diff --git a/notebooks/service_health.ipynb b/notebooks/service_health.ipynb
index f4e8976..f382847 100644
--- a/notebooks/service_health.ipynb
+++ b/notebooks/service_health.ipynb
@@ -47,7 +47,9 @@
    },
    "source": [
     "# Service Health\n",
-    "Fetch the typed `/healthz` payload from the configured JointFM deployment and display every field. Use this notebook to diagnose health-endpoint issues: version pins, advertised modes, `decoding_strategy` (parallel vs autoregressive horizon decoding), sample-count budgets, and the optional `data_generation` capability block."
+    "Fetch the typed `/healthz` payload from the configured JointFM deployment and display every field. Use this notebook to diagnose health-endpoint issues: version pins, advertised modes, `decoding_strategy` (parallel vs autoregressive horizon decoding), sample-count budgets, and the optional `data_generation` capability block.\n",
+    "\n",
+    "`JOINTFM_DEPLOYMENT_IDS` configures a round-robin pool of hosted deployments. Each endpoint's health payload describes only that endpoint, so the client probes every configured peer and aggregates locally: `health()` returns consensus metadata whose `max_sample_count` is the **minimum** reachable cap, which is the sample-batch size used to split oversized sample requests. `cache=True` stores this probe so the topology section below reuses it instead of issuing a second round of requests."
    ]
   },
   {
@@ -66,17 +68,67 @@
     "from jointfm_client import JointFMClient\n",
     "\n",
     "client = JointFMClient.from_env()\n",
-    "health = client.health(refresh=True)\n",
-    "pprint(asdict(health), sort_dicts=False, width=100)\n",
-    "health"
+    "health = client.health(cache=True, refresh=True)\n",
+    "pprint(asdict(health), sort_dicts=False, width=100)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "4",
+   "metadata": {
+    "id": "service-health-topology-description",
+    "language": "markdown"
+   },
+   "source": [
+    "## Deployment Topology\n",
+    "`health_instances()` returns the per-endpoint results behind that consensus: one entry per configured deployment ID, including peers skipped as unreachable or contract-incompatible. Its `max_sample_count` is the **sum** of reachable caps, the overall parallel capacity of the pool, and `topology_label` groups those caps as `<count>x<cap>`. Reachable peers must agree on `model_version` and `checkpoint_version`; a mismatch fails the probe rather than silently mixing models across requests."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "5",
+   "metadata": {
+    "id": "service-health-topology",
+    "language": "python"
+   },
+   "outputs": [],
+   "source": [
+    "settings = client.settings\n",
+    "instances = client.health_instances(cache=True)\n",
+    "# A pool repeats its primary endpoint in `instances`; the seed covers single-endpoint configs.\n",
+    "predict_url_by_id: dict[str | None, str] = {\n",
+    "    settings.deployment_id: settings.predict_url\n",
+    "}\n",
+    "for instance in settings.instances:\n",
+    "    predict_url_by_id[instance.deployment_id] = instance.predict_url\n",
+    "\n",
+    "print(f\"selector:             {settings.deployment_selector}\")\n",
+    "print(f\"configured endpoints: {len(instances.instances)}\")\n",
+    "print(f\"topology:             {instances.topology_label}\")\n",
+    "print(f\"parallel capacity:    {instances.max_sample_count} (sum of reachable caps)\")\n",
+    "print(f\"sample-batch cap:     {health.max_sample_count} (minimum reachable cap)\")\n",
+    "\n",
+    "for instance in instances.instances:\n",
+    "    print()\n",
+    "    if instance.metadata is None:\n",
+    "        print(f\"{instance.deployment_id}: unavailable\")\n",
+    "        print(f\"  error:      {instance.error}\")\n",
+    "        continue\n",
+    "    print(f\"{instance.deployment_id}: available\")\n",
+    "    print(f\"  url:        {predict_url_by_id[instance.deployment_id]}\")\n",
+    "    print(f\"  device:     {instance.metadata.device}\")\n",
+    "    print(f\"  image:      {instance.metadata.image_version}\")\n",
+    "    print(f\"  checkpoint: {instance.metadata.checkpoint_version}\")\n",
+    "    print(f\"  samples:    {instance.metadata.max_sample_count}\")"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "joint-client-python (3.13.3)",
+   "display_name": "Python (joint-client-python)",
    "language": "python",
-   "name": "python3"
+   "name": "joint-client-python"
   },
   "language_info": {
    "codemirror_mode": {