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

"""מדדי הצלחה (KPIs) לתהליך כתיבת החלטות.

מדדים:
1. אחוז שינוי — השוואת טיוטה לגרסה סופית (יעד: <10%)
2. אפס הזיות — ספירת הפניות לא מבוססות
3. מענה לכל טענה — כיסוי טענות בדיון
4. משקלות בטווח — עמידה ביחסי הזהב
5. רקע ניטרלי — ללא מילות שיפוט
6. זמן עיבוד — מקליטה עד טיוטה
"""

from __future__ import annotations

import json
import logging
from datetime import datetime
from uuid import UUID

from legal_mcp.services import db

logger = logging.getLogger(__name__)


async def get_case_metrics(case_id: UUID) -> dict:
    """חישוב מדדים לתיק בודד."""
    case = await db.get_case(case_id)
    if not case:
        raise ValueError(f"Case {case_id} not found")

    decision = await db.get_decision_by_case(case_id)
    pool = await db.get_pool()

    metrics = {
        "case_number": case["case_number"],
        "title": case.get("title", ""),
        "status": case.get("status", ""),
    }

    # 1. Change percentage (if final version exists)
    if decision and decision.get("status") == "final":
        async with pool.acquire() as conn:
            # Get draft word count
            draft_words = await conn.fetchval(
                "SELECT SUM(word_count) FROM decision_blocks WHERE decision_id = $1",
                UUID(decision["id"]),
            )
        metrics["draft_words"] = draft_words or 0
        # Change percent is stored during learning loop
        metrics["change_percent"] = None  # populated from learning_loop results
    else:
        metrics["draft_words"] = 0
        metrics["change_percent"] = None

    # 2. QA results
    async with pool.acquire() as conn:
        qa_rows = await conn.fetch(
            "SELECT check_name, passed, severity, errors FROM qa_results WHERE case_id = $1",
            case_id,
        )

    if qa_rows:
        qa_results = {}
        for row in qa_rows:
            errors = json.loads(row["errors"]) if isinstance(row["errors"], str) else row["errors"]
            qa_results[row["check_name"]] = {
                "passed": row["passed"],
                "severity": row["severity"],
                "error_count": len(errors) if errors else 0,
            }
        metrics["qa"] = qa_results
        metrics["qa_passed"] = all(r["passed"] for r in qa_results.values())
        metrics["qa_critical_failures"] = sum(
            1 for r in qa_results.values()
            if not r["passed"] and r["severity"] == "critical"
        )
    else:
        metrics["qa"] = None
        metrics["qa_passed"] = None

    # 3. Claims coverage
    claims = await db.get_claims(case_id)
    metrics["total_claims"] = len(claims)

    # 4. Documents
    docs = await db.list_documents(case_id)
    metrics["total_documents"] = len(docs)

    # 5. Processing time
    if docs and decision:
        first_doc_time = min(
            d.get("created_at", datetime.max) for d in docs
            if d.get("created_at")
        )
        decision_time = decision.get("created_at")
        if first_doc_time and decision_time:
            delta = decision_time - first_doc_time
            metrics["processing_hours"] = round(delta.total_seconds() / 3600, 1)
        else:
            metrics["processing_hours"] = None
    else:
        metrics["processing_hours"] = None

    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"
    )
    counts = {r["review_status"]: r["n"] for r in rows}
    oldest = await conn.fetchval(
        "SELECT MIN(created_at) FROM halachot WHERE review_status = 'pending_review'"
    )
    # #84.7 — split the pending bucket: how many are genuine candidates (clean)
    # vs flagged 'needs extraction fix', and the breakdown by flag, so the chair
    # sees how much of the backlog is real review vs extraction noise.
    pending_clean = await conn.fetchval(
        "SELECT COUNT(*) FROM halachot WHERE review_status = 'pending_review' "
        "AND COALESCE(array_length(quality_flags, 1), 0) = 0"
    )
    flag_rows = await conn.fetch(
        "SELECT flag, COUNT(*) AS n FROM ("
        "  SELECT unnest(quality_flags) AS flag FROM halachot "
        "  WHERE review_status = 'pending_review'"
        ") t GROUP BY flag ORDER BY n DESC"
    )
    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)
        "pending_flagged": pending_total - pending_clean,  # needs-fix bucket
        "approved": counts.get("approved", 0),
        "rejected": counts.get("rejected", 0),
        "deferred": counts.get("deferred", 0),
        "published": counts.get("published", 0),
        "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,
    }


async def get_dashboard() -> dict:
    """דשבורד כולל — סיכום מדדים על כל התיקים."""
    pool = await db.get_pool()

    async with pool.acquire() as conn:
        # Case counts by status
        status_rows = await conn.fetch(
            "SELECT status, COUNT(*) as cnt FROM cases GROUP BY status ORDER BY cnt DESC"
        )
        cases_by_status = {r["status"]: r["cnt"] for r in status_rows}

        # Total counts
        total_cases = await conn.fetchval("SELECT COUNT(*) FROM cases")
        total_docs = await conn.fetchval("SELECT COUNT(*) FROM documents")
        total_claims = await conn.fetchval("SELECT COUNT(*) FROM claims")
        total_chunks = await conn.fetchval("SELECT COUNT(*) FROM document_chunks")
        total_decisions = await conn.fetchval("SELECT COUNT(*) FROM decisions")
        total_corpus = await conn.fetchval("SELECT COUNT(*) FROM style_corpus")
        total_patterns = await conn.fetchval("SELECT COUNT(*) FROM style_patterns")
        total_case_law = await conn.fetchval("SELECT COUNT(*) FROM case_law")
        non_searchable_case_law = await conn.fetchval(
            "SELECT COUNT(*) FROM case_law WHERE NOT searchable"
        )
        cases_with_stale_blocks = await conn.fetchval(
            "SELECT COUNT(*) FROM cases WHERE blocks_stale"
        )
        stale_embedding_case_law = await conn.fetchval(
            "SELECT COUNT(*) FROM case_law "
            "WHERE coalesce(full_text,'') <> '' AND content_hash IS DISTINCT FROM indexed_hash")

        # QA summary
        qa_total = await conn.fetchval("SELECT COUNT(DISTINCT case_id) FROM qa_results")
        qa_passed = await conn.fetchval(
            """SELECT COUNT(DISTINCT case_id) FROM qa_results
               WHERE case_id NOT IN (
                   SELECT case_id FROM qa_results WHERE passed = false AND severity = 'critical'
               )"""
        )

        # Final decisions
        final_count = await conn.fetchval(
            "SELECT COUNT(*) FROM decisions WHERE status = 'final'"
        )

        # Average words per decision
        avg_words = await conn.fetchval(
            "SELECT AVG(total_words) FROM decisions WHERE total_words > 0"
        )

        # Halacha review backlog (GAP-14 / INV-QA1 / G10)
        backlog = await halacha_backlog(conn)

    return {
        "summary": {
            "total_cases": total_cases,
            "total_documents": total_docs,
            "total_claims": total_claims,
            "total_chunks": total_chunks,
            "total_decisions": total_decisions,
            "final_decisions": final_count,
            "style_corpus": total_corpus,
            "style_patterns": total_patterns,
            "case_law_entries": total_case_law,
            "non_searchable_case_law": non_searchable_case_law,
            "cases_with_stale_blocks": cases_with_stale_blocks,
            "stale_embedding_case_law": stale_embedding_case_law,
        },
        "cases_by_status": cases_by_status,
        "halacha_backlog": backlog,
        "qa": {
            "cases_validated": qa_total,
            "cases_passed": qa_passed,
            "pass_rate": round(qa_passed / qa_total * 100, 1) if qa_total else None,
        },
        "avg_decision_words": round(avg_words) if avg_words else None,
    }