feat(system): instrument worker pipeline stages

tests/test_e15_shadow_inference.py

@@ -394,3 +394,31 @@ def test_runtime_telemetry_captures_bounded_queue_snapshots() -> None:
     row = json.loads(stream.getvalue())
     assert row["schema_version"] == "missioncore.worker-runtime-telemetry/v1"
     assert row["queues"]["detector"]["maximum_depth"] == 1
+
+
+def test_stage_execution_telemetry_measures_named_spans_without_process_claims() -> None:
+    module = _module()
+    telemetry = module._StageExecutionTelemetry(("detector", "tracking"))
+
+    with telemetry.measure("detector", frame_index=42):
+        pass
+    with telemetry.measure("tracking", frame_index=42):
+        pass
+
+    snapshot = telemetry.snapshot()
+    assert snapshot["current_stage"] is None
+    assert snapshot["active_stages"] == []
+    assert snapshot["active_frame_index"] == 42
+    assert snapshot["stages"]["detector"]["activations"] == 1
+    assert snapshot["stages"]["tracking"]["activations"] == 1
+    assert snapshot["stages"]["detector"]["elapsed_seconds"] >= 0
+    assert snapshot["stages"]["tracking"]["elapsed_seconds"] >= 0
+    assert sum(
+        stage["share_percent"] for stage in snapshot["stages"].values()
+    ) == pytest.approx(100)
+
+    with (
+        pytest.raises(RuntimeError, match="unknown pipeline stage"),
+        telemetry.measure("unregistered"),
+    ):
+        pass

tests/test_system_telemetry_api.py

@@ -121,11 +121,24 @@ def _probe(
             "perception": {
                 "state": "busy",
                 "current_stage": "detector",
+                "active_stages": ["detector", "semantic-model"],
                 "active_request_id": "run-001",
                 "active_frame_index": 42,
                 "completed_runs": 3,
                 "failed_runs": 0,
                 "model_load_seconds": 10.5,
+                "stage_metrics": {
+                    "detector": {
+                        "elapsed_seconds": 2.5,
+                        "activations": 42,
+                        "share_percent": 62.5,
+                    },
+                    "semantic-model": {
+                        "elapsed_seconds": 1.5,
+                        "activations": 11,
+                        "share_percent": 37.5,
+                    },
+                },
             },
         },
     }
@@ -203,16 +216,25 @@ def test_worker_telemetry_separates_mission_core_and_external_load(
     assert runtimes["sentinel-frigate"]["external"] is True
     assert runtimes["sentinel-frigate"]["cpu_percent"] == 150
     assert document["pipeline"]["active_request_id"] == "run-001"
-    assert next(
+    detector_stage = next(
         stage
         for stage in document["pipeline"]["stages"]
         if stage["id"] == "detector"
-    )["state"] == "active"
+    )
+    assert detector_stage["state"] == "active"
+    assert detector_stage["elapsed_seconds"] == 2.5
+    assert detector_stage["activations"] == 42
+    assert detector_stage["share_percent"] == 62.5
     assert next(
         stage
         for stage in document["pipeline"]["stages"]
         if stage["id"] == "semantic-model"
     )["state"] == "active"
+    assert next(
+        stage
+        for stage in document["pipeline"]["stages"]
+        if stage["id"] == "preprocessing"
+    )["share_percent"] is None
 
 
 def test_profile_apply_fails_closed_on_wrong_node_and_keeps_old_profile(