Faktencheck-Deduplizierung und Auto-Resolve implementiert

3-Ebenen-System gegen Duplikate:
1. Pre-Dedup: LLM-Antwort wird vor DB-Insert dedupliziert (deduplicate_new_facts)
2. Auto-Resolve: Bestaetigte Fakten loesen automatisch stale developing/unconfirmed Fakten auf
3. Periodische Konsolidierung: Haiku clustert alle 6h semantische Duplikate und entfernt sie

Verbessertes Claim-Matching: SequenceMatcher (70%) + Jaccard-Keyword-Overlap (30%)
statt reinem SequenceMatcher. Threshold von 0.7 auf 0.75 erhoeht.

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

src/agents/factchecker.py

@@ -168,49 +168,120 @@ Jedes Element hat:
 Antworte NUR mit dem JSON-Array."""
 
 
+# --- Stopwords fuer Keyword-Extraktion ---
+_STOPWORDS = frozenset({
+    "der", "die", "das", "ein", "eine", "und", "oder", "von", "nach", "bei", "mit",
+    "wurde", "wird", "haben", "sein", "dass", "ist", "sind", "hat", "vor", "fuer",
+    "den", "dem", "des", "sich", "auf", "als", "auch", "noch", "nicht", "aber",
+    "ueber", "durch", "einer", "einem", "eines", "werden", "wurde", "waren",
+    "the", "and", "was", "has", "been", "have", "that", "with", "from", "for",
+    "are", "were", "this", "which", "into", "their", "than", "about",
+})
+
+STATUS_PRIORITY = {
+    "confirmed": 5, "established": 5,
+    "contradicted": 4, "disputed": 4,
+    "unconfirmed": 3, "unverified": 3,
+    "developing": 1,
+}
+
+
 def normalize_claim(claim: str) -> str:
-    """Normalisiert einen Claim für Ähnlichkeitsvergleich."""
+    """Normalisiert einen Claim fuer Aehnlichkeitsvergleich."""
     c = claim.lower().strip()
-    # Umlaute normalisieren
+    c = c.replace("\u00e4", "ae").replace("\u00f6", "oe").replace("\u00fc", "ue").replace("\u00df", "ss")
-    c = c.replace("ä", "ae").replace("ö", "oe").replace("ü", "ue").replace("ß", "ss")
     c = re.sub(r'[^\w\s]', '', c)
     c = re.sub(r'\s+', ' ', c).strip()
     return c
 
 
-def find_matching_claim(new_claim: str, existing_claims: list[dict], threshold: float = 0.7) -> dict | None:
+def _keyword_set(text: str) -> set[str]:
-    """Findet den besten passenden bestehenden Claim per Fuzzy-Matching.
+    """Extrahiert signifikante Woerter fuer Overlap-Vergleich."""
+    words = set(normalize_claim(text).split())
+    return {w for w in words if len(w) >= 4 and w not in _STOPWORDS}
 
-    Args:
-        new_claim: Der neue Claim-Text
-        existing_claims: Liste von Dicts mit mindestens {"id", "claim", "status"}
-        threshold: Mindest-Ähnlichkeit (0.0-1.0), Standard 0.7
 
-    Returns:
+def find_matching_claim(new_claim: str, existing_claims: list[dict], threshold: float = 0.75) -> dict | None:
-        Das passende Dict oder None wenn kein Match über dem Schwellwert
+    """Findet den besten passenden bestehenden Claim per kombiniertem Scoring.
+
+    Verwendet SequenceMatcher (70%) + Jaccard-Keyword-Overlap (30%) fuer robusteres Matching.
     """
     norm_new = normalize_claim(new_claim)
     if not norm_new:
         return None
 
+    kw_new = _keyword_set(new_claim)
     best_match = None
-    best_ratio = 0.0
+    best_score = 0.0
 
     for existing in existing_claims:
         norm_existing = normalize_claim(existing.get("claim", ""))
         if not norm_existing:
             continue
-        ratio = SequenceMatcher(None, norm_new, norm_existing).ratio()
+
-        if ratio > best_ratio:
+        # Fruehzeitiger Abbruch bei grossem Laengenunterschied
-            best_ratio = ratio
+        len_ratio = len(norm_new) / len(norm_existing) if norm_existing else 0
+        if len_ratio > 2.5 or len_ratio < 0.4:
+            continue
+
+        seq_ratio = SequenceMatcher(None, norm_new, norm_existing).ratio()
+
+        kw_existing = _keyword_set(existing.get("claim", ""))
+        kw_union = kw_new | kw_existing
+        jaccard = len(kw_new & kw_existing) / len(kw_union) if kw_union else 0.0
+
+        combined = 0.7 * seq_ratio + 0.3 * jaccard
+
+        if combined > best_score:
+            best_score = combined
             best_match = existing
 
-    if best_ratio >= threshold:
+    if best_score >= threshold:
-        logger.debug(f"Claim-Match ({best_ratio:.2f}): '{new_claim[:50]}...' → '{best_match['claim'][:50]}...'")
+        logger.debug(
+            f"Claim-Match ({best_score:.2f}): "
+            f"'{new_claim[:50]}...' -> '{best_match['claim'][:50]}...'"
+        )
         return best_match
     return None
 
 
+def deduplicate_new_facts(facts: list[dict], threshold: float = 0.70) -> list[dict]:
+    """Dedupliziert Fakten aus einer einzelnen LLM-Antwort vor dem DB-Insert.
+
+    Clustert aehnliche Claims und behaelt pro Cluster den mit dem
+    hoechsten Status und den meisten Quellen.
+    """
+    if not facts:
+        return []
+
+    clusters: list[list[dict]] = []
+    for fact in facts:
+        matched_cluster = None
+        for cluster in clusters:
+            if find_matching_claim(fact.get("claim", ""), cluster, threshold=threshold):
+                matched_cluster = cluster
+                break
+        if matched_cluster is not None:
+            matched_cluster.append(fact)
+        else:
+            clusters.append([fact])
+
+    result = []
+    for cluster in clusters:
+        best = max(cluster, key=lambda f: (
+            STATUS_PRIORITY.get(f.get("status", "developing"), 0),
+            f.get("sources_count", 0),
+        ))
+        result.append(best)
+
+    if len(result) < len(facts):
+        logger.info(
+            f"Fakten-Dedup: {len(facts)} -> {len(result)} "
+            f"(-{len(facts) - len(result)} Duplikate)"
+        )
+    return result
+
+
 class FactCheckerAgent:
     """Prüft Fakten über Claude CLI gegen unabhängige Quellen."""
 

src/agents/orchestrator.py

@@ -9,7 +9,7 @@ from typing import Optional
 from urllib.parse import urlparse, urlunparse
 
 from agents.claude_client import UsageAccumulator
-from agents.factchecker import find_matching_claim
+from agents.factchecker import find_matching_claim, deduplicate_new_facts
 from source_rules import (
     _detect_category,
     _extract_domain,
@@ -890,6 +890,9 @@ class AgentOrchestrator:
                     all_articles_for_fc = [dict(row) for row in await cursor.fetchall()]
                     fact_checks, fc_usage = await factchecker.check(title, all_articles_for_fc, incident_type)
 
+                # Pre-Dedup: Duplikate aus LLM-Antwort entfernen
+                fact_checks = deduplicate_new_facts(fact_checks)
+
                 if fc_usage:
                     usage_acc.add(fc_usage)
 

src/main.py

@@ -21,6 +21,7 @@ from auth import decode_token
 from agents.orchestrator import orchestrator
 from services.source_health import run_health_checks, get_health_summary
 from services.source_suggester import generate_suggestions
+from services.fact_consolidation import consolidate_fact_checks
 
 # Logging
 os.makedirs(LOG_DIR, exist_ok=True)

src/services/fact_consolidation.py

@@ -0,0 +1,243 @@
+"""Periodische Faktencheck-Konsolidierung via Haiku.
+
+Erkennt und merged semantische Duplikate unter Faktenchecks.
+Laeuft als Scheduler-Job alle 6 Stunden.
+"""
+import json
+import logging
+import re
+from datetime import datetime
+
+from config import CLAUDE_MODEL_FAST, TIMEZONE
+from database import get_db
+from agents.claude_client import call_claude
+
+logger = logging.getLogger("osint.fact_consolidation")
+
+STATUS_PRIORITY = {
+    "confirmed": 5, "established": 5,
+    "contradicted": 4, "disputed": 4,
+    "unconfirmed": 3, "unverified": 3,
+    "developing": 1,
+}
+
+CONSOLIDATION_PROMPT = (
+    "Du bist ein Deduplizierungs-Agent. Du bekommst eine Liste von Faktenchecks (ID + Claim + Status).\n"
+    "Finde Gruppen von Fakten, die inhaltlich DASSELBE aussagen (auch bei unterschiedlicher Formulierung).\n\n"
+    "REGELN:\n"
+    '- Gleicher Sachverhalt = gleiche Gruppe (z.B. "Khamenei wurde getoetet" und "Chamenei bei Angriff ums Leben gekommen")\n'
+    "- Unterschiedliche Detailtiefe zum SELBEN Fakt = gleiche Gruppe\n"
+    '- VERSCHIEDENE Sachverhalte = verschiedene Gruppen (z.B. "Angriff auf Isfahan" vs "Angriff auf Teheran")\n'
+    "- Eine Gruppe muss mindestens 2 Eintraege haben\n\n"
+    "Antworte NUR als JSON-Array von Gruppen. Jede Gruppe ist ein Array von IDs:\n"
+    "[[1,5,12], [3,8], [20,25,30]]\n\n"
+    "Wenn keine Duplikate: antworte mit []\n\n"
+    "FAKTEN:\n{facts_text}"
+)
+
+
+async def _ask_haiku_for_clusters(facts: list[dict]) -> list[list[int]]:
+    """Fragt Haiku welche Fakten semantische Duplikate sind."""
+    facts_text = "\n".join(
+        f'ID={f["id"]} [{f["status"]}]: {f["claim"]}'
+        for f in facts
+    )
+    prompt = CONSOLIDATION_PROMPT.format(facts_text=facts_text)
+
+    try:
+        result, usage = await call_claude(prompt, tools=None, model=CLAUDE_MODEL_FAST)
+        data = json.loads(result)
+        if isinstance(data, list) and all(isinstance(g, list) for g in data):
+            return data
+    except json.JSONDecodeError:
+        match = re.search(r'\[.*\]', result, re.DOTALL)
+        if match:
+            try:
+                data = json.loads(match.group())
+                if isinstance(data, list):
+                    return data
+            except json.JSONDecodeError:
+                pass
+    except Exception as e:
+        logger.error(f"Haiku-Cluster-Anfrage fehlgeschlagen: {e}")
+
+    return []
+
+
+async def consolidate_fact_checks(max_per_incident: int = 25):
+    """Konsolidiert doppelte Faktenchecks via Haiku-Clustering."""
+    db = await get_db()
+    try:
+        cursor = await db.execute(
+            "SELECT incident_id, COUNT(*) as cnt FROM fact_checks "
+            "GROUP BY incident_id HAVING cnt > ?",
+            (max_per_incident,),
+        )
+        bloated = [dict(row) for row in await cursor.fetchall()]
+
+        if not bloated:
+            logger.info("Faktencheck-Konsolidierung: keine aufgeblaehten Incidents gefunden")
+            return 0
+
+        total_removed = 0
+
+        for row in bloated:
+            incident_id = row["incident_id"]
+
+            # Pruefe ob gerade ein Refresh laeuft
+            cursor_rl = await db.execute(
+                "SELECT COUNT(*) as cnt FROM refresh_log "
+                "WHERE incident_id = ? AND status = 'running'",
+                (incident_id,),
+            )
+            rl_row = await cursor_rl.fetchone()
+            if rl_row and rl_row["cnt"] > 0:
+                logger.info(
+                    f"Incident {incident_id} hat laufenden Refresh, ueberspringe"
+                )
+                continue
+
+            cursor2 = await db.execute(
+                "SELECT id, claim, status, sources_count, evidence, "
+                "checked_at, status_history "
+                "FROM fact_checks WHERE incident_id = ? "
+                "ORDER BY checked_at DESC",
+                (incident_id,),
+            )
+            all_facts = [dict(r) for r in await cursor2.fetchall()]
+
+            if len(all_facts) <= max_per_incident:
+                continue
+
+            # Haiku in Batches fragen
+            all_clusters = []
+            batch_size = 80
+            for i in range(0, len(all_facts), batch_size):
+                batch = all_facts[i:i + batch_size]
+                clusters = await _ask_haiku_for_clusters(batch)
+                all_clusters.extend(clusters)
+
+            # Pro Cluster: besten behalten, Rest loeschen
+            ids_to_delete = []
+            facts_by_id = {f["id"]: f for f in all_facts}
+
+            for cluster_ids in all_clusters:
+                valid_ids = [cid for cid in cluster_ids if cid in facts_by_id]
+                if len(valid_ids) <= 1:
+                    continue
+
+                cluster_facts = [facts_by_id[cid] for cid in valid_ids]
+                best = max(cluster_facts, key=lambda f: (
+                    STATUS_PRIORITY.get(f["status"], 0),
+                    f.get("sources_count", 0),
+                    f.get("checked_at", ""),
+                ))
+
+                for fact in cluster_facts:
+                    if fact["id"] != best["id"]:
+                        ids_to_delete.append(fact["id"])
+
+            if ids_to_delete:
+                unique_ids = list(set(ids_to_delete))
+                placeholders = ",".join("?" * len(unique_ids))
+                await db.execute(
+                    f"DELETE FROM fact_checks WHERE id IN ({placeholders})",
+                    unique_ids,
+                )
+                total_removed += len(unique_ids)
+                logger.info(
+                    f"Incident {incident_id}: {len(unique_ids)} Duplikate entfernt, "
+                    f"{len(all_facts) - len(unique_ids)} verbleiben"
+                )
+
+        await db.commit()
+        if total_removed > 0:
+            logger.info(
+                f"Faktencheck-Konsolidierung: {total_removed} Duplikate entfernt"
+            )
+        return total_removed
+
+    except Exception as e:
+        logger.error(
+            f"Faktencheck-Konsolidierung Fehler: {e}", exc_info=True
+        )
+        return 0
+    finally:
+        await db.close()
+
+
+async def auto_resolve_stale_facts(incident_id: int, confirmed_claims: list[dict], db):
+    """Loest veraltete developing/unconfirmed Fakten automatisch auf,
+    wenn ein bestaetigter Match gefunden wird.
+
+    Wird vom Orchestrator nach jedem Faktencheck aufgerufen.
+    """
+    if not confirmed_claims:
+        return 0
+
+    from agents.factchecker import find_matching_claim
+
+    now = datetime.now(TIMEZONE).strftime('%Y-%m-%d %H:%M:%S')
+
+    cursor = await db.execute(
+        "SELECT id, claim, status, status_history FROM fact_checks "
+        "WHERE incident_id = ? "
+        "AND status IN ('developing', 'unconfirmed', 'unverified')",
+        (incident_id,),
+    )
+    stale_facts = [dict(row) for row in await cursor.fetchall()]
+
+    if not stale_facts:
+        return 0
+
+    resolved_count = 0
+    resolved_ids = set()
+
+    for confirmed_fc in confirmed_claims:
+        confirmed_claim_text = confirmed_fc.get("claim", "")
+        for stale in stale_facts:
+            if stale["id"] in resolved_ids:
+                continue
+            # Niedrigerer Threshold (0.65) fuer aggressiveres Auto-Resolve
+            if find_matching_claim(
+                confirmed_claim_text, [stale], threshold=0.65
+            ):
+                try:
+                    history = json.loads(
+                        stale.get("status_history") or "[]"
+                    )
+                except (ValueError, TypeError):
+                    history = []
+
+                new_status = (
+                    "confirmed"
+                    if confirmed_fc.get("status") == "confirmed"
+                    else "established"
+                )
+                history.append({
+                    "status": new_status,
+                    "at": now,
+                    "reason": "auto-resolved",
+                })
+
+                await db.execute(
+                    "UPDATE fact_checks SET status = ?, "
+                    "evidence = COALESCE(evidence, '') "
+                    "|| ' [Auto-aufgeloest: uebereinstimmender Fakt bestaetigt]', "
+                    "status_history = ?, checked_at = ? WHERE id = ?",
+                    (new_status, json.dumps(history), now, stale["id"]),
+                )
+                resolved_ids.add(stale["id"])
+                resolved_count += 1
+                logger.info(
+                    f"Auto-resolved Fakt #{stale['id']}: "
+                    f"'{stale['claim'][:60]}...' -> {new_status}"
+                )
+
+    if resolved_count > 0:
+        logger.info(
+            f"Auto-Resolve: {resolved_count} veraltete Fakten "
+            f"fuer Incident {incident_id} aufgeloest"
+        )
+
+    return resolved_count