Feat: Matching-Logik mit gewichtetem Scoring & Interaktiv-Modus (v3.0)

Diese Version überarbeitet den Kern des Matching-Algorithmus grundlegend, um die Genauigkeit drastisch zu erhöhen und die manuelle Nachbearbeitung zu reduzieren. Die Änderungen basieren auf der Analyse eines umfangreichen Testdatensatzes und setzen die neue Philosophie des "großzügigen Matchens" von wirtschaftlichen Einheiten um.

Gewichtetes Namens-Scoring (TF-IDF):
- Einzigartige Namensbestandteile (z.B. "Warema") erhalten nun ein höheres Gewicht als generische Füllwörter (z.B. "Stadtwerke", "Gruppe").
- Dies löst das Problem von Fehlzuordnungen bei häufig vorkommenden, aber nicht-identifizierenden Begriffen und verbessert die Treffsicherheit bei unklaren Firmennamen signifikant.

Golden-Rule für exakte Namens-Matches:
- Eine Namensübereinstimmung von >98% führt zu einem sofortigen "Golden Match" mit einem sehr hohen Score.
- Damit wird verhindert, dass klare Treffer durch abweichende Signale (z.B. unterschiedliche URLs von Tochterfirmen) fälschlicherweise bestraft werden.

Optionaler Interaktiver Modus:
- Kann mit dem Flag --interactive gestartet werden.
- Bei uneindeutigen Ergebnissen, bei denen die Top-Kandidaten sehr ähnliche Scores haben, hält das Skript an und ermöglicht dem Benutzer die direkte Auswahl des korrekten Matches aus einer übersichtlichen Liste.

Überarbeitete Scoring-Formel:
- Die Gesamtbewertung wurde neu balanciert, um dem jetzt deutlich aussagekräftigeren Namens-Score mehr Gewicht zu verleihen.

duplicate_checker.py

@@ -1,3 +1,12 @@
+# duplicate_checker.py v3.0
+# Build timestamp is injected into logfile name.
+
+# --- NEUE FEATURES v3.0 ---
+# - Golden-Rule: Fast exakte Namens-Matches (>98%) werden immer als Treffer gewertet.
+# - Weighted Scoring (TF-IDF): Einzigartige Wörter im Firmennamen erhalten mehr Gewicht als häufige Füllwörter.
+# - Interaktiver Modus: Bei unklaren Fällen kann der Nutzer manuell den besten Kandidaten auswählen.
+# - Umfassend überarbeitete Scoring-Logik für höhere Präzision.
+
 import os
 import sys
 import re
@@ -5,6 +14,7 @@ import argparse
 import json
 import logging
 import pandas as pd
+import math
 from datetime import datetime
 from collections import Counter
 from thefuzz import fuzz
@@ -18,7 +28,6 @@ def update_status(job_id, status, progress_message):
     if not job_id: return
     status_file = os.path.join(STATUS_DIR, f"{job_id}.json")
     try:
-        # Lese alte Daten, um Action und Startzeit zu erhalten
         try:
             with open(status_file, 'r') as f:
                 data = json.load(f)
@@ -32,25 +41,26 @@ def update_status(job_id, status, progress_message):
     except Exception as e:
         logging.error(f"Konnte Statusdatei für Job {job_id} nicht schreiben: {e}")
 
-
-# duplicate_checker.py v2.15
-# Quality-first ++: Domain-Gate, Location-Penalties, Smart Blocking (IDF-light),
-# Serp-Trust, Weak-Threshold, City-Bias-Guard, Prefilter tightened, Metrics
-# Build timestamp is injected into logfile name.
-
 # --- Konfiguration ---
 CRM_SHEET_NAME      = "CRM_Accounts"
 MATCHING_SHEET_NAME = "Matching_Accounts"
-SCORE_THRESHOLD     = 80      # Standard-Schwelle
+LOG_DIR             = "Log"
-SCORE_THRESHOLD_WEAK= 95      # Schwelle, wenn weder Domain noch (City&Country) matchen
-MIN_NAME_FOR_DOMAIN = 70      # Domain-Score nur, wenn Name >= 70 ODER Ort+Land matchen
-CITY_MISMATCH_PENALTY    = 30
-COUNTRY_MISMATCH_PENALTY = 40
-PREFILTER_MIN_PARTIAL    = 70  # (vorher 60)
-PREFILTER_LIMIT          = 30  # (vorher 50)
-LOG_DIR                  = "Log"
 now = datetime.now().strftime('%Y-%m-%d_%H-%M')
-LOG_FILE                 = f"{now}_duplicate_check_v2.15.txt"
+LOG_FILE            = f"{now}_duplicate_check_v3.0.txt"
+
+# Scoring-Konfiguration
+SCORE_THRESHOLD     = 100     # Standard-Schwelle für einen Match
+SCORE_THRESHOLD_WEAK= 130     # Schwelle für Matches ohne Domain oder Ort
+GOLDEN_MATCH_RATIO  = 98      # Ratio, ab der ein Namens-Match als "Golden Match" gilt
+GOLDEN_MATCH_SCORE  = 300     # Score, der bei einem Golden Match vergeben wird
+
+# Interaktiver Modus Konfiguration
+INTERACTIVE_SCORE_MIN   = 100 # Mindestscore des besten Kandidaten, um den interaktiven Modus zu triggern
+INTERACTIVE_SCORE_DIFF  = 20  # Maximaler Score-Unterschied zum zweitbesten Kandidaten, um den Modus zu triggern
+
+# Prefilter-Konfiguration
+PREFILTER_MIN_PARTIAL   = 70
+PREFILTER_LIMIT         = 30
 
 # --- Logging Setup ---
 if not os.path.exists(LOG_DIR):
@@ -71,7 +81,7 @@ fh.setFormatter(formatter)
 root.addHandler(fh)
 logger = logging.getLogger(__name__)
 logger.info(f"Logging to console and file: {log_path}")
-logger.info(f"Starting duplicate_checker.py v2.15 | Build: {now}")
+logger.info(f"Starting duplicate_checker.py v3.0 | Build: {now}")
 
 # --- SerpAPI Key laden ---
 try:
@@ -89,119 +99,122 @@ STOP_TOKENS_BASE = {
     'holding','gruppe','group','international','solutions','solution','service','services',
     'deutschland','austria','germany','technik','technology','technologies','systems','systeme',
     'logistik','logistics','industries','industrie','management','consulting','vertrieb','handel',
-    'international','company','gesellschaft','mbh&co','mbhco','werke','werk','renkhoff','sonnenschutztechnik'
+    'international','company','gesellschaft','mbh&co','mbhco','werke','werk'
 }
-CITY_TOKENS = set()  # dynamisch befüllt nach Datennormalisierung
+CITY_TOKENS = set()  # dynamisch befüllt
 
 # --- Utilities ---
 def _tokenize(s: str):
-    if not s:
+    if not s: return []
-        return []
+    return re.split(r"[^a-z0-9äöüß]+", str(s).lower())
-    return re.split(r"[^a-z0-9]+", str(s).lower())
 
-def split_tokens(name: str):
+def clean_name_for_scoring(norm_name: str, dynamic_stopwords: set):
-    """Tokens für Indexing/Scoring (Basis-Stop + dynamische City-Tokens)."""
+    """Entfernt Stop- & City-Tokens sowie dynamische Stopwords."""
-    if not name:
+    if not norm_name: return "", set()
-        return []
+    
-    tokens = [t for t in _tokenize(name) if len(t) >= 3]
+    tokens = [t for t in _tokenize(norm_name) if len(t) >= 3]
-    stop_union = STOP_TOKENS_BASE | CITY_TOKENS
+    stop_union = STOP_TOKENS_BASE | CITY_TOKENS | dynamic_stopwords
-    return [t for t in tokens if t not in stop_union]
+    
+    final_tokens = [t for t in tokens if t not in stop_union]
+    return " ".join(final_tokens), set(final_tokens)
 
-def clean_name_for_scoring(norm_name: str):
+# --- NEU: TF-IDF Logik (vereinfacht) ---
-    """Entfernt Stop- & City-Tokens. Leerer Output => kein sinnvoller Namevergleich."""
+def build_term_weights(crm_df: pd.DataFrame, dynamic_stopwords: set):
-    toks = split_tokens(norm_name)
+    """Erstellt ein Gewichts-Wörterbuch basierend auf der Seltenheit der Wörter (IDF)."""
-    return " ".join(toks), set(toks)
+    logger.info("Starte Berechnung der Wortgewichte (TF-IDF)...")
+    token_counts = Counter()
+    total_docs = len(crm_df)
+    
+    for name in crm_df['normalized_name']:
+        _, tokens = clean_name_for_scoring(name, dynamic_stopwords)
+        for token in set(tokens): # Zähle jedes Wort nur einmal pro Firmenname
+            token_counts[token] += 1
+            
+    term_weights = {}
+    for token, count in token_counts.items():
+        # IDF-Formel: log(N / df) - je seltener das Wort, desto höher das Gewicht
+        idf = math.log(total_docs / (count + 1)) # +1 zur Glättung
+        term_weights[token] = idf
+        
+    logger.info(f"Wortgewichte für {len(term_weights)} Tokens berechnet.")
+    return term_weights
 
-def assess_serp_trust(company_name: str, url: str) -> str:
+def get_dynamic_stopwords(crm_df: pd.DataFrame, threshold_percent=0.01):
-    """Vertrauen 'hoch/mittel/niedrig' anhand Token-Vorkommen in Domain."""
+    """Identifiziert häufige Wörter im CRM-Datensatz, die als Stopwords behandelt werden sollen."""
-    if not url:
+    logger.info("Sammle dynamische Stopwords...")
-        return 'n/a'
+    token_counts = Counter()
-    host = simple_normalize_url(url) or ''
+    for name in crm_df['normalized_name']:
-    host = host.replace('www.', '')
+        tokens = [t for t in _tokenize(name) if len(t) >= 3 and t not in (STOP_TOKENS_BASE | CITY_TOKENS)]
-    name_toks = [t for t in split_tokens(normalize_company_name(company_name)) if len(t) >= 3]
+        for token in set(tokens):
-    if any(t in host for t in name_toks if len(t) >= 4):
+            token_counts[token] += 1
-        return 'hoch'
+            
-    if any(t in host for t in name_toks if len(t) == 3):
+    limit = len(crm_df) * threshold_percent
-        return 'mittel'
+    stopwords = {token for token, count in token_counts.items() if count > limit}
-    return 'niedrig'
+    logger.info(f"{len(stopwords)} dynamische Stopwords identifiziert (z.B. 'stadtwerke', 'werke', ...)")
+    return stopwords
 
 # --- Similarity ---
-def calculate_similarity(mrec: dict, crec: dict, token_freq: Counter):
+def calculate_similarity(mrec: dict, crec: dict, term_weights: dict, dynamic_stopwords: set):
+    
+    # --- NEU: Golden-Rule für exakten Namens-Match ---
+    n1_raw = mrec.get('normalized_name', '')
+    n2_raw = crec.get('normalized_name', '')
+    if fuzz.ratio(n1_raw, n2_raw) >= GOLDEN_MATCH_RATIO:
+        return GOLDEN_MATCH_SCORE, {'reason': f'Golden Match (Name Ratio >= {GOLDEN_MATCH_RATIO}%)'}
+
     # Domain (mit Gate)
     dom1 = mrec.get('normalized_domain','')
     dom2 = crec.get('normalized_domain','')
-    m_domain_use = mrec.get('domain_use_flag', 0)
+    domain_match = 1 if (dom1 and dom1 == dom2) else 0
-    domain_flag_raw = 1 if (m_domain_use == 1 and dom1 and dom1 == dom2) else 0
 
-    # Location flags
+    # Location
     city_match = 1 if (mrec.get('CRM Ort') and crec.get('CRM Ort') and mrec.get('CRM Ort') == crec.get('CRM Ort')) else 0
     country_match = 1 if (mrec.get('CRM Land') and crec.get('CRM Land') and mrec.get('CRM Land') == crec.get('CRM Land')) else 0
+    
+    # Name (bereinigt)
+    clean1, toks1 = clean_name_for_scoring(n1_raw, dynamic_stopwords)
+    clean2, toks2 = clean_name_for_scoring(n2_raw, dynamic_stopwords)
 
-    # Name (nur sinnvolle Tokens)
+    # --- NEU: Gewichteter Name-Score basierend auf TF-IDF ---
-    n1 = mrec.get('normalized_name','')
+    name_score = 0
-    n2 = crec.get('normalized_name','')
+    overlapping_tokens = toks1 & toks2
-    clean1, toks1 = clean_name_for_scoring(n1)
+    if overlapping_tokens:
-    clean2, toks2 = clean_name_for_scoring(n2)
+        # Score ist die Summe der Gewichte der übereinstimmenden Wörter
+        name_score = sum(term_weights.get(token, 0) for token in overlapping_tokens)
+        # Bonus für hohe prozentuale Übereinstimmung der seltenen Wörter
+        if toks1:
+            overlap_percentage = len(overlapping_tokens) / len(toks1)
+            name_score *= (1 + overlap_percentage)
 
-    # Overlaps
+    # --- NEU: Überarbeitete Gesamt-Score-Berechnung ---
-    overlap_clean = toks1 & toks2
+    # Basis-Score-Komponenten
-    # city-only overlap check (wenn nach Clean nichts übrig, aber Roh-Overlap evtl. Städte; wir cappen Score)
+    score_domain = 100 if domain_match else 0
-    raw_overlap = set(_tokenize(n1)) & set(_tokenize(n2))
+    score_location = 20 if (city_match and country_match) else 0
-    city_only_overlap = (not overlap_clean) and any(t in CITY_TOKENS for t in raw_overlap)
 
-    # Name-Score
+    # Gesamtscore
-    if clean1 and clean2:
+    total = name_score * 15 + score_domain + score_location # Name hat jetzt viel mehr Einfluss
-        ts = fuzz.token_set_ratio(clean1, clean2)
-        pr = fuzz.partial_ratio(clean1, clean2)
-        ss = fuzz.token_sort_ratio(clean1, clean2)
-        name_score = max(ts, pr, ss)
-    else:
-        name_score = 0
 
-    if city_only_overlap and name_score > 70:
+    # Strafen
-        name_score = 70  # cap
-
-    # Rare-token-overlap (IDF-light): benutze seltensten Token aus mrec
-    rtoks_sorted = sorted(list(toks1), key=lambda t: (token_freq.get(t, 10**9), -len(t)))
-    rare_token = rtoks_sorted[0] if rtoks_sorted else None
-    rare_overlap = 1 if (rare_token and rare_token in toks2) else 0
-
-    # Domain Gate
-    domain_gate_ok = (name_score >= MIN_NAME_FOR_DOMAIN) or (city_match and country_match)
-    domain_used = 1 if (domain_flag_raw and domain_gate_ok) else 0
-
-    # Basisscore
-    total = domain_used*100 + name_score*1.0 + (1 if (city_match and country_match) else 0)*20
-
-    # Penalties
     penalties = 0
     if mrec.get('CRM Land') and crec.get('CRM Land') and not country_match:
-        penalties += COUNTRY_MISMATCH_PENALTY
+        penalties += 40
     if mrec.get('CRM Ort') and crec.get('CRM Ort') and not city_match:
-        penalties += CITY_MISMATCH_PENALTY
+        penalties += 30
     total -= penalties
-
+    
-    # Bonus für starke Name-only Fälle
-    name_bonus = 1 if (domain_used == 0 and not (city_match and country_match) and name_score >= 85 and rare_overlap==1) else 0
-    if name_bonus:
-        total += 20
-
     comp = {
-        'domain_raw': domain_flag_raw,
+        'name_score': round(name_score,1),
-        'domain_used': domain_used,
+        'domain_match': domain_match,
-        'domain_gate_ok': int(domain_gate_ok),
-        'name': round(name_score,1),
         'city_match': city_match,
         'country_match': country_match,
         'penalties': penalties,
-        'name_bonus': name_bonus,
+        'overlapping_tokens': list(overlapping_tokens)
-        'rare_overlap': rare_overlap,
-        'city_only_overlap': int(city_only_overlap)
     }
-    return round(total), comp
+    
+    return max(0, round(total)), comp
 
 # --- Indexe ---
-def build_indexes(crm_df: pd.DataFrame):
+def build_indexes(crm_df: pd.DataFrame, dynamic_stopwords: set):
     records = list(crm_df.to_dict('records'))
     # Domain-Index
     domain_index = {}
@@ -209,178 +222,126 @@ def build_indexes(crm_df: pd.DataFrame):
         d = r.get('normalized_domain')
         if d:
             domain_index.setdefault(d, []).append(r)
-    # Token-Frequenzen (auf gereinigten Tokens)
+    
-    token_freq = Counter()
-    for r in records:
-        _, toks = clean_name_for_scoring(r.get('normalized_name',''))
-        for t in set(toks):
-            token_freq[t] += 1
     # Token-Index
     token_index = {}
-    for r in records:
+    for idx, r in enumerate(records):
-        _, toks = clean_name_for_scoring(r.get('normalized_name',''))
+        _, toks = clean_name_for_scoring(r.get('normalized_name',''), dynamic_stopwords)
         for t in set(toks):
-            token_index.setdefault(t, []).append(r)
+            token_index.setdefault(t, []).append(idx) # Speichere Index statt ganzem Record
-    return records, domain_index, token_freq, token_index
+            
+    return records, domain_index, token_index
 
-
+def choose_rarest_token(norm_name: str, term_weights: dict, dynamic_stopwords: set):
-def choose_rarest_token(norm_name: str, token_freq: Counter):
+    _, toks = clean_name_for_scoring(norm_name, dynamic_stopwords)
-    _, toks = clean_name_for_scoring(norm_name)
+    if not toks: return None
-    if not toks:
+    # Seltenstes Token hat höchstes Gewicht (höchsten IDF-Score)
-        return None
+    rarest = max(toks, key=lambda t: term_weights.get(t, 0))
-    lst = sorted(list(toks), key=lambda x: (token_freq.get(x, 10**9), -len(x)))
+    return rarest if term_weights.get(rarest, 0) > 0 else None
-    return lst[0] if lst else None
 
 # --- Hauptfunktion ---
-def main(job_id=None):
+def main(job_id=None, interactive=False):
-    logger.info("Starte Duplikats-Check v2.15 (Quality-first++)")
+    logger.info("Starte Duplikats-Check v3.0 (Weighted Scoring & Interactive Mode)")
-    # NEU: Status-Update
     update_status(job_id, "Läuft", "Initialisiere GoogleSheetHandler...")
     try:
         sheet = GoogleSheetHandler()
         logger.info("GoogleSheetHandler initialisiert")
     except Exception as e:
         logger.critical(f"Init GoogleSheetHandler fehlgeschlagen: {e}")
-        # NEU: Status-Update bei Fehler
         update_status(job_id, "Fehlgeschlagen", f"Init GoogleSheetHandler fehlgeschlagen: {e}")
         sys.exit(1)
 
     # Daten laden
-    # NEU: Status-Update
     update_status(job_id, "Läuft", "Lade CRM- und Matching-Daten...")
     crm_df = sheet.get_sheet_as_dataframe(CRM_SHEET_NAME)
     match_df = sheet.get_sheet_as_dataframe(MATCHING_SHEET_NAME)
-    total = len(match_df) if match_df is not None else 0 # NEU: total hier definieren
+    total = len(match_df) if match_df is not None else 0
     logger.info(f"{0 if crm_df is None else len(crm_df)} CRM-Datensätze | {total} Matching-Datensätze")
     if crm_df is None or crm_df.empty or match_df is None or match_df.empty:
         logger.critical("Leere Daten in einem der Sheets. Abbruch.")
-        # NEU: Status-Update bei Fehler
         update_status(job_id, "Fehlgeschlagen", "Leere Daten in einem der Sheets.")
         return
 
-    # SerpAPI nur für Matching (B und E leer)
+    # Normalisierung
-    if Config.API_KEYS.get('serpapi'): # Sicherer Zugriff auf den Key
-        if 'Gefundene Website' not in match_df.columns:
-            match_df['Gefundene Website'] = ''
-        b_empty = match_df['CRM Website'].fillna('').astype(str).str.strip().str.lower().isin(['','k.a.','k.a','n/a','na'])
-        e_empty = match_df['Gefundene Website'].fillna('').astype(str).str.strip().str.lower().isin(['','k.a.','k.a','n/a','na'])
-        empty_mask = b_empty & e_empty
-        empty_count = int(empty_mask.sum())
-        if empty_count > 0:
-            # NEU: Status-Update
-            update_status(job_id, "Läuft", f"Suche Websites für {empty_count} Firmen via SerpAPI...")
-            logger.info(f"Serp-Fallback für Matching: {empty_count} Firmen ohne URL in B/E")
-            found_cnt = 0
-            trust_stats = Counter()
-            for idx, row in match_df[empty_mask].iterrows():
-                company = row['CRM Name']
-                try:
-                    url = serp_website_lookup(company)
-                    if url and 'k.A.' not in url:
-                        if not str(url).startswith(('http://','https://')):
-                            url = 'https://' + str(url).lstrip()
-                        trust = assess_serp_trust(company, url)
-                        match_df.at[idx, 'Gefundene Website'] = url
-                        match_df.at[idx, 'Serp Vertrauen'] = trust
-                        trust_stats[trust] += 1
-                        logger.info(f"  ✓ URL gefunden: '{company}' -> {url} (Vertrauen: {trust})")
-                        found_cnt += 1
-                    else:
-                        logger.debug(f"  ✗ Keine eindeutige URL: '{company}' -> {url}")
-                except Exception as e:
-                    logger.warning(f"  ! Serp-Fehler für '{company}': {e}")
-            logger.info(f"Serp-Fallback beendet: {found_cnt}/{empty_count} URLs ergänzt | Trust: {dict(trust_stats)}")
-        else:
-            logger.info("Serp-Fallback übersprungen: B oder E bereits befüllt (keine fehlenden Matching-URLs)")
-
-    # Normalisierung CRM
     update_status(job_id, "Läuft", "Normalisiere Daten...")
     crm_df['normalized_name']   = crm_df['CRM Name'].astype(str).apply(normalize_company_name)
     crm_df['normalized_domain'] = crm_df['CRM Website'].astype(str).apply(simple_normalize_url)
     crm_df['CRM Ort']           = crm_df['CRM Ort'].astype(str).str.lower().str.strip()
     crm_df['CRM Land']          = crm_df['CRM Land'].astype(str).str.lower().str.strip()
-    crm_df['block_key']         = crm_df['normalized_name'].apply(lambda x: x.split()[0] if x else None)
+    
-    crm_df['domain_use_flag']   = 1
-
-    # Normalisierung Matching
-    match_df['Gefundene Website'] = match_df.get('Gefundene Website', pd.Series(index=match_df.index, dtype=object))
-    match_df['Serp Vertrauen']    = match_df.get('Serp Vertrauen', pd.Series(index=match_df.index, dtype=object))
-    match_df['Effektive Website'] = match_df['CRM Website'].fillna('').astype(str).str.strip()
-    mask_eff = match_df['Effektive Website'] == ''
-    match_df.loc[mask_eff, 'Effektive Website'] = match_df['Gefundene Website'].fillna('').astype(str).str.strip()
-
     match_df['normalized_name']   = match_df['CRM Name'].astype(str).apply(normalize_company_name)
-    match_df['normalized_domain'] = match_df['Effektive Website'].astype(str).apply(simple_normalize_url)
+    match_df['normalized_domain'] = match_df['CRM Website'].astype(str).apply(simple_normalize_url)
     match_df['CRM Ort']           = match_df['CRM Ort'].astype(str).str.lower().str.strip()
     match_df['CRM Land']          = match_df['CRM Land'].astype(str).str.lower().str.strip()
-    match_df['block_key']         = match_df['normalized_name'].apply(lambda x: x.split()[0] if x else None)
+    
-
-    def _domain_use(row):
-        if str(row.get('CRM Website','')).strip():
-            return 1
-        trust = str(row.get('Serp Vertrauen','')).lower()
-        return 1 if trust == 'hoch' else 0
-    match_df['domain_use_flag'] = match_df.apply(_domain_use, axis=1)
-
     def build_city_tokens(crm_df, match_df):
         cities = set()
         for s in pd.concat([crm_df['CRM Ort'], match_df['CRM Ort']], ignore_index=True).dropna().unique():
             for t in _tokenize(s):
-                if len(t) >= 3:
+                if len(t) >= 3: cities.add(t)
-                    cities.add(t)
         return cities
     global CITY_TOKENS
     CITY_TOKENS = build_city_tokens(crm_df, match_df)
     logger.info(f"City tokens gesammelt: {len(CITY_TOKENS)}")
 
-    # Blocking-Indizes
+    # --- NEU: TF-IDF und Index-Erstellung ---
-    update_status(job_id, "Läuft", "Erstelle Blocking-Indizes...")
+    dynamic_stopwords = get_dynamic_stopwords(crm_df)
-    crm_records, domain_index, token_freq, token_index = build_indexes(crm_df)
+    term_weights = build_term_weights(crm_df, dynamic_stopwords)
+    crm_records, domain_index, token_index = build_indexes(crm_df, dynamic_stopwords)
     logger.info(f"Blocking: Domains={len(domain_index)} | TokenKeys={len(token_index)}")
 
-    # Matching
+    # --- Matching ---
     results = []
-    metrics = Counter()
     logger.info("Starte Matching-Prozess…")
     
     for idx, mrow in match_df.to_dict('index').items():
         processed = idx + 1
         progress_message = f"Prüfe {processed}/{total}: '{mrow.get('CRM Name','')}'"
         logger.info(progress_message)
-        # NEU: Status-Update in der Schleife
         if processed % 5 == 0 or processed == total:
             update_status(job_id, "Läuft", progress_message)
 
-        candidates = []
+        candidate_indices = set()
         used_block = ''
-        if mrow.get('normalized_domain') and mrow.get('domain_use_flag') == 1:
+        
-            candidates = domain_index.get(mrow['normalized_domain'], [])
+        # Blocking via Domain
-            used_block = f"domain:{mrow['normalized_domain']}"
+        if mrow.get('normalized_domain'):
-        if not candidates:
+            candidates_from_domain = domain_index.get(mrow['normalized_domain'], [])
-            rtok = choose_rarest_token(mrow.get('normalized_name',''), token_freq)
+            for c in candidates_from_domain:
+                # Finde den Index des Records, um Duplikate zu vermeiden
+                # Dies ist ineffizient, für eine genaue Index-Logik müsste der domain_index auch indices speichern
+                for i, record in enumerate(crm_records):
+                    if record['CRM Name'] == c['CRM Name'] and record['CRM Website'] == c['CRM Website']:
+                         candidate_indices.add(i)
+                         break
+            if candidate_indices: used_block = f"domain:{mrow['normalized_domain']}"
+        
+        # Blocking via seltenstes Token
+        if not candidate_indices:
+            rtok = choose_rarest_token(mrow.get('normalized_name',''), term_weights, dynamic_stopwords)
             if rtok:
-                candidates = token_index.get(rtok, [])
+                indices_from_token = token_index.get(rtok, [])
+                candidate_indices.update(indices_from_token)
                 used_block = f"token:{rtok}"
-        if not candidates:
+
+        # Prefilter als Fallback
+        if not candidate_indices:
             pf = []
             n1 = mrow.get('normalized_name','')
-            rtok = choose_rarest_token(n1, token_freq)
+            clean1, _ = clean_name_for_scoring(n1, dynamic_stopwords)
-            clean1, toks1 = clean_name_for_scoring(n1)
             if clean1:
-                for r in crm_records:
+                for i, r in enumerate(crm_records):
                     n2 = r.get('normalized_name','')
-                    clean2, toks2 = clean_name_for_scoring(n2)
+                    clean2, _ = clean_name_for_scoring(n2, dynamic_stopwords)
-                    if not clean2:
+                    if not clean2: continue
-                        continue
-                    if rtok and rtok not in toks2:
-                        continue
                     pr = fuzz.partial_ratio(clean1, clean2)
                     if pr >= PREFILTER_MIN_PARTIAL:
-                        pf.append((pr, r))
+                        pf.append((pr, i))
             pf.sort(key=lambda x: x[0], reverse=True)
-            candidates = [r for _, r in pf[:PREFILTER_LIMIT]]
+            candidate_indices.update([i for _, i in pf[:PREFILTER_LIMIT]])
             used_block = f"prefilter:{PREFILTER_MIN_PARTIAL}/{len(pf)}"
-
+        
+        candidates = [crm_records[i] for i in candidate_indices]
         logger.info(f"Prüfe {processed}/{total}: '{mrow.get('CRM Name','')}' -> {len(candidates)} Kandidaten (Block={used_block})")
         if not candidates:
             results.append({'Match':'', 'Score':0, 'Match_Grund':'keine Kandidaten'})
@@ -388,31 +349,65 @@ def main(job_id=None):
 
         scored = []
         for cr in candidates:
-            score, comp = calculate_similarity(mrow, cr, token_freq)
+            score, comp = calculate_similarity(mrow, cr, term_weights, dynamic_stopwords)
-            scored.append((cr.get('CRM Name',''), score, comp))
+            scored.append({'name': cr.get('CRM Name',''), 'score': score, 'comp': comp, 'record': cr})
-        scored.sort(key=lambda x: x[1], reverse=True)
+        scored.sort(key=lambda x: x['score'], reverse=True)
+        
+        for cand in scored[:5]:
+            logger.debug(f"  Kandidat: {cand['name']} | Score={cand['score']} | Comp={cand['comp']}")
 
-        for cand_name, sc, comp in scored[:5]:
+        best_match = scored[0] if scored else None
-            logger.debug(f"  Kandidat: {cand_name} | Score={sc} | Comp={comp}")
+        
+        # --- NEU: Interaktiver Modus ---
+        if interactive and best_match and len(scored) > 1:
+            best_score = best_match['score']
+            second_best_score = scored[1]['score']
+            if best_score > INTERACTIVE_SCORE_MIN and (best_score - second_best_score) < INTERACTIVE_SCORE_DIFF:
+                print("\n" + "="*50)
+                print(f"AMBIGUOUS MATCH for '{mrow['CRM Name']}'")
+                print(f"Top candidates have very similar scores.")
+                print(f"  - Match:    '{mrow['CRM Name']}' | {mrow['normalized_domain']} | {mrow['CRM Ort']}, {mrow['CRM Land']}")
+                print("-"*50)
+                for i, item in enumerate(scored[:5]):
+                    cr = item['record']
+                    print(f"[{i+1}] Candidate: '{cr['CRM Name']}' | {cr['normalized_domain']} | {cr['CRM Ort']}, {cr['CRM Land']}")
+                    print(f"    Score: {item['score']} | Details: {item['comp']}")
+                print("[0] No match")
+                
+                choice = -1
+                while choice < 0 or choice > len(scored[:5]):
+                    try:
+                        choice = int(input(f"Please select the best match (1-{len(scored[:5])}) or 0 for no match: "))
+                    except ValueError:
+                        choice = -1
+                
+                if choice > 0:
+                    best_match = scored[choice-1]
+                    logger.info(f"User selected candidate {choice}: '{best_match['name']}'")
+                elif choice == 0:
+                    best_match = None # User decided no match
+                    logger.info("User selected no match.")
+                print("="*50 + "\n")
 
-        best_name, best_score, best_comp = scored[0]
 
-        weak = (best_comp.get('domain_used') == 0 and not (best_comp.get('city_match') and best_comp.get('country_match')))
+        if best_match and best_match['score'] >= SCORE_THRESHOLD:
-        applied_threshold = SCORE_THRESHOLD_WEAK if weak else SCORE_THRESHOLD
+            # Schwache Matches (ohne Domain/Ort) brauchen höheren Threshold
-        weak_guard_fail = (weak and best_comp.get('rare_overlap') == 0)
+            is_weak = best_match['comp'].get('domain_match', 0) == 0 and not (best_match['comp'].get('city_match', 0) and best_match['comp'].get('country_match', 0))
+            applied_threshold = SCORE_THRESHOLD_WEAK if is_weak else SCORE_THRESHOLD
 
-        if not weak_guard_fail and best_score >= applied_threshold:
+            if best_match['score'] >= applied_threshold:
-            results.append({'Match': best_name, 'Score': best_score, 'Match_Grund': str(best_comp)})
+                results.append({'Match': best_match['name'], 'Score': best_match['score'], 'Match_Grund': str(best_match['comp'])})
-            metrics['matches_total'] += 1
+                logger.info(f" --> Match: '{best_match['name']}' ({best_match['score']}) | TH={applied_threshold}{' (weak)' if is_weak else ''}")
-            if best_comp.get('domain_used') == 1: metrics['matches_domain'] += 1
+            else:
-            if best_comp.get('city_match') and best_comp.get('country_match'): metrics['matches_with_loc'] += 1
+                results.append({'Match':'', 'Score': best_match['score'], 'Match_Grund': f"Below WEAK threshold | {str(best_match['comp'])}"})
-            if best_comp.get('domain_used') == 0 and best_comp.get('name') >= 85 and not (best_comp.get('city_match') and best_comp.get('country_match')):
+                logger.info(f" --> No Match (below weak TH): '{best_match['name']}' ({best_match['score']}) | TH={applied_threshold}")
-                metrics['matches_name_only'] += 1
+        elif best_match:
-            logger.info(f" --> Match: '{best_name}' ({best_score}) {best_comp} | TH={applied_threshold}{' weak' if weak else ''}")
+             results.append({'Match':'', 'Score': best_match['score'], 'Match_Grund': f"Below threshold | {str(best_match['comp'])}"})
+             logger.info(f" --> No Match (below TH): '{best_match['name']}' ({best_match['score']})")
         else:
-            reason = 'weak_guard_no_rare' if weak_guard_fail else 'below_threshold'
+             results.append({'Match':'', 'Score':0, 'Match_Grund':'No valid candidates or user override'})
-            results.append({'Match':'', 'Score': best_score, 'Match_Grund': f"{best_comp} | {reason} TH={applied_threshold}"})
+             logger.info(f" --> No Match (no candidates)")
-            logger.info(f" --> Kein Match (Score={best_score}) {best_comp} | {reason} TH={applied_threshold}")
+
 
     # --- Ergebnisse zurückschreiben ---
     logger.info("Matching-Prozess abgeschlossen. Bereite Ergebnisse für den Upload vor...")
@@ -420,27 +415,20 @@ def main(job_id=None):
 
     result_df = pd.DataFrame(results)
     
-    # KORREKTUR: Füge die Ergebnisspalten zum MODIFIZIERTEN match_df hinzu,
+    # Löschen der Ergebnisspalten, falls sie bereits im Sheet existieren
-    # der die neuen URLs aus der SerpAPI-Suche enthält.
+    cols_to_drop_from_match = ['Match', 'Score', 'Match_Grund']
-    # Wir benutzen die Original-Indizes, um sicherzustellen, dass alles passt.
+    match_df_clean = match_df.drop(columns=[col for col in cols_to_drop_from_match if col in match_df.columns], errors='ignore')
-    final_df = match_df.join(result_df)
+    
+    final_df = pd.concat([match_df_clean, result_df], axis=1)
 
-    # Bereinige die temporären Spalten für eine saubere Ausgabe
+    cols_to_drop = ['normalized_name', 'normalized_domain']
-    # KORREKTUR: 'block_key' statt 'block_keys'
-    cols_to_drop = ['normalized_name', 'normalized_domain', 'block_key', 'Effektive Website', 'domain_use_flag']
     final_df = final_df.drop(columns=[col for col in cols_to_drop if col in final_df.columns], errors='ignore')
 
-    # NEU: Robuster Schreibprozess zur Vermeidung von Typ-Fehlern
-    # 1. Alle Spalten explizit in String konvertieren, um Inkompatibilitäten mit der API (z.B. numpy-Typen) zu vermeiden.
-    # 2. NaN/None-Werte mit einem leeren String füllen.
     upload_df = final_df.astype(str).replace({'nan': '', 'None': ''})
-    
-    # Konvertiere in Liste von Listen für den Upload
     data_to_write = [upload_df.columns.tolist()] + upload_df.values.tolist()
     
     logger.info(f"Versuche, {len(data_to_write) - 1} Ergebniszeilen in das Sheet '{MATCHING_SHEET_NAME}' zu schreiben...")
     
-    # KORREKTUR: 'sheet' statt 'sheet_handler' verwenden
     ok = sheet.clear_and_write_data(MATCHING_SHEET_NAME, data_to_write)
     
     if ok:
@@ -450,20 +438,12 @@ def main(job_id=None):
         logger.error("Fehler beim Schreiben der Ergebnisse ins Google Sheet.")
         update_status(job_id, "Fehlgeschlagen", "Fehler beim Schreiben ins Google Sheet.")
 
-    # Summary
-    # KORREKTUR: 'final_df' statt 'write_df' verwenden
-    serp_counts = Counter((str(x).lower() for x in final_df.get('Serp Vertrauen', [])))
-    logger.info("===== Summary =====")
-    logger.info(f"Matches total: {metrics['matches_total']} | mit Domain: {metrics['matches_domain']} | mit Ort: {metrics['matches_with_loc']} | nur Name: {metrics['matches_name_only']}")
-    logger.info(f"Serp Vertrauen: {dict(serp_counts)}")
-    logger.info(f"Config: TH={SCORE_THRESHOLD}, TH_WEAK={SCORE_THRESHOLD_WEAK}, MIN_NAME_FOR_DOMAIN={MIN_NAME_FOR_DOMAIN}, Penalties(city={CITY_MISMATCH_PENALTY},country={COUNTRY_MISMATCH_PENALTY}), Prefilter(partial>={PREFILTER_MIN_PARTIAL}, limit={PREFILTER_LIMIT})")
-
 if __name__=='__main__':
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(description="Duplicate Checker v3.0")
     parser.add_argument("--job-id", type=str, help="Eindeutige ID für den Job-Status.")
+    parser.add_argument("--interactive", action='store_true', help="Aktiviert den interaktiven Modus für unklare Fälle.")
     args = parser.parse_args()
     
-    # Lade API-Keys, bevor die main-Funktion startet
     Config.load_api_keys()
     
-    main(job_id=args.job_id)
+    main(job_id=args.job_id, interactive=args.interactive)