test(fusion): end-to-end smoke + README section

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

README.md

@@ -41,6 +41,19 @@ short_description: Living decision system for BBB, EEG, and MRI clinical ML
 | 8 | Grand Finale (Multi-Modal Agents, Track 5 & Public Deploy) | Multi-modal explainers + experiments + deploy surface | Shipped |
 | 9 | Agent/RAG hardening + MRI DL decision layer | Guarded orchestration + `POST /predict/mri` ONNX surface | Shipped — 242 passed, 2 skipped |
 
+### Fusion Engine
+
+`POST /fusion/predict` (and the agent tool `run_fusion`) combines whichever of
+MRI, EEG, and clinical-test scores (MMSE, MoCA, UPDRS, gait, age) the doctor
+has uploaded into a per-disease confidence (Alzheimer's, Parkinson's, other)
+with full attribution showing how much each modality contributed. Missing
+modalities are skipped, not imputed — the engine renormalises onto whichever
+inputs are present so absence naturally lowers confidence rather than
+silently inflating it. Weights live in `src/fusion/weights.py` and are
+heuristic — adjust there. **BBB is intentionally NOT a fusion modality**:
+it is a researcher-side concern (drug permeability) and stays decoupled
+from disease classification.
+
 ## Quick Start
 
 **Prerequisite:** Python 3.10–3.12. The pinned `requirements.txt` has no cp313+ wheels;

tests/fusion/test_end_to_end.py

@@ -0,0 +1,46 @@
+"""End-to-end: agent calls run_fusion with realistic inputs, top disease is sane."""
+from __future__ import annotations
+
+import pytest
+
+from src.agents.tools import build_default_tools
+
+
+@pytest.mark.parametrize(
+    "scenario,expected_top",
+    [
+        # Strong AD signal: low MMSE + MRI flags alzheimers.
+        (
+            {
+                "mri": {
+                    "label_text": "alzheimers", "label": 1, "confidence": 0.85,
+                    "probabilities": [
+                        {"label_text": "control", "probability": 0.15},
+                        {"label_text": "alzheimers", "probability": 0.85},
+                    ],
+                },
+                "clinical": {"mmse": 14.0, "age_years": 79.0},
+            },
+            "alzheimers",
+        ),
+        # Strong PD signal: high UPDRS + slow gait + EEG flags parkinsons.
+        (
+            {
+                "eeg": {
+                    "label_text": "parkinsons", "label": 1, "confidence": 0.78,
+                    "probabilities": [
+                        {"label_text": "control", "probability": 0.22},
+                        {"label_text": "parkinsons", "probability": 0.78},
+                    ],
+                },
+                "clinical": {"updrs": 80.0, "gait_speed_m_s": 0.4, "age_years": 70.0},
+            },
+            "parkinsons",
+        ),
+    ],
+)
+def test_realistic_scenarios_pick_correct_top_disease(scenario, expected_top) -> None:
+    tools = {t.name: t for t in build_default_tools(rag_index_dir=None)}
+    tool = tools["run_fusion"]
+    out = tool.execute(tool.input_model.model_validate(scenario))
+    assert out.top_disease == expected_top