Sync main.py from blood-brain-omics repo

main.py

@@ -541,6 +541,92 @@ def h2h_blood_summary(
     }
 
 
+@app.get("/api/v1/h2h/blood/per_outcome")
+def h2h_blood_per_outcome(
+    pair: str = Query(..., description="H2H pair, e.g. ClinLabs_vs_Metabolon"),
+    metric: str = Query("pearson"),
+    covariates: str = Query("included", enum=["none", "included"]),
+    model: str = Query("TabPFN"),
+):
+    """Per-outcome head-to-head comparison for one blood pair.
+
+    Returns one row per brain_target with the mean and best metric value
+    for each side of the pair, plus per-target win counts. Lets the H2H
+    page show "for ClinLabs vs Metabolon, here's how each performs on
+    every outcome group" instead of just the global win count.
+    """
+    validate_metric(metric)
+    phase = "h2h_blood_withcov" if covariates == "included" else "h2h_blood"
+
+    rows = q(f"""
+        SELECT brain_target, target, blood_platform, {metric}
+        FROM results
+        WHERE phase = ? AND model = ? AND h2h_pair = ?
+        ORDER BY brain_target, target, blood_platform
+    """, [phase, model, pair]).fetchall()
+
+    # Group by (brain_target, target) → {platform: value}
+    per_target: dict[tuple[str, str], dict[str, Optional[float]]] = {}
+    platforms_seen: set[str] = set()
+    for brain_target, target, platform, val in rows:
+        per_target.setdefault((brain_target, target), {})[platform] = val
+        platforms_seen.add(platform)
+
+    if len(platforms_seen) != 2:
+        return {"pair": pair, "metric": metric, "covariates": covariates,
+                "platforms": sorted(platforms_seen), "outcomes": []}
+
+    p_a, p_b = sorted(platforms_seen)
+    asc = metric in ("mse", "mae")
+
+    # Aggregate per brain_target
+    by_outcome: dict[str, dict] = {}
+    for (brain_target, target), vals in per_target.items():
+        v_a = vals.get(p_a)
+        v_b = vals.get(p_b)
+        if v_a is None or v_b is None:
+            continue
+        bucket = by_outcome.setdefault(brain_target, {
+            "n_targets": 0, "sum_a": 0.0, "sum_b": 0.0,
+            "wins_a": 0, "wins_b": 0,
+            "best_a": None, "best_b": None,
+        })
+        bucket["n_targets"] += 1
+        bucket["sum_a"] += v_a
+        bucket["sum_b"] += v_b
+        if v_a < v_b if asc else v_a > v_b:
+            bucket["wins_a"] += 1
+        else:
+            bucket["wins_b"] += 1
+        if bucket["best_a"] is None or (v_a < bucket["best_a"] if asc else v_a > bucket["best_a"]):
+            bucket["best_a"] = v_a
+        if bucket["best_b"] is None or (v_b < bucket["best_b"] if asc else v_b > bucket["best_b"]):
+            bucket["best_b"] = v_b
+
+    outcomes = []
+    for brain_target, b in sorted(by_outcome.items()):
+        n = b["n_targets"]
+        outcomes.append({
+            "brain_target": brain_target,
+            "n_targets": n,
+            "mean_a": round(b["sum_a"] / n, 4),
+            "mean_b": round(b["sum_b"] / n, 4),
+            "best_a": round(b["best_a"], 4),
+            "best_b": round(b["best_b"], 4),
+            "wins_a": b["wins_a"],
+            "wins_b": b["wins_b"],
+        })
+
+    return {
+        "pair": pair,
+        "metric": metric,
+        "covariates": covariates,
+        "platform_a": p_a,
+        "platform_b": p_b,
+        "outcomes": outcomes,
+    }
+
+
 @app.get("/api/v1/h2h/brain")
 def h2h_brain(
     blood: str = Query(..., description="Blood platform name"),