#!/usr/bin/env python3
"""Evaluate all pending Engrams (verdict != confirmed_true/false) and set verdict automatically."""

import sys
import json
from pathlib import Path

BRAIN_DIR = Path("/root/.openclaw/workspace/second-brain")
sys.path.insert(0, str(BRAIN_DIR))
from src.store import EngramStore

DB_PATH = BRAIN_DIR / "data" / "brain.sqlite"
store = EngramStore(str(DB_PATH))

# Hole alle Engrams, die nicht confirmed_true oder confirmed_false sind
cursor = store._conn.execute("""
    SELECT id, metadata_json, correctness_json FROM engrams
    WHERE json_extract(correctness_json, '$.verdict') NOT IN ('confirmed_true', 'confirmed_false')
""")
rows = cursor.fetchall()
print(f"Pendings (nicht confirmed_true/false): {len(rows)}")

evaluated = 0
true_count = 0
false_count = 0
skipped = 0

for eid, meta_json, corr_json in rows:
    try:
        meta = json.loads(meta_json) if meta_json else {}
        corr = json.loads(corr_json) if corr_json else {}
        source = meta.get("source", "")
        tags = meta.get("tags", [])
        if isinstance(tags, str):
            tags = [tags]

        # Entscheidungsregeln
        verdict = None
        reason = None

        if source == "worker":
            verdict = "confirmed_true"
            reason = "source=worker (system task)"
        elif source == "memory":
            safe_tags = ["ops", "housekeeping", "sop", "meta", "system", "documentation", "guide"]
            if any(t in safe_tags for t in tags):
                verdict = "confirmed_true"
                reason = f"memory with safe tags"
            else:
                # Memory ohne bedenkliche Tags → tendenziell true
                verdict = "confirmed_true"
                reason = "memory (no negative tags)"
        elif source == "agent":
            verdict = "confirmed_true"
            reason = "source=agent (AI output)"
        else:
            # Prüfe auf Fehler-Tags
            error_tags = ["error", "failure", "exception", "bug", "critical", "issue", "problem"]
            if any(t in error_tags for t in tags):
                verdict = "confirmed_false"
                reason = f"error tags present"
            else:
                # Default: true (dokumentarisch)
                verdict = "confirmed_true"
                reason = "default (no negative indicators)"

        if verdict:
            eg = store.get(eid)
            if eg is None:
                skipped += 1
                continue
            eg.correctness.verdict = verdict
            if verdict == "confirmed_true":
                eg.correctness.confirmed = True
                true_count += 1
            else:
                eg.correctness.confirmed = False
                false_count += 1
            store.save(eg)
            evaluated += 1
            if evaluated % 100 == 0:
                print(f"  ... {evaluated} evaluiert (true={true_count}, false={false_count})")
    except Exception as e:
        print(f"Fehler bei {eid}: {e}")

print(f"Evaluierte Engrams: {evaluated}")
print(f"  -> confirmed_true: {true_count}")
print(f"  -> confirmed_false: {false_count}")
print(f"  -> übersprungen: {skipped}")