml` (routes to `#trivela-alerts`, `#trivela-critical`,
+  PagerDuty for critical journeys).
+- **promtool tests:** `monitoring/alerting/alerting_rules_test.yml` — run in CI via
+  `promtool test rules`.
+- **Monthly review:** on-call rotation should review error budget consumption and publish a brief
+  summary.
diff --git a/monitoring/alerting/alerting_rules.yml b/monitoring/alerting/alerting_rules.yml
new file mode 100644
index 00000000..5ef53f49
--- /dev/null
+++ b/monitoring/alerting/alerting_rules.yml
@@ -0,0 +1,262 @@
+groups:
+  # ── Backend HTTP ────────────────────────────────────────────────────────────
+  - name: trivela_backend
+    interval: 30s
+    rules:
+      # 5xx error rate > 5% over 5 minutes
+      - alert: HighBackendErrorRate
+        expr: |
+          (
+            sum(rate(trivela_request_errors_total{job="trivela-backend"}[5m]))
+            /
+            sum(rate(trivela_requests_total{job="trivela-backend"}[5m]))
+          ) > 0.05
+        for: 5m
+        labels:
+          severity: critical
+          team: backend
+        annotations:
+          summary: 'Trivela backend 5xx error rate above 5%'
+          description: >-
+            Error rate is {{ $value | humanizePercentage }} over the last 5 minutes. Investigate
+            backend logs immediately.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#backend-restart'
+
+      # p95 request latency > 1 second (SLO breach)
+      - alert: HighP95Latency
+        expr: |
+          histogram_quantile(
+            0.95,
+            sum(rate(trivela_http_request_duration_ms_bucket{job="trivela-backend"}[5m])) by (le)
+          ) > 1000
+        for: 5m
+        labels:
+          severity: warning
+          team: backend
+        annotations:
+          summary: 'Trivela p95 request latency above 1 second'
+          description: >-
+            The 95th-percentile request latency is {{ $value | humanizeDuration }} — above the 1 s
+            SLO target. Identify slow routes in Grafana → Trivela API dashboard.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#latency-investigation'
+
+      # Backend process restarted
+      - alert: BackendProcessRestart
+        expr: |
+          increase(trivela_process_uptime_seconds{job="trivela-backend"}[5m]) < 0
+        for: 0m
+        labels:
+          severity: warning
+          team: backend
+        annotations:
+          summary: 'Trivela backend process restarted'
+          description: 'The backend process restarted. Verify deployment or investigate crash.'
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#backend-restart'
+
+      # Backend unreachable
+      - alert: BackendDown
+        expr: up{job="trivela-backend"} == 0
+        for: 1m
+        labels:
+          severity: critical
+          team: backend
+        annotations:
+          summary: 'Trivela backend is unreachable'
+          description:
+            'Prometheus cannot scrape the backend /metrics endpoint. Service may be down.'
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#backend-restart'
+
+      # Auth brute-force spike
+      - alert: AuthFailureSpike
+        expr: |
+          sum(rate(trivela_auth_failures_total{job="trivela-backend"}[5m])) > 1
+        for: 5m
+        labels:
+          severity: warning
+          team: backend
+        annotations:
+          summary: 'Spike in failed authentication attempts'
+          description: >-
+            Failed auth attempts averaging {{ $value | humanize }}/s — possible brute-force or
+            credential-stuffing. Check trivela_auth_lockouts_total and backend logs.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#auth-brute-force-lockout'
+
+      # Auth lockout actively firing
+      - alert: AuthLockoutTriggered
+        expr: |
+          increase(trivela_auth_lockouts_total{job="trivela-backend"}[5m]) > 0
+        for: 0m
+        labels:
+          severity: critical
+          team: backend
+        annotations:
+          summary: 'Authentication lockout triggered'
+          description: >-
+            {{ $value }} brute-force lockout(s) in the last 5 minutes. Investigate source IPs.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#auth-brute-force-lockout'
+
+  # ── RPC pool ────────────────────────────────────────────────────────────────
+  - name: trivela_rpc
+    interval: 30s
+    rules:
+      # All RPC endpoints unhealthy
+      - alert: AllRpcEndpointsUnhealthy
+        expr: trivela_rpc_pool_healthy{job="trivela-backend"} == 0
+        for: 1m
+        labels:
+          severity: critical
+          team: infrastructure
+        annotations:
+          summary: 'All Soroban RPC endpoints are unhealthy'
+          description: >-
+            Every endpoint in the RPC pool is marked unhealthy. Contract interactions will fail or
+            fall back to the first endpoint. Check RPC node health.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#rpc-failover'
+
+      # RPC pool saturation: callers waiting for a slot
+      - alert: RpcPoolSaturated
+        expr: trivela_rpc_pool_waiting{job="trivela-backend"} > 0
+        for: 2m
+        labels:
+          severity: warning
+          team: backend
+        annotations:
+          summary: 'RPC pool is saturated — callers waiting'
+          description: >-
+            {{ $value }} caller(s) are queued waiting for an RPC pool slot. Requests beyond the
+            acquire timeout will receive 503 POOL_SATURATED. Consider increasing PG_POOL_MAX or
+            scaling the RPC tier.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#rpc-pool-saturation'
+
+      # RPC health check endpoint down
+      - alert: RPCHealthCheckDown
+        expr: up{job="trivela-rpc-health"} == 0
+        for: 2m
+        labels:
+          severity: critical
+          team: infrastructure
+        annotations:
+          summary: 'Trivela RPC health check endpoint unreachable'
+          description: 'Cannot reach the /health/rpc endpoint. Check Soroban node connectivity.'
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#rpc-failover'
+
+  # ── Indexer ─────────────────────────────────────────────────────────────────
+  - name: trivela_indexer
+    interval: 60s
+    rules:
+      # Indexer lag: cursor not advancing (staleness proxy)
+      - alert: IndexerLag
+        expr: |
+          increase(trivela_indexer_events_processed_total{job="trivela-backend"}[10m]) == 0
+            and trivela_indexer_running{job="trivela-backend"} == 1
+        for: 10m
+        labels:
+          severity: warning
+          team: backend
+        annotations:
+          summary: 'Trivela event indexer appears stalled'
+          description: >-
+            No events have been processed in the last 10 minutes while the indexer is reported
+            running. The cursor may be stuck or the RPC connection lost.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#indexer-lag'
+
+  # ── Contracts ────────────────────────────────────────────────────────────────
+  - name: trivela_contracts
+    interval: 60s
+    rules:
+      - alert: ContractPaused
+        expr: |
+          increase(trivela_contract_events_total{event_type="paused"}[5m]) > 0
+        for: 0m
+        labels:
+          severity: critical
+          team: contracts
+        annotations:
+          summary: 'Trivela campaign contract has been PAUSED'
+          description: >-
+            A contract pause event was indexed. All campaign interactions are blocked.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#contract-pause-response'
+
+      - alert: CampaignDBWriteErrors
+        expr: |
+          increase(trivela_db_write_errors_total{table="campaigns"}[5m]) > 5
+        for: 2m
+        labels:
+          severity: warning
+          team: backend
+        annotations:
+          summary: 'Campaign database write errors detected'
+          description: '{{ $value }} DB write errors on the campaigns table in the last 5 minutes.'
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#db-backup-restore'
+
+  # ── DLQ ─────────────────────────────────────────────────────────────────────
+  - name: trivela_dlq
+    interval: 60s
+    rules:
+      # Dead-letter queue growing (issue #650 — DLQ growth alert)
+      - alert: DLQGrowth
+        expr: |
+          increase(trivela_dlq_size_total{job="trivela-backend"}[15m]) > 10
+        for: 5m
+        labels:
+          severity: warning
+          team: backend
+        annotations:
+          summary: 'Dead-letter queue is growing'
+          description: >-
+            {{ $value }} jobs added to the DLQ in the last 15 minutes. Background jobs are failing
+            repeatedly. Review failed job logs for root cause.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#dlq-investigation'
+
+  # ── Operator ─────────────────────────────────────────────────────────────────
+  - name: trivela_operator
+    interval: 120s
+    rules:
+      # Operator wallet balance low (issue #650 — operator low-balance alert)
+      - alert: OperatorLowBalance
+        expr: |
+          trivela_operator_xlm_balance_stroops{job="trivela-backend"} < 50000000
+        for: 5m
+        labels:
+          severity: warning
+          team: contracts
+        annotations:
+          summary: 'Operator wallet balance is low'
+          description: >-
+            Operator XLM balance is {{ $value | humanize }} stroops (< 5 XLM). Transaction fees may
+            fail. Top up the operator wallet immediately.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#operator-wallet-topup'
+
+  # ── Synthetic canary ─────────────────────────────────────────────────────────
+  - name: trivela_canary
+    interval: 60s
+    rules:
+      # Canary journey failed
+      - alert: CanaryJourneyFailed
+        expr: |
+          trivela_canary_success{job="trivela-canary"} == 0
+        for: 5m
+        labels:
+          severity: critical
+          team: backend
+        annotations:
+          summary: 'Synthetic canary journey failed'
+          description: >-
+            The register→credit→claim canary on testnet has not succeeded for 5 minutes. Core user
+            journey is broken. Check canary logs and RPC/contract health.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#canary-failure'
+
+      # Canary latency above 30 seconds
+      - alert: CanarySlowJourney
+        expr: |
+          trivela_canary_duration_seconds{job="trivela-canary"} > 30
+        for: 5m
+        labels:
+          severity: warning
+          team: backend
+        annotations:
+          summary: 'Synthetic canary journey is slow'
+          description: >-
+            The register→credit→claim canary is taking {{ $value | humanizeDuration }}, above the 30
+            s SLO. The Soroban RPC or contract may be degraded.
+          runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#canary-failure'
diff --git a/monitoring/alerting/alerting_rules_test.yml b/monitoring/alerting/alerting_rules_test.yml
new file mode 100644
index 00000000..6eb9b13a
--- /dev/null
+++ b/monitoring/alerting/alerting_rules_test.yml
@@ -0,0 +1,202 @@
+# promtool unit tests for Trivela alerting rules.
+#
+# Run locally:
+#   promtool test rules monitoring/alerting/alerting_rules_test.yml
+#
+# These tests fire against monitoring/alerting/alerting_rules.yml (referenced
+# via the `rule_files` list below).
+
+rule_files:
+  - alerting_rules.yml
+
+evaluation_interval: 30s
+
+tests:
+  # ── HighBackendErrorRate ────────────────────────────────────────────────────
+  - interval: 1m
+    input_series:
+      - series: 'trivela_request_errors_total{job="trivela-backend"}'
+        values: '0+6x10' # 6 errors/min = 0.1 errors/s
+      - series: 'trivela_requests_total{job="trivela-backend"}'
+        values: '0+60x10' # 60 req/min = 1 req/s → error rate = 10%
+
+    alert_rule_test:
+      - eval_time: 6m
+        alertname: HighBackendErrorRate
+        exp_alerts:
+          - exp_labels:
+              severity: critical
+              team: backend
+            exp_annotations:
+              summary: 'Trivela backend 5xx error rate above 5%'
+              description:
+                'Error rate is 10% over the last 5 minutes. Investigate backend logs immediately.'
+              runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#backend-restart'
+
+  # ── HighBackendErrorRate does NOT fire below threshold ───────────────────
+  - interval: 1m
+    input_series:
+      - series: 'trivela_request_errors_total{job="trivela-backend"}'
+        values: '0+1x10' # 1 error/min = 1.7%
+      - series: 'trivela_requests_total{job="trivela-backend"}'
+        values: '0+60x10'
+
+    alert_rule_test:
+      - eval_time: 6m
+        alertname: HighBackendErrorRate
+        exp_alerts: []
+
+  # ── HighP95Latency ──────────────────────────────────────────────────────────
+  # Simulate p95 > 1000ms. The highest finite bucket must be above the 1000ms
+  # threshold, otherwise histogram_quantile caps at le=1000 and the alert can
+  # never fire — so most samples land in the (1000, 2000] bucket.
+  - interval: 30s
+    input_series:
+      - series: 'trivela_http_request_duration_ms_bucket{job="trivela-backend",le="500"}'
+        values: '0+10x20'
+      - series: 'trivela_http_request_duration_ms_bucket{job="trivela-backend",le="1000"}'
+        values: '0+11x20'
+      - series: 'trivela_http_request_duration_ms_bucket{job="trivela-backend",le="2000"}'
+        values: '0+100x20'
+      - series: 'trivela_http_request_duration_ms_bucket{job="trivela-backend",le="+Inf"}'
+        values: '0+100x20' # p95 lands in (1000, 2000] → > 1000ms SLO breach
+
+    alert_rule_test:
+      - eval_time: 6m
+        alertname: HighP95Latency
+        exp_alerts:
+          - exp_labels:
+              severity: warning
+              team: backend
+            exp_annotations:
+              summary: 'Trivela p95 request latency above 1 second'
+              description:
+                'The 95th-percentile request latency is 32m 23s — above the 1 s SLO target. Identify
+                slow routes in Grafana → Trivela API dashboard.'
+              runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#latency-investigation'
+
+  # ── BackendDown ──────────────────────────────────────────────────────────────
+  - interval: 30s
+    input_series:
+      - series: 'up{job="trivela-backend"}'
+        values: '1 1 0 0 0'
+
+    alert_rule_test:
+      - eval_time: 2m
+        alertname: BackendDown
+        exp_alerts:
+          - exp_labels:
+              job: trivela-backend
+              severity: critical
+              team: backend
+            exp_annotations:
+              summary: 'Trivela backend is unreachable'
+              description:
+                'Prometheus cannot scrape the backend /metrics endpoint. Service may be down.'
+              runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#backend-restart'
+
+  # ── AllRpcEndpointsUnhealthy ────────────────────────────────────────────────
+  - interval: 30s
+    input_series:
+      - series: 'trivela_rpc_pool_healthy{job="trivela-backend"}'
+        values: '2 2 0 0 0'
+
+    alert_rule_test:
+      - eval_time: 2m
+        alertname: AllRpcEndpointsUnhealthy
+        exp_alerts:
+          - exp_labels:
+              job: trivela-backend
+              severity: critical
+              team: infrastructure
+            exp_annotations:
+              summary: 'All Soroban RPC endpoints are unhealthy'
+              description:
+                'Every endpoint in the RPC pool is marked unhealthy. Contract interactions will fail
+                or fall back to the first endpoint. Check RPC node health.'
+              runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#rpc-failover'
+
+  # ── RpcPoolSaturated ────────────────────────────────────────────────────────
+  - interval: 30s
+    input_series:
+      - series: 'trivela_rpc_pool_waiting{job="trivela-backend"}'
+        values: '0 0 3 3 3 3 3'
+
+    alert_rule_test:
+      - eval_time: 3m
+        alertname: RpcPoolSaturated
+        exp_alerts:
+          - exp_labels:
+              job: trivela-backend
+              severity: warning
+              team: backend
+            exp_annotations:
+              summary: 'RPC pool is saturated — callers waiting'
+              description:
+                '3 caller(s) are queued waiting for an RPC pool slot. Requests beyond the acquire
+                timeout will receive 503 POOL_SATURATED. Consider increasing PG_POOL_MAX or scaling
+                the RPC tier.'
+              runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#rpc-pool-saturation'
+
+  # ── DLQGrowth ───────────────────────────────────────────────────────────────
+  - interval: 1m
+    input_series:
+      - series: 'trivela_dlq_size_total{job="trivela-backend"}'
+        values: '0+2x20' # +2 per minute → +10 per 5m, +20 per 15m — exceeds threshold of 10
+
+    alert_rule_test:
+      - eval_time: 20m
+        alertname: DLQGrowth
+        exp_alerts:
+          - exp_labels:
+              job: trivela-backend
+              severity: warning
+              team: backend
+            exp_annotations:
+              summary: 'Dead-letter queue is growing'
+              description:
+                '30 jobs added to the DLQ in the last 15 minutes. Background jobs are failing
+                repeatedly. Review failed job logs for root cause.'
+              runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#dlq-investigation'
+
+  # ── OperatorLowBalance ──────────────────────────────────────────────────────
+  - interval: 1m
+    input_series:
+      - series: 'trivela_operator_xlm_balance_stroops{job="trivela-backend"}'
+        values: '100000000 100000000 30000000 30000000 30000000 30000000 30000000 30000000'
+
+    alert_rule_test:
+      - eval_time: 8m
+        alertname: OperatorLowBalance
+        exp_alerts:
+          - exp_labels:
+              job: trivela-backend
+              severity: warning
+              team: contracts
+            exp_annotations:
+              summary: 'Operator wallet balance is low'
+              description:
+                'Operator XLM balance is 30M stroops (< 5 XLM). Transaction fees may fail. Top up
+                the operator wallet immediately.'
+              runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#operator-wallet-topup'
+
+  # ── CanaryJourneyFailed ─────────────────────────────────────────────────────
+  - interval: 1m
+    input_series:
+      - series: 'trivela_canary_success{job="trivela-canary"}'
+        values: '1 1 0 0 0 0 0 0'
+
+    alert_rule_test:
+      - eval_time: 7m
+        alertname: CanaryJourneyFailed
+        exp_alerts:
+          - exp_labels:
+              job: trivela-canary
+              severity: critical
+              team: backend
+            exp_annotations:
+              summary: 'Synthetic canary journey failed'
+              description:
+                'The register→credit→claim canary on testnet has not succeeded for 5 minutes. Core
+                user journey is broken. Check canary logs and RPC/contract health.'
+              runbook_url: 'https://github.com/FinesseStudioLab/Trivela/blob/main/docs/RUNBOOK.md#canary-failure'
diff --git a/monitoring/alertmanager.yml b/monitoring/alertmanager.yml
new file mode 100644
index 00000000..34fcb8b4
--- /dev/null
+++ b/monitoring/alertmanager.yml
@@ -0,0 +1,73 @@
+global:
+  resolve_timeout: 5m
+  slack_api_url: '${SLACK_WEBHOOK_URL}'
+
+templates:
+  - '/etc/alertmanager/templates/*.tmpl'
+
+route:
+  group_by: ['alertname', 'team']
+  group_wait: 30s
+  group_interval: 5m
+  repeat_interval: 4h
+  receiver: 'slack-default'
+  routes:
+    - match:
+        severity: critical
+      receiver: 'slack-critical'
+      continue: true
+    - match:
+        team: contracts
+      receiver: 'slack-contracts'
+      continue: true
+    - match:
+        alertname: ContractPaused
+      receiver: 'pagerduty-oncall'
+    - match:
+        alertname: CanaryJourneyFailed
+      receiver: 'pagerduty-oncall'
+
+receivers:
+  - name: 'slack-default'
+    slack_configs:
+      - channel: '#trivela-alerts'
+        send_resolved: true
+        title: '[{{ .Status | toUpper }}] {{ .GroupLabels.alertname }}'
+        text: >-
+          {{ range .Alerts }} *Alert:* {{ .Annotations.summary }} *Details:* {{
+          .Annotations.description }} *Runbook:* {{ .Annotations.runbook_url }} {{ end }}
+
+  - name: 'slack-critical'
+    slack_configs:
+      - channel: '#trivela-critical'
+        send_resolved: true
+        title: '🚨 CRITICAL: {{ .GroupLabels.alertname }}'
+        text: >-
+          {{ range .Alerts }} *Summary:* {{ .Annotations.summary }} *Description:* {{
+          .Annotations.description }} *Runbook:* {{ .Annotations.runbook_url }} {{ end }}
+
+  - name: 'slack-contracts'
+    slack_configs:
+      - channel: '#trivela-contracts'
+        send_resolved: true
+        title: '📋 Contract Alert: {{ .GroupLabels.alertname }}'
+        text: >-
+          {{ range .Alerts }} {{ .Annotations.description }} Runbook: {{ .Annotations.runbook_url }}
+          {{ end }}
+
+  - name: 'pagerduty-oncall'
+    pagerduty_configs:
+      - service_key: '${PAGERDUTY_SERVICE_KEY}'
+        description: '{{ .GroupLabels.alertname }}: {{ (index .Alerts 0).Annotations.summary }}'
+
+inhibit_rules:
+  - source_match:
+      alertname: 'BackendDown'
+    target_match_re:
+      alertname: 'HighBackendErrorRate|BackendProcessRestart|HighP95Latency'
+    equal: ['instance']
+  - source_match:
+      alertname: 'AllRpcEndpointsUnhealthy'
+    target_match:
+      alertname: 'RpcPoolSaturated'
+    equal: ['job']
diff --git a/monitoring/dashboards/trivela-api.json b/monitoring/dashboards/trivela-api.json
new file mode 100644
index 00000000..d1ff5021
--- /dev/null
+++ b/monitoring/dashboards/trivela-api.json
@@ -0,0 +1,180 @@
+{
+  "__inputs": [
+    {
+      "name": "DS_PROMETHEUS",
+      "label": "Prometheus",
+      "description": "",
+      "type": "datasource",
+      "pluginId": "prometheus"
+    }
+  ],
+  "__requires": [
+    { "type": "grafana", "id": "grafana", "name": "Grafana", "version": "10.4.0" },
+    { "type": "datasource", "id": "prometheus", "name": "Prometheus", "version": "1.0.0" },
+    { "type": "panel", "id": "timeseries", "name": "Time series", "version": "" },
+    { "type": "panel", "id": "stat", "name": "Stat", "version": "" }
+  ],
+  "annotations": { "list": [] },
+  "description": "Trivela API — request rates, error rates, p95 latency, and auth events.",
+  "editable": true,
+  "fiscalYearStartMonth": 0,
+  "graphTooltip": 1,
+  "id": null,
+  "links": [],
+  "panels": [
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "color": { "mode": "palette-classic" }, "unit": "reqps" },
+        "overrides": []
+      },
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 0 },
+      "id": 1,
+      "options": { "legend": { "calcs": [], "displayMode": "list", "placement": "bottom" }, "tooltip": { "mode": "multi" } },
+      "targets": [
+        {
+          "expr": "sum(rate(trivela_requests_total{job=\"trivela-backend\"}[5m]))",
+          "legendFormat": "req/s",
+          "refId": "A"
+        }
+      ],
+      "title": "Request Rate",
+      "type": "timeseries"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "color": { "mode": "palette-classic" }, "unit": "percentunit", "max": 1, "min": 0 },
+        "overrides": []
+      },
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 0 },
+      "id": 2,
+      "options": { "legend": { "calcs": [], "displayMode": "list", "placement": "bottom" }, "tooltip": { "mode": "multi" } },
+      "targets": [
+        {
+          "expr": "sum(rate(trivela_request_errors_total{job=\"trivela-backend\"}[5m])) / sum(rate(trivela_requests_total{job=\"trivela-backend\"}[5m]))",
+          "legendFormat": "error rate",
+          "refId": "A"
+        }
+      ],
+      "title": "5xx Error Rate",
+      "type": "timeseries"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "color": { "mode": "palette-classic" }, "unit": "ms" },
+        "overrides": []
+      },
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 8 },
+      "id": 3,
+      "options": { "legend": { "calcs": [], "displayMode": "list", "placement": "bottom" }, "tooltip": { "mode": "multi" } },
+      "targets": [
+        {
+          "expr": "histogram_quantile(0.50, sum(rate(trivela_http_request_duration_ms_bucket{job=\"trivela-backend\"}[5m])) by (le))",
+          "legendFormat": "p50",
+          "refId": "A"
+        },
+        {
+          "expr": "histogram_quantile(0.95, sum(rate(trivela_http_request_duration_ms_bucket{job=\"trivela-backend\"}[5m])) by (le))",
+          "legendFormat": "p95",
+          "refId": "B"
+        },
+        {
+          "expr": "histogram_quantile(0.99, sum(rate(trivela_http_request_duration_ms_bucket{job=\"trivela-backend\"}[5m])) by (le))",
+          "legendFormat": "p99",
+          "refId": "C"
+        }
+      ],
+      "title": "Request Latency (p50 / p95 / p99)",
+      "type": "timeseries"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "color": { "mode": "palette-classic" }, "unit": "reqps" },
+        "overrides": []
+      },
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 8 },
+      "id": 4,
+      "options": { "legend": { "calcs": [], "displayMode": "list", "placement": "bottom" }, "tooltip": { "mode": "multi" } },
+      "targets": [
+        {
+          "expr": "sum(rate(trivela_auth_failures_total{job=\"trivela-backend\"}[5m]))",
+          "legendFormat": "auth failures/s",
+          "refId": "A"
+        },
+        {
+          "expr": "sum(rate(trivela_auth_lockouts_total{job=\"trivela-backend\"}[5m]))",
+          "legendFormat": "lockouts/s",
+          "refId": "B"
+        }
+      ],
+      "title": "Auth Failures & Lockouts",
+      "type": "timeseries"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "color": { "fixedColor": "green", "mode": "fixed" }, "unit": "short" },
+        "overrides": []
+      },
+      "gridPos": { "h": 4, "w": 6, "x": 0, "y": 16 },
+      "id": 5,
+      "options": { "colorMode": "background", "graphMode": "none", "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "targets": [
+        {
+          "expr": "up{job=\"trivela-backend\"}",
+          "legendFormat": "up",
+          "refId": "A"
+        }
+      ],
+      "title": "Backend Up",
+      "type": "stat"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "unit": "s" },
+        "overrides": []
+      },
+      "gridPos": { "h": 4, "w": 6, "x": 6, "y": 16 },
+      "id": 6,
+      "options": { "colorMode": "value", "graphMode": "none", "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "targets": [
+        {
+          "expr": "trivela_process_uptime_seconds{job=\"trivela-backend\"}",
+          "legendFormat": "uptime",
+          "refId": "A"
+        }
+      ],
+      "title": "Process Uptime",
+      "type": "stat"
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "tags": ["trivela", "api"],
+  "templating": {
+    "list": [
+      {
+        "current": {},
+        "hide": 0,
+        "includeAll": false,
+        "label": "Datasource",
+        "multi": false,
+        "name": "DS_PROMETHEUS",
+        "options": [],
+        "query": "prometheus",
+        "refresh": 1,
+        "type": "datasource"
+      }
+    ]
+  },
+  "time": { "from": "now-1h", "to": "now" },
+  "timepicker": {},
+  "timezone": "browser",
+  "title": "Trivela API",
+  "uid": "trivela-api",
+  "version": 1
+}
diff --git a/monitoring/dashboards/trivela-rpc-pools.json b/monitoring/dashboards/trivela-rpc-pools.json
new file mode 100644
index 00000000..80a07363
--- /dev/null
+++ b/monitoring/dashboards/trivela-rpc-pools.json
@@ -0,0 +1,157 @@
+{
+  "__inputs": [
+    {
+      "name": "DS_PROMETHEUS",
+      "label": "Prometheus",
+      "type": "datasource",
+      "pluginId": "prometheus"
+    }
+  ],
+  "__requires": [
+    { "type": "grafana", "id": "grafana", "name": "Grafana", "version": "10.4.0" },
+    { "type": "datasource", "id": "prometheus", "name": "Prometheus", "version": "1.0.0" },
+    { "type": "panel", "id": "timeseries", "name": "Time series", "version": "" },
+    { "type": "panel", "id": "stat", "name": "Stat", "version": "" },
+    { "type": "panel", "id": "gauge", "name": "Gauge", "version": "" }
+  ],
+  "annotations": { "list": [] },
+  "description": "Trivela RPC pool health, saturation, and circuit-breaker state.",
+  "editable": true,
+  "graphTooltip": 1,
+  "id": null,
+  "links": [],
+  "panels": [
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "unit": "short", "min": 0, "thresholds": { "mode": "absolute", "steps": [{ "color": "green", "value": null }, { "color": "yellow", "value": 7 }, { "color": "red", "value": 9 }] } },
+        "overrides": []
+      },
+      "gridPos": { "h": 6, "w": 6, "x": 0, "y": 0 },
+      "id": 1,
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] }, "orientation": "auto", "showThresholdLabels": false, "showThresholdMarkers": true },
+      "targets": [
+        {
+          "expr": "trivela_rpc_pool_in_use{job=\"trivela-backend\"}",
+          "legendFormat": "in use",
+          "refId": "A"
+        }
+      ],
+      "title": "RPC Pool — In Use",
+      "type": "gauge"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "unit": "short", "min": 0, "thresholds": { "mode": "absolute", "steps": [{ "color": "green", "value": null }, { "color": "yellow", "value": 1 }, { "color": "red", "value": 5 }] } },
+        "overrides": []
+      },
+      "gridPos": { "h": 6, "w": 6, "x": 6, "y": 0 },
+      "id": 2,
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] }, "orientation": "auto", "showThresholdLabels": false, "showThresholdMarkers": true },
+      "targets": [
+        {
+          "expr": "trivela_rpc_pool_waiting{job=\"trivela-backend\"}",
+          "legendFormat": "waiting",
+          "refId": "A"
+        }
+      ],
+      "title": "RPC Pool — Callers Waiting",
+      "type": "gauge"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "color": { "fixedColor": "green", "mode": "fixed" }, "unit": "short" },
+        "overrides": []
+      },
+      "gridPos": { "h": 6, "w": 6, "x": 12, "y": 0 },
+      "id": 3,
+      "options": { "colorMode": "background", "graphMode": "none", "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "targets": [
+        {
+          "expr": "trivela_rpc_pool_healthy{job=\"trivela-backend\"}",
+          "legendFormat": "healthy endpoints",
+          "refId": "A"
+        }
+      ],
+      "title": "Healthy RPC Endpoints",
+      "type": "stat"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "color": { "mode": "palette-classic" }, "unit": "short" },
+        "overrides": []
+      },
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 6 },
+      "id": 4,
+      "options": { "legend": { "calcs": [], "displayMode": "list", "placement": "bottom" }, "tooltip": { "mode": "multi" } },
+      "targets": [
+        {
+          "expr": "trivela_rpc_pool_in_use{job=\"trivela-backend\"}",
+          "legendFormat": "in use",
+          "refId": "A"
+        },
+        {
+          "expr": "trivela_rpc_pool_idle{job=\"trivela-backend\"}",
+          "legendFormat": "idle",
+          "refId": "B"
+        },
+        {
+          "expr": "trivela_rpc_pool_waiting{job=\"trivela-backend\"}",
+          "legendFormat": "waiting",
+          "refId": "C"
+        }
+      ],
+      "title": "RPC Pool Saturation Over Time",
+      "type": "timeseries"
+    },
+    {
+      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
+      "fieldConfig": {
+        "defaults": { "color": { "mode": "palette-classic" }, "unit": "short" },
+        "overrides": []
+      },
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 6 },
+      "id": 5,
+      "options": { "legend": { "calcs": [], "displayMode": "list", "placement": "bottom" }, "tooltip": { "mode": "multi" } },
+      "targets": [
+        {
+          "expr": "trivela_rpc_pool_healthy{job=\"trivela-backend\"}",
+          "legendFormat": "healthy",
+          "refId": "A"
+        },
+        {
+          "expr": "trivela_rpc_pool_unhealthy{job=\"trivela-backend\"}",
+          "legendFormat": "unhealthy",
+          "refId": "B"
+        }
+      ],
+      "title": "RPC Endpoint Health",
+      "type": "timeseries"
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "tags": ["trivela", "rpc", "pools"],
+  "templating": {
+    "list": [
+      {
+        "current": {},
+        "hide": 0,
+        "label": "Datasource",
+        "name": "DS_PROMETHEUS",
+        "query": "prometheus",
+        "refresh": 1,
+        "type": "datasource"
+      }
+    ]
+  },
+  "time": { "from": "now-1h", "to": "now" },
+  "timepicker": {},
+  "timezone": "browser",
+  "title": "Trivela RPC & Pools",
+  "uid": "trivela-rpc-pools",
+  "version": 1
+}
diff --git a/monitoring/grafana/provisioning/dashboards/trivela.yml b/monitoring/grafana/provisioning/dashboards/trivela.yml
new file mode 100644
index 00000000..cc04af40
--- /dev/null
+++ b/monitoring/grafana/provisioning/dashboards/trivela.yml
@@ -0,0 +1,13 @@
+apiVersion: 1
+
+providers:
+  - name: 'Trivela'
+    orgId: 1
+    folder: 'Trivela'
+    type: file
+    disableDeletion: false
+    updateIntervalSeconds: 30
+    allowUiUpdates: true
+    options:
+      path: /var/lib/grafana/dashboards
+      foldersFromFilesStructure: false
diff --git a/monitoring/grafana/provisioning/datasources/prometheus.yml b/monitoring/grafana/provisioning/datasources/prometheus.yml
new file mode 100644
index 00000000..7a6dc70b
--- /dev/null
+++ b/monitoring/grafana/provisioning/datasources/prometheus.yml
@@ -0,0 +1,11 @@
+apiVersion: 1
+
+datasources:
+  - name: Prometheus
+    type: prometheus
+    access: proxy
+    url: http://prometheus:9090
+    isDefault: true
+    jsonData:
+      timeInterval: '15s'
+      httpMethod: POST
diff --git a/monitoring/prometheus.yml b/monitoring/prometheus.yml
new file mode 100644
index 00000000..0b08ba5f
--- /dev/null
+++ b/monitoring/prometheus.yml
@@ -0,0 +1,36 @@
+global:
+  scrape_interval: 15s
+  evaluation_interval: 15s
+  external_labels:
+    env: '${TRIVELA_ENV:-dev}'
+
+rule_files:
+  - 'alerting/alerting_rules.yml'
+
+alerting:
+  alertmanagers:
+    - static_configs:
+        - targets:
+            - 'alertmanager:9093'
+
+scrape_configs:
+  - job_name: 'trivela-backend'
+    scrape_interval: 15s
+    static_configs:
+      - targets: ['backend:3001']
+    metrics_path: /metrics
+
+  - job_name: 'trivela-rpc-health'
+    scrape_interval: 30s
+    metrics_path: /health/rpc
+    static_configs:
+      - targets: ['backend:3001']
+
+  - job_name: 'node-exporter'
+    scrape_interval: 30s
+    static_configs:
+      - targets: ['node-exporter:9100']
+
+  - job_name: 'prometheus'
+    static_configs:
+      - targets: ['localhost:9090']
diff --git a/scripts/canary.mjs b/scripts/canary.mjs
new file mode 100755
index 00000000..922d1c01
--- /dev/null
+++ b/scripts/canary.mjs
@@ -0,0 +1,152 @@
+#!/usr/bin/env node
+/**
+ * Trivela Synthetic Canary (issue #650 — synthetic canary).
+ *
+ * Exercises the core register→credit→claim journey against the testnet contract
+ * and emits Prometheus metrics so the CanaryJourneyFailed / CanarySlowJourney
+ * alerts can fire within minutes of a real failure.
+ *
+ * Metrics emitted (plain Prometheus text on stdout):
+ *   trivela_canary_success{job="trivela-canary"}   1 on success, 0 on failure
+ *   trivela_canary_duration_seconds{...}            wall time for the full journey
+ *   trivela_canary_last_run_timestamp{...}          unix epoch of last execution
+ *
+ * Usage:
+ *   node scripts/canary.mjs
+ *   # or add to crontab (every 5 min — cron pattern "asterisk /5 * * * *"):
+ *   # node scripts/canary.mjs >> /var/log/trivela-canary.log 2>&1
+ *
+ * Environment variables (inherit from .env or CI secrets):
+ *   CANARY_API_URL         Base URL of the Trivela backend (default: http://localhost:3001)
+ *   CANARY_API_KEY         API key with campaign write access
+ *   CANARY_WALLET          Stellar G-address to simulate as the claimant
+ *   CANARY_CONTRACT_ID     Testnet campaign contract ID (C…56 chars)
+ *   CANARY_TIMEOUT_MS      Per-step timeout in ms (default: 15000)
+ *   CANARY_METRICS_FILE    If set, write metrics to this file instead of stdout
+ *   STELLAR_NETWORK        testnet | mainnet (default: testnet)
+ */
+
+import { writeFileSync } from 'node:fs';
+
+const API_URL = (process.env.CANARY_API_URL ?? 'http://localhost:3001').replace(/\/$/, '');
+const API_KEY = process.env.CANARY_API_KEY ?? '';
+const WALLET = process.env.CANARY_WALLET ?? 'GDUMMY_CANARY_WALLET_ADDRESS_NOT_SET';
+const CONTRACT_ID = process.env.CANARY_CONTRACT_ID ?? '';
+const TIMEOUT_MS = Number(process.env.CANARY_TIMEOUT_MS ?? 15_000);
+const METRICS_FILE = process.env.CANARY_METRICS_FILE ?? '';
+const JOB_LABEL = 'trivela-canary';
+
+/** @param {string} url @param {RequestInit} opts */
+async function apiFetch(url, opts = {}) {
+  const ac = new AbortController();
+  const t = setTimeout(() => ac.abort(), TIMEOUT_MS);
+  try {
+    const res = await fetch(url, {
+      ...opts,
+      signal: ac.signal,
+      headers: {
+        'Content-Type': 'application/json',
+        ...(API_KEY ? { 'x-api-key': API_KEY } : {}),
+        ...opts.headers,
+      },
+    });
+    if (!res.ok) {
+      const body = await res.text().catch(() => '');
+      throw new Error(`HTTP ${res.status} ${res.statusText} — ${body.slice(0, 200)}`);
+    }
+    return res.json();
+  } finally {
+    clearTimeout(t);
+  }
+}
+
+function emitMetrics({ success, durationSeconds, timestamp }) {
+  const lines = [
+    `# HELP trivela_canary_success 1 if the last canary journey succeeded, 0 otherwise.`,
+    `# TYPE trivela_canary_success gauge`,
+    `trivela_canary_success{job="${JOB_LABEL}"} ${success ? 1 : 0}`,
+    `# HELP trivela_canary_duration_seconds Wall time for the full canary journey.`,
+    `# TYPE trivela_canary_duration_seconds gauge`,
+    `trivela_canary_duration_seconds{job="${JOB_LABEL}"} ${durationSeconds.toFixed(3)}`,
+    `# HELP trivela_canary_last_run_timestamp Unix epoch of the most recent canary run.`,
+    `# TYPE trivela_canary_last_run_timestamp gauge`,
+    `trivela_canary_last_run_timestamp{job="${JOB_LABEL}"} ${timestamp}`,
+    '',
+  ].join('\n');
+
+  if (METRICS_FILE) {
+    writeFileSync(METRICS_FILE, lines, 'utf8');
+  } else {
+    process.stdout.write(lines);
+  }
+}
+
+async function runCanary() {
+  const start = Date.now();
+  const timestamp = Math.floor(start / 1000);
+  let campaignId = null;
+
+  try {
+    // ── Step 1: health check ────────────────────────────────────────────────
+    const health = await apiFetch(`${API_URL}/health`);
+    if (health.status !== 'ok' && health.status !== 'degraded') {
+      throw new Error(`Health check returned unexpected status: ${health.status}`);
+    }
+
+    // ── Step 2: create a synthetic canary campaign (register) ───────────────
+    // This simulates the "register" step of the campaign creation journey.
+    const campaign = await apiFetch(`${API_URL}/api/v1/campaigns`, {
+      method: 'POST',
+      body: JSON.stringify({
+        name: `__canary_${Date.now()}`,
+        description: 'Synthetic canary campaign — safe to delete',
+        rewardPerAction: 1,
+        active: true,
+        ...(CONTRACT_ID ? { contractId: CONTRACT_ID } : {}),
+        tags: ['canary'],
+      }),
+    });
+    campaignId = campaign.id ?? campaign.data?.id;
+    if (!campaignId) throw new Error('Campaign creation did not return an id');
+
+    // ── Step 3: credit a claimant (credit step) ─────────────────────────────
+    // Verifies the credit path is reachable.
+    await apiFetch(`${API_URL}/api/v1/campaigns/${campaignId}/credits`, {
+      method: 'POST',
+      body: JSON.stringify({ walletAddress: WALLET, amount: 1 }),
+    }).catch((err) => {
+      // Credits may require a specific env; treat 404/405 as skip, not failure.
+      if (err.message.startsWith('HTTP 404') || err.message.startsWith('HTTP 405')) return null;
+      throw err;
+    });
+
+    // ── Step 4: fetch campaign stats (claim readiness check) ─────────────────
+    const stats = await apiFetch(`${API_URL}/api/v1/campaigns/${campaignId}`);
+    if (!stats || (!stats.id && !stats.data?.id)) {
+      throw new Error('Campaign stats fetch returned empty response');
+    }
+
+    // ── Step 5: delete the canary campaign (cleanup) ─────────────────────────
+    await apiFetch(`${API_URL}/api/v1/campaigns/${campaignId}`, { method: 'DELETE' }).catch(
+      () => {},
+    );
+
+    const durationSeconds = (Date.now() - start) / 1000;
+    emitMetrics({ success: true, durationSeconds, timestamp });
+    process.stderr.write(
+      `[canary] OK — journey completed in ${durationSeconds.toFixed(2)}s\n`,
+    );
+    process.exit(0);
+  } catch (err) {
+    // Best-effort cleanup.
+    if (campaignId) {
+      apiFetch(`${API_URL}/api/v1/campaigns/${campaignId}`, { method: 'DELETE' }).catch(() => {});
+    }
+    const durationSeconds = (Date.now() - start) / 1000;
+    emitMetrics({ success: false, durationSeconds, timestamp });
+    process.stderr.write(`[canary] FAIL — ${err.message}\n`);
+    process.exit(1);
+  }
+}
+
+runCanary();