Merge pull request 'feat(halacha): #84.7 — מדדי-תור (throughput + איכות)' (#193) from worktree-halacha-queue-metrics into main

mcp-server/src/legal_mcp/services/metrics.py

@@ -103,12 +103,25 @@ async def get_case_metrics(case_id: UUID) -> dict:
     return metrics
 
 
+def _median(values: list[float]) -> float | None:
+    """Median of a numeric list (None if empty). Pure — unit-tested."""
+    s = sorted(v for v in values if v is not None)
+    if not s:
+        return None
+    mid = len(s) // 2
+    return s[mid] if len(s) % 2 else (s[mid - 1] + s[mid]) / 2
+
+
 async def halacha_backlog(conn) -> dict:
     """תור אישור-ההלכות (GAP-14 / INV-QA1 / G10) — נראות ה-backlog האנושי.
 
     הלכות נכנסות כ-`pending_review` ובלתי-נראות לחיפוש עד אישור היו"ר; בלי ספירה
     גלויה, אישור-חסר נשאר סמוי (10/19 התגלה במקרה). מקבל connection פתוח כדי
     שאפשר יהיה לשלב בסנאפ-שוט קיים (get_dashboard, /api/system/diagnostics).
+
+    כולל גם מדדי-תור (#84.7): throughput (24ש'/7ימים), יחסי approve/reject/defer,
+    זמן-חציוני-לפריט (פער בין החלטות עוקבות בתוך session של 30 דק'), ופילוח
+    מי-החליט (panel/auto/chair) — כדי לראות גם מהירות וגם איכות, לא רק backlog.
     """
     rows = await conn.fetch(
         "SELECT review_status, COUNT(*) AS n FROM halachot GROUP BY review_status"
@@ -132,6 +145,38 @@ async def halacha_backlog(conn) -> dict:
     )
     pending_total = counts.get("pending_review", 0)
     reviewed = counts.get("approved", 0) + counts.get("rejected", 0) + counts.get("published", 0)
+
+    # ── #84.7 queue throughput + quality ──────────────────────────────────────
+    # throughput windows (decisions = anything with a reviewed_at stamp)
+    tp = await conn.fetchrow(
+        "SELECT COUNT(*) FILTER (WHERE reviewed_at >= now() - interval '24 hours') AS d24, "
+        "       COUNT(*) FILTER (WHERE reviewed_at >= now() - interval '7 days') AS d7 "
+        "FROM halachot WHERE reviewed_at IS NOT NULL"
+    )
+    # who decided — panel (tri-model), auto (confidence gate), chair (human), other
+    who_rows = await conn.fetch(
+        "SELECT CASE "
+        "  WHEN reviewer LIKE 'panel:%' THEN 'panel' "
+        "  WHEN reviewer LIKE 'auto-approved%' THEN 'auto' "
+        "  WHEN reviewer LIKE 'chair%' THEN 'chair' "
+        "  ELSE 'other' END AS who, COUNT(*) AS n "
+        "FROM halachot WHERE reviewed_at IS NOT NULL GROUP BY 1"
+    )
+    by_reviewer = {r["who"]: r["n"] for r in who_rows}
+    # time-per-item proxy: median seconds between consecutive HAND-PACED
+    # decisions — gaps in [1s, 30min]. Excludes 0-second gaps (batch operations
+    # like panel/auto stamp many rows with the same reviewed_at) and >30-min gaps
+    # (between sessions), so the number reflects interactive review pacing, not
+    # machine throughput. None when the queue is entirely batch-decided.
+    gap_rows = await conn.fetch(
+        "SELECT EXTRACT(EPOCH FROM (reviewed_at - prev)) AS gap FROM ("
+        "  SELECT reviewed_at, LAG(reviewed_at) OVER (ORDER BY reviewed_at) AS prev "
+        "  FROM halachot WHERE reviewed_at IS NOT NULL"
+        ") t WHERE prev IS NOT NULL "
+        "AND reviewed_at - prev BETWEEN interval '1 second' AND interval '30 minutes'"
+    )
+    median_secs = _median([float(r["gap"]) for r in gap_rows if r["gap"] is not None])
+
     return {
         "pending_review": pending_total,
         "pending_clean": pending_clean,           # real review candidates (#84.1)
@@ -143,8 +188,16 @@ async def halacha_backlog(conn) -> dict:
         "total": sum(counts.values()),
         "reviewed_total": reviewed,
         "approve_ratio": round(counts.get("approved", 0) / reviewed, 3) if reviewed else None,
+        "reject_ratio": round(counts.get("rejected", 0) / reviewed, 3) if reviewed else None,
+        "defer_ratio": (round(counts.get("deferred", 0) / (reviewed + counts.get("deferred", 0)), 3)
+                        if (reviewed + counts.get("deferred", 0)) else None),
         "pending_by_flag": {r["flag"]: r["n"] for r in flag_rows},
         "oldest_pending_at": oldest.isoformat() if oldest else None,
+        # #84.7 throughput + quality
+        "throughput_24h": tp["d24"] if tp else 0,
+        "throughput_7d": tp["d7"] if tp else 0,
+        "median_seconds_per_decision": round(median_secs, 1) if median_secs is not None else None,
+        "by_reviewer": by_reviewer,
     }
 
 

mcp-server/tests/test_metrics_median.py

@@ -0,0 +1,24 @@
+"""Test for #84.7 — _median helper used by the queue-metrics time-per-item proxy."""
+
+from __future__ import annotations
+
+import pytest
+
+from legal_mcp.services import metrics
+
+
+def test_median_empty_is_none():
+    assert metrics._median([]) is None
+    assert metrics._median([None, None]) is None
+
+
+def test_median_odd():
+    assert metrics._median([3.0, 1.0, 2.0]) == 2.0
+
+
+def test_median_even_averages_middle():
+    assert metrics._median([4.0, 1.0, 3.0, 2.0]) == pytest.approx(2.5)
+
+
+def test_median_ignores_none():
+    assert metrics._median([None, 5.0, None, 1.0, 3.0]) == 3.0