diff --git a/.github/workflows/imp-dashboard-metrics-demo.yml b/.github/workflows/imp-dashboard-metrics-demo.yml
new file mode 100644
index 0000000..aed9859
--- /dev/null
+++ b/.github/workflows/imp-dashboard-metrics-demo.yml
@@ -0,0 +1,35 @@
+name: Imp dashboard metrics demo
+
+on:
+  workflow_dispatch:
+    inputs:
+      hold_seconds:
+        description: Seconds to keep the runner handoff active for dashboard inspection
+        required: true
+        default: "120"
+        type: choice
+        options: ["60", "120", "300"]
+
+permissions:
+  contents: read
+
+jobs:
+  dashboard-demo:
+    name: Create observable runner activity
+    runs-on:
+      group: omni-runner
+      labels: [self-hosted, imp]
+    timeout-minutes: 10
+    steps:
+      - name: Report the dashboard correlation fields
+        run: |
+          printf '%s\n' \
+            "github_repository=$GITHUB_REPOSITORY" \
+            "github_run_id=$GITHUB_RUN_ID" \
+            "github_run_attempt=$GITHUB_RUN_ATTEMPT" \
+            "github_job=$GITHUB_JOB" \
+            "runner_name=$RUNNER_NAME"
+      - name: Hold the runner handoff open
+        run: |
+          echo "holding runner activity for ${{ inputs.hold_seconds }} seconds"
+          sleep "${{ inputs.hold_seconds }}"
diff --git a/charts/imp/dashboards/imp-admin.json b/charts/imp/dashboards/imp-admin.json
new file mode 100644
index 0000000..7402484
--- /dev/null
+++ b/charts/imp/dashboards/imp-admin.json
@@ -0,0 +1,351 @@
+{
+  "annotations": {
+    "list": []
+  },
+  "description": "Technical health. Missing data means unknown or missing telemetry, not healthy.",
+  "editable": true,
+  "refresh": "30s",
+  "schemaVersion": 38,
+  "tags": [
+    "imp",
+    "administrator",
+    "microvm"
+  ],
+  "time": {
+    "from": "now-1h",
+    "to": "now"
+  },
+  "timezone": "browser",
+  "title": "Imp administrator",
+  "uid": "imp-admin-v1",
+  "version": 1,
+  "templating": {
+    "list": [
+      {
+        "name": "datasource",
+        "label": "Data source",
+        "type": "datasource",
+        "query": "prometheus",
+        "refresh": 1,
+        "current": {}
+      },
+      {
+        "name": "namespace",
+        "label": "Namespace",
+        "type": "query",
+        "datasource": {
+          "type": "prometheus",
+          "uid": "${datasource}"
+        },
+        "definition": "label_values(imp_vm_state, namespace)",
+        "query": {
+          "query": "label_values(imp_vm_state, namespace)",
+          "refId": "NamespaceVariableQuery"
+        },
+        "includeAll": true,
+        "allValue": ".*",
+        "multi": true,
+        "refresh": 2,
+        "current": {}
+      },
+      {
+        "name": "vm",
+        "label": "VM",
+        "type": "query",
+        "datasource": {
+          "type": "prometheus",
+          "uid": "${datasource}"
+        },
+        "definition": "label_values(imp_vm_state{namespace=~\"${namespace:regex}\"}, impvm)",
+        "query": {
+          "query": "label_values(imp_vm_state{namespace=~\"${namespace:regex}\"}, impvm)",
+          "refId": "VMVariableQuery"
+        },
+        "includeAll": true,
+        "allValue": ".*",
+        "multi": true,
+        "refresh": 2,
+        "current": {}
+      }
+    ]
+  },
+  "panels": [
+    {
+      "id": 1,
+      "type": "stat",
+      "title": "Runtime ready",
+      "description": "Missing series means unknown or missing telemetry.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 6,
+        "w": 8,
+        "x": 0,
+        "y": 0
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "imp_runtime_ready"
+        }
+      ],
+      "options": {
+        "reduceOptions": {
+          "calcs": [
+            "lastNotNull"
+          ]
+        },
+        "orientation": "auto",
+        "textMode": "auto",
+        "colorMode": "value",
+        "graphMode": "none"
+      },
+      "fieldConfig": {
+        "defaults": {},
+        "overrides": []
+      }
+    },
+    {
+      "id": 2,
+      "type": "timeseries",
+      "title": "Observed VMs by state",
+      "description": "Missing series means no observed VMs or missing telemetry.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 8,
+        "x": 8,
+        "y": 0
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum by (state) (imp_vm_state{namespace=~\"${namespace:regex}\",impvm=~\"${vm:regex}\"} == 1)",
+          "legendFormat": "{{state}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          }
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    },
+    {
+      "id": 3,
+      "type": "timeseries",
+      "title": "Observed VMs by node",
+      "description": "Missing series means no observed VMs or missing telemetry.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 8,
+        "x": 16,
+        "y": 0
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum by (node) (imp_vm_state{namespace=~\"${namespace:regex}\",impvm=~\"${vm:regex}\"} == 1)",
+          "legendFormat": "{{node}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          }
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    },
+    {
+      "id": 4,
+      "type": "timeseries",
+      "title": "Provisioning latency p95",
+      "description": "Cluster-wide; namespace and VM filters do not apply.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 12,
+        "x": 0,
+        "y": 8
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "histogram_quantile(0.95, sum by (le) (rate(imp_vm_scheduling_latency_seconds_bucket[$__rate_interval])))",
+          "legendFormat": "scheduling p95"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          },
+          "unit": "s"
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    },
+    {
+      "id": 5,
+      "type": "timeseries",
+      "title": "Boot latency p95",
+      "description": "Cluster-wide; namespace and VM filters do not apply.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 12,
+        "x": 12,
+        "y": 8
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "histogram_quantile(0.95, sum by (le) (rate(imp_vm_boot_latency_seconds_bucket[$__rate_interval])))",
+          "legendFormat": "boot p95"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          },
+          "unit": "s"
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    },
+    {
+      "id": 6,
+      "type": "stat",
+      "title": "Recorded scheduling transitions",
+      "description": "Observed histogram events; not unique VM counts.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 6,
+        "w": 12,
+        "x": 0,
+        "y": 16
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum(increase(imp_vm_scheduling_latency_seconds_count[$__range]))"
+        }
+      ],
+      "options": {
+        "reduceOptions": {
+          "calcs": [
+            "lastNotNull"
+          ]
+        },
+        "orientation": "auto",
+        "textMode": "auto",
+        "colorMode": "value",
+        "graphMode": "none"
+      },
+      "fieldConfig": {
+        "defaults": {},
+        "overrides": []
+      }
+    },
+    {
+      "id": 7,
+      "type": "stat",
+      "title": "Recorded boot transitions",
+      "description": "Observed histogram events; not unique VM counts.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 6,
+        "w": 12,
+        "x": 12,
+        "y": 16
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum(increase(imp_vm_boot_latency_seconds_count[$__range]))"
+        }
+      ],
+      "options": {
+        "reduceOptions": {
+          "calcs": [
+            "lastNotNull"
+          ]
+        },
+        "orientation": "auto",
+        "textMode": "auto",
+        "colorMode": "value",
+        "graphMode": "none"
+      },
+      "fieldConfig": {
+        "defaults": {},
+        "overrides": []
+      }
+    }
+  ]
+}
diff --git a/charts/imp/dashboards/imp-github-runners.json b/charts/imp/dashboards/imp-github-runners.json
new file mode 100644
index 0000000..f7f2848
--- /dev/null
+++ b/charts/imp/dashboards/imp-github-runners.json
@@ -0,0 +1,232 @@
+{
+  "annotations": {
+    "list": []
+  },
+  "description": "Runner VM infrastructure and guest runner handoffs. GitHub registration, job assignment, and cleanup need separate GitHub telemetry.",
+  "editable": true,
+  "refresh": "30s",
+  "schemaVersion": 38,
+  "tags": [
+    "imp",
+    "github",
+    "runner",
+    "microvm"
+  ],
+  "time": {
+    "from": "now-1h",
+    "to": "now"
+  },
+  "timezone": "browser",
+  "title": "Imp GitHub runners",
+  "uid": "imp-github-runners-v1",
+  "version": 2,
+  "templating": {
+    "list": [
+      {
+        "name": "datasource",
+        "label": "Data source",
+        "type": "datasource",
+        "query": "prometheus",
+        "refresh": 1,
+        "current": {}
+      },
+      {
+        "name": "runner_vm",
+        "label": "Runner VM",
+        "type": "query",
+        "datasource": {
+          "type": "prometheus",
+          "uid": "${datasource}"
+        },
+        "definition": "label_values(imp_vm_state, impvm)",
+        "query": {
+          "query": "label_values(imp_vm_state, impvm)",
+          "refId": "RunnerVMVariableQuery"
+        },
+        "includeAll": false,
+        "multi": true,
+        "refresh": 2,
+        "current": {}
+      }
+    ]
+  },
+  "panels": [
+    {
+      "id": 1,
+      "type": "text",
+      "title": "Scope",
+      "gridPos": {
+        "h": 5,
+        "w": 24,
+        "x": 0,
+        "y": 0
+      },
+      "options": {
+        "mode": "markdown",
+        "content": "**Runner VM infrastructure and handoff activity.** The handoff panels show Imp delivering a one-time runner configuration to a guest. They do not prove GitHub registration, job assignment, or cleanup."
+      }
+    },
+    {
+      "id": 2,
+      "type": "timeseries",
+      "title": "Selected runner VM observations",
+      "description": "No selected runner VM telemetry.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 0,
+        "y": 5
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "imp_vm_state{impvm=~\"${runner_vm:regex}\"} == 1",
+          "legendFormat": "{{impvm}} \u00b7 {{namespace}} \u00b7 {{node}} \u00b7 {{state}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          }
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    },
+    {
+      "id": 3,
+      "type": "timeseries",
+      "title": "Selected runner VMs by state",
+      "description": "No selected runner VM telemetry.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 9,
+        "w": 12,
+        "x": 12,
+        "y": 5
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum by (state) (imp_vm_state{impvm=~\"${runner_vm:regex}\"} == 1)",
+          "legendFormat": "{{state}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          }
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    },
+    {
+      "id": 4,
+      "type": "stat",
+      "title": "Active runner handoffs",
+      "description": "A nonzero value means Imp is waiting for a guest runner process to exit.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 12,
+        "x": 0,
+        "y": 14
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum(imp_runner_handoffs_active{impvm=~\"${runner_vm:regex}\"})"
+        }
+      ],
+      "options": {
+        "reduceOptions": {
+          "calcs": [
+            "lastNotNull"
+          ],
+          "fields": "",
+          "values": false
+        },
+        "orientation": "auto",
+        "textMode": "auto"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "decimals": 0
+        },
+        "overrides": []
+      }
+    },
+    {
+      "id": 5,
+      "type": "timeseries",
+      "title": "Completed runner handoffs",
+      "description": "Successful and failed handoffs observed by this agent.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 12,
+        "x": 12,
+        "y": 14
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum by (outcome) (increase(imp_runner_handoffs_total{impvm=~\"${runner_vm:regex}\"}[$__rate_interval]))",
+          "legendFormat": "{{outcome}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "always"
+          }
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    }
+  ]
+}
diff --git a/charts/imp/dashboards/imp-workloads.json b/charts/imp/dashboards/imp-workloads.json
new file mode 100644
index 0000000..2a0a4d6
--- /dev/null
+++ b/charts/imp/dashboards/imp-workloads.json
@@ -0,0 +1,197 @@
+{
+  "annotations": {
+    "list": []
+  },
+  "description": "Lifecycle observations for workload owners. Running does not prove application readiness.",
+  "editable": true,
+  "refresh": "30s",
+  "schemaVersion": 38,
+  "tags": [
+    "imp",
+    "workload",
+    "microvm"
+  ],
+  "time": {
+    "from": "now-1h",
+    "to": "now"
+  },
+  "timezone": "browser",
+  "title": "Imp workloads",
+  "uid": "imp-workloads-v1",
+  "version": 1,
+  "templating": {
+    "list": [
+      {
+        "name": "datasource",
+        "label": "Data source",
+        "type": "datasource",
+        "query": "prometheus",
+        "refresh": 1,
+        "current": {}
+      },
+      {
+        "name": "namespace",
+        "label": "Namespace",
+        "type": "query",
+        "datasource": {
+          "type": "prometheus",
+          "uid": "${datasource}"
+        },
+        "definition": "label_values(imp_vm_state, namespace)",
+        "query": {
+          "query": "label_values(imp_vm_state, namespace)",
+          "refId": "NamespaceVariableQuery"
+        },
+        "includeAll": true,
+        "allValue": ".*",
+        "multi": true,
+        "refresh": 2,
+        "current": {}
+      },
+      {
+        "name": "vm",
+        "label": "VM",
+        "type": "query",
+        "datasource": {
+          "type": "prometheus",
+          "uid": "${datasource}"
+        },
+        "definition": "label_values(imp_vm_state{namespace=~\"${namespace:regex}\"}, impvm)",
+        "query": {
+          "query": "label_values(imp_vm_state{namespace=~\"${namespace:regex}\"}, impvm)",
+          "refId": "VMVariableQuery"
+        },
+        "includeAll": true,
+        "allValue": ".*",
+        "multi": true,
+        "refresh": 2,
+        "current": {}
+      }
+    ]
+  },
+  "panels": [
+    {
+      "id": 1,
+      "type": "timeseries",
+      "title": "Current observed VM inventory",
+      "description": "No matching VM observations can mean filters are wrong or telemetry is missing.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 24,
+        "x": 0,
+        "y": 0
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "imp_vm_state{namespace=~\"${namespace:regex}\",impvm=~\"${vm:regex}\"} == 1",
+          "legendFormat": "{{namespace}} \u00b7 {{impvm}} \u00b7 {{node}} \u00b7 {{state}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          }
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    },
+    {
+      "id": 2,
+      "type": "timeseries",
+      "title": "Selected VMs by state",
+      "description": "No matching VM observations can mean filters are wrong or telemetry is missing.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 12,
+        "x": 0,
+        "y": 8
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum by (state) (imp_vm_state{namespace=~\"${namespace:regex}\",impvm=~\"${vm:regex}\"} == 1)",
+          "legendFormat": "{{state}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          }
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    },
+    {
+      "id": 3,
+      "type": "timeseries",
+      "title": "Observed lifecycle history",
+      "description": "Gaps are missing samples. The chart does not infer transitions between scrapes.",
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${datasource}"
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 12,
+        "x": 12,
+        "y": 8
+      },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "imp_vm_state{namespace=~\"${namespace:regex}\",impvm=~\"${vm:regex}\"} == 1",
+          "legendFormat": "{{impvm}} \u00b7 {{state}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": {
+          "custom": {
+            "lineWidth": 1,
+            "showPoints": "never"
+          }
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": {
+          "displayMode": "table",
+          "placement": "bottom"
+        },
+        "tooltip": {
+          "mode": "multi"
+        }
+      }
+    }
+  ]
+}
diff --git a/charts/imp/templates/grafana/dashboard.yaml b/charts/imp/templates/grafana/dashboard.yaml
index 32781d1..eab1696 100644
--- a/charts/imp/templates/grafana/dashboard.yaml
+++ b/charts/imp/templates/grafana/dashboard.yaml
@@ -10,6 +10,12 @@ metadata:
 data:
   imp-overview.json: |-
 {{ .Files.Get "dashboards/imp-overview.json" | indent 4 }}
+  imp-admin.json: |-
+{{ .Files.Get "dashboards/imp-admin.json" | indent 4 }}
+  imp-workloads.json: |-
+{{ .Files.Get "dashboards/imp-workloads.json" | indent 4 }}
+  imp-github-runners.json: |-
+{{ .Files.Get "dashboards/imp-github-runners.json" | indent 4 }}
   imp-traces.json: |-
 {{ .Files.Get "dashboards/imp-traces.json" | indent 4 }}
 {{- end }}
diff --git a/charts/imp/tests/grafana_test.yaml b/charts/imp/tests/grafana_test.yaml
index 4525765..588e0ea 100644
--- a/charts/imp/tests/grafana_test.yaml
+++ b/charts/imp/tests/grafana_test.yaml
@@ -14,6 +14,12 @@ tests:
           value: "1"
       - isNotEmpty:
           path: data["imp-overview.json"]
+      - isNotEmpty:
+          path: data["imp-admin.json"]
+      - isNotEmpty:
+          path: data["imp-workloads.json"]
+      - isNotEmpty:
+          path: data["imp-github-runners.json"]
 
   - it: renders no resources when grafana.enabled=false
     template: templates/grafana/dashboard.yaml
diff --git a/internal/agent/metrics.go b/internal/agent/metrics.go
index d06b9ec..b9128aa 100644
--- a/internal/agent/metrics.go
+++ b/internal/agent/metrics.go
@@ -36,6 +36,8 @@ type VMMetricsCollector struct {
 	vmStates     map[string]vmStateEntry // "ns/name" → {state, node}
 	guestMetrics map[string]*guestData   // "ns/name" → data
 	gatherer     prometheus.Gatherer
+	runnerActive metric.Int64UpDownCounter
+	runnerTotal  metric.Int64Counter
 }
 
 // NewVMMetricsCollector creates a new collector using the provided OTel meter.
@@ -47,6 +49,14 @@ func NewVMMetricsCollector(meter metric.Meter, gatherer prometheus.Gatherer) *VM
 		guestMetrics: make(map[string]*guestData),
 		gatherer:     gatherer,
 	}
+	c.runnerActive, _ = meter.Int64UpDownCounter(
+		"imp_runner_handoffs_active",
+		metric.WithDescription("Runner handoffs currently executing in guest VMs."),
+	)
+	c.runnerTotal, _ = meter.Int64Counter(
+		"imp_runner_handoffs",
+		metric.WithDescription("Completed runner handoffs by outcome."),
+	)
 
 	_, _ = meter.Float64ObservableGauge(
 		"imp_vm_state",
@@ -175,6 +185,33 @@ func (c *VMMetricsCollector) ClearVM(key string) {
 	delete(c.guestMetrics, key)
 }
 
+// RecordRunnerHandoffStarted records a guest runner handoff that has begun.
+func (c *VMMetricsCollector) RecordRunnerHandoffStarted(key, node string) {
+	ns, name := splitKey(key)
+	c.runnerActive.Add(context.Background(), 1, metric.WithAttributes(
+		attribute.String("namespace", ns),
+		attribute.String("impvm", name),
+		attribute.String("node", node),
+	))
+}
+
+// RecordRunnerHandoffFinished records a completed guest runner handoff.
+func (c *VMMetricsCollector) RecordRunnerHandoffFinished(key, node, outcome string) {
+	ns, name := splitKey(key)
+	attrs := metric.WithAttributes(
+		attribute.String("namespace", ns),
+		attribute.String("impvm", name),
+		attribute.String("node", node),
+	)
+	c.runnerActive.Add(context.Background(), -1, attrs)
+	c.runnerTotal.Add(context.Background(), 1, metric.WithAttributes(
+		attribute.String("namespace", ns),
+		attribute.String("impvm", name),
+		attribute.String("node", node),
+		attribute.String("outcome", outcome),
+	))
+}
+
 // NewMetricsHandlerWithCollector returns an HTTP handler for the collector's Prometheus registry.
 func NewMetricsHandlerWithCollector(c *VMMetricsCollector) http.Handler {
 	return promhttp.HandlerFor(c.gatherer, promhttp.HandlerOpts{})
diff --git a/internal/agent/metrics_test.go b/internal/agent/metrics_test.go
index bfc051b..c0e6027 100644
--- a/internal/agent/metrics_test.go
+++ b/internal/agent/metrics_test.go
@@ -81,3 +81,20 @@ func TestVMMetrics_guestMetricsAppear(t *testing.T) {
 		}
 	}
 }
+
+func TestVMMetrics_runnerHandoffMetricsAppear(t *testing.T) {
+	mc, h := newTestCollector(t)
+	mc.RecordRunnerHandoffStarted("default/test-vm", "test-node")
+	mc.RecordRunnerHandoffFinished("default/test-vm", "test-node", "success")
+
+	req := httptest.NewRequest(http.MethodGet, "/metrics", nil)
+	w := httptest.NewRecorder()
+	h.ServeHTTP(w, req)
+
+	body := w.Body.String()
+	for _, metric := range []string{"imp_runner_handoffs_active", "imp_runner_handoffs_total"} {
+		if !strings.Contains(body, metric) {
+			t.Errorf("expected %s in output, got:\n%s", metric, body)
+		}
+	}
+}
diff --git a/internal/agent/reconciler.go b/internal/agent/reconciler.go
index d57cb56..121f9c3 100644
--- a/internal/agent/reconciler.go
+++ b/internal/agent/reconciler.go
@@ -264,6 +264,19 @@ func (r *ImpVMReconciler) maybeLaunchRunner(ctx context.Context, vm *impdevv1alp
 		Reader: r.APIReader,
 		Sock:   sockPath,
 	}
+	if r.Metrics != nil {
+		key := vm.Namespace + "/" + vm.Name
+		launch.OnStarted = func() {
+			r.Metrics.RecordRunnerHandoffStarted(key, r.NodeName)
+		}
+		launch.OnFinished = func(res runnerlaunch.Result) {
+			outcome := "success"
+			if res.Err != nil || res.ExitCode != 0 {
+				outcome = "failed"
+			}
+			r.Metrics.RecordRunnerHandoffFinished(key, r.NodeName, outcome)
+		}
+	}
 	go func() {
 		launchRunner := r.runnerLaunch
 		if launchRunner == nil {
diff --git a/internal/agent/runnerlaunch/launcher.go b/internal/agent/runnerlaunch/launcher.go
index d6cd63e..afeb18f 100644
--- a/internal/agent/runnerlaunch/launcher.go
+++ b/internal/agent/runnerlaunch/launcher.go
@@ -125,12 +125,16 @@ type Launcher struct {
 	Reader client.Reader // uncached reader; avoids a cluster-wide Secret informer
 	Sock   string        // path to the VM's VSOCK unix socket proxy
 	Log    *slog.Logger
+	// OnStarted and OnFinished observe a real, de-duplicated handoff attempt.
+	// They must not read or log the JIT payload.
+	OnStarted  func()
+	OnFinished func(Result)
 }
 
 // Run performs the handoff and blocks until the runner exits (or fails).
 // It is safe to call in a goroutine. Cancel ctx to abandon (the guest keeps
 // running; Firecracker exit handling stays with the reconciler).
-func (l *Launcher) Run(ctx context.Context, vm *impv1alpha1.ImpVM) Result {
+func (l *Launcher) Run(ctx context.Context, vm *impv1alpha1.ImpVM) (result Result) {
 	log := l.Log
 	if log == nil {
 		log = slog.Default()
@@ -144,6 +148,14 @@ func (l *Launcher) Run(ctx context.Context, vm *impv1alpha1.ImpVM) Result {
 		return Result{}
 	}
 	defer launches.Delete(launchKey)
+	if l.OnStarted != nil {
+		l.OnStarted()
+	}
+	defer func() {
+		if l.OnFinished != nil {
+			l.OnFinished(result)
+		}
+	}()
 
 	// The accepted marker is written before Exec because the unary RPC has no
 	// acknowledgement boundary: it returns only when the guest process exits.