app.py

# ============================================================================
# GERMAN LINGUISTICS HUB (CONSOLIDATED APP V23)
#
# This script combines multiple NLP tools into a single Gradio interface.
#
# ============================================================================
# TABS & FUNCTIONALITY:
# ============================================================================
#
# --- PRIMARY TABS ---
#
# 1. Word Encyclopedia (DE):
#    - NON-CONTEXTUAL analysis of single words.
#    - Multi-engine dispatcher with user selection and automatic fallback:
#      (Wiktionary -> DWDSmor -> HanTa -> IWNLP)
#    - Aggregates all grammatical (Wiktionary, Pattern) and semantic 
#      (Wiktionary, OdeNet, ConceptNet) possibilities, grouped by Part-of-Speech.
#    - Validates and filters artifacts (e.g., "abgeschnitten", "lauf").
#
# 2. Comprehensive Analyzer (DE):
#    - CONTEXTUAL analysis of full sentences.
#    - Uses the Word Encyclopedia's dispatcher for robust lemma analysis.
#    - Ranks all semantic senses (Wiktionary, OdeNet) by relevance to the sentence.
#
# --- STANDALONE TOOL TABS ---
#
# 3. spaCy Analyzer (Multi-lingual): 
#    - Direct, raw spaCy output (NER, POS, dependencies) for multiple languages.
#
# 4. Grammar Check (DE): 
#    - Direct LanguageTool output.
#
# --- RAW ENGINE TABS (for debugging & comparison) ---
#
# 5. Engine: Wiktionary (DE): 
#    - Standalone access to the Wiktionary DB (Primary) engine.
#
# 6. Engine: DWDSmor (DE):
#    - Standalone access to the DWDSmor (Fallback 1) engine.
#
# 7. Engine: HanTa (DE):
#    - Standalone access to the HanTa (Fallback 2) engine.
#
# 8. Engine: IWNLP-spaCy (DE):
#    - Standalone access to the IWNLP-spaCy (Fallback 3) engine.
#
# --- RAW COMPONENT TABS (for debugging & comparison) ---
#
# 9. Component: Inflections (DE): 
#    - Direct access to the `pattern.de` library.
#
# 10. Component: Thesaurus (DE): 
#     - Direct access to the `OdeNet` library.
#
# 11. Component: ConceptNet (Direct): 
#     - Direct access to the ConceptNet API.
#
# ============================================================================

# ============================================================================
# 1. CONSOLIDATED IMPORTS
# ============================================================================
import gradio as gr
import spacy
from spacy import displacy
import base64
import traceback
import subprocess
import sys
import os
from pathlib import Path
import importlib
import site
import threading
import queue
from dataclasses import dataclass
from enum import Enum
from typing import Dict, Any, List, Set, Optional, Tuple
import requests  
import zipfile  
import re
import sqlite3 
import json
from huggingface_hub import hf_hub_download

# --- Requests and gradio Import (for ConceptNet) ---
try:
    import requests
    from requests.exceptions import RequestException, HTTPError, ConnectionError, Timeout
    REQUESTS_AVAILABLE = True
except ImportError:
    REQUESTS_AVAILABLE = False
    print("="*70)
    print("CRITICAL WARNING: `requests` library not found.")
    print("ConceptNet features will not function.")
    print("="*70)


try:
    from gradio_client import Client
    GRADIO_CLIENT_AVAILABLE = True

except ImportError:
    GRADIO_CLIENT_AVAILABLE = False
    print("="*70)
    print("CRITICAL WARNING: `gradio_client` library not found.")
    print("ConceptNet features will not function.")
    print("Install with: pip install gradio_client")
    print("="*70)

# --- IWNLP (spaCy Extension) Import ---
try:
    from spacy_iwnlp import spaCyIWNLP
    IWNLP_AVAILABLE = True
    print("✓ Successfully imported spacy-iwnlp")
except ImportError:
    IWNLP_AVAILABLE = False
    spaCyIWNLP = object # Dummy definition for error case
    print("="*70)
    print("WARNING: `spacy-iwnlp` library not found.")
    print("The 'Word Encyclopedia' tab will be less accurate.")
    print("Install with: pip install spacy-iwnlp")
    print("="*70)

# --- LanguageTool Import ---
try:
    import language_tool_python
    LT_AVAILABLE = True
    print("✓ Successfully imported language_tool")
except ImportError:
    LT_AVAILABLE = False
    print("="*70)
    print("CRITICAL WARNING: `language-tool-python` library not found.")
    print("The 'German Grammar Check' tab will not function.")
    print("="*70)

# --- OdeNet (wn) Import ---
try:
    import wn
    WN_AVAILABLE = True
    print("✓ Successfully imported wordnet for odenet")
except ImportError:
    WN_AVAILABLE = False
    print("="*70)
    print("CRITICAL WARNING: `wn` library not found.")
    print("The 'German Thesaurus' tab will not function.")
    print("="*70)

# --- Pattern.de Import ---
try:
    from pattern.de import (
        pluralize, singularize, conjugate, tenses, lemma, lexeme,
        attributive, predicative,
        article, gender, MALE, FEMALE, NEUTRAL, PLURAL,
        INFINITIVE, PRESENT, PAST, PARTICIPLE,
        FIRST, SECOND, THIRD, SINGULAR, PLURAL as PL,
        INDICATIVE, IMPERATIVE, SUBJUNCTIVE,
        NOMINATIVE, ACCUSATIVE, DATIVE, GENITIVE,
        SUBJECT, OBJECT, INDIRECT, PROPERTY,
        DEFINITE, INDEFINITE,
        comparative, superlative,
        NOUN, VERB, ADJECTIVE,
        parse, split
    )
    PATTERN_DE_AVAILABLE = True
    print("✓ Successfully imported pattern.de")
except ImportError as e:
    PATTERN_DE_AVAILABLE = False
    # --- FIX: Define dummy constants to prevent NameError in other functions ---
    MALE, FEMALE, NEUTRAL, PLURAL, SINGULAR = 1, 2, 3, "pl", "sg"
    NOMINATIVE, ACCUSATIVE, DATIVE, GENITIVE = "nom", "acc", "dat", "gen"
    INFINITIVE, PRESENT, PAST, PARTICIPLE = "inf", "pres", "pst", "part"
    DEFINITE, INDEFINITE = "def", "indef"
    
    print("="*70)
    print(f"WARNING: `pattern.de` library not found: {e}")

# --- HanTa Tagger Import ---
try:
    from HanTa.HanoverTagger import HanoverTagger
    import HanTa.HanoverTagger
    # This sys.modules line is critical for pickle compatibility
    sys.modules['HanoverTagger'] = HanTa.HanoverTagger
    HANTA_AVAILABLE = True
    print("✓ Successfully imported HanTa")
except ImportError:
    HANTA_AVAILABLE = False
    HanoverTagger = object # Dummy definition
    print("="*70)
    print("CRITICAL WARNING: `HanTa` library not found.")
    print("The 'Word Encyclopedia' tab will NOT function.")
    print("Install with: pip install HanTa")
    print("="*70)

# --- DWDSmor Import ---
DWDSMOR_AVAILABLE = False
DwdsmorLemmatizerClass = object # Dummy definition
try:
    import dwdsmor
    import dwdsmor.spacy # Test this import
    DWDSMOR_AVAILABLE = True
    print("✓ Successfully imported dwdsmor")
except ImportError as e:
    DWDSMOR_AVAILABLE = False
    print("="*70)
    print(f"WARNING: `dwdsmor` or a dependency failed to import: {e}")
    print("The DWDSmor engine will not be available.")
    print("On macOS, run: brew install sfst")
    print("On Debian/Ubuntu, run: apt-get install sfst")
    print("Then, run: pip install dwdsmor")
    print("="*70)

# ============================================================================
# 2. SHARED GLOBALS & CONFIG
# ============================================================================
VERBOSE = True  # Enable verbose debug output for Pattern.de
def log(msg):
    """Print debug messages if verbose mode is on."""
    if VERBOSE:
        print(f"[DEBUG] {msg}")

# --- Wiktionary Cache & Lock ---
WIKTIONARY_DB_PATH = "de_wiktionary_normalized_full.db"
WIKTIONARY_REPO_ID = "cstr/de-wiktionary-sqlite-full"
WIKTIONARY_CONN: Optional[sqlite3.Connection] = None
WIKTIONARY_CONN_LOCK = threading.Lock()
WIKTIONARY_AVAILABLE = False

# --- ConceptNet Cache & Lock ---
CONCEPTNET_CACHE: Dict[Tuple[str, str], Any] = {}
CONCEPTNET_LOCK = threading.Lock()

CONCEPTNET_CLIENT: Optional[Client] = None 
CONCEPTNET_CLIENT_LOCK = threading.Lock() 

# --- HanTa Tagger Cache & Lock ---
HANTA_TAGGER_INSTANCE: Optional[HanoverTagger] = None
HANTA_TAGGER_LOCK = threading.Lock()

# --- DWDSmor Cache & Lock ---
DWDSMOR_LEMMATIZER: Optional[Any] = None
DWDSMOR_LEMMATIZER_LOCK = threading.Lock()

# --- Helper ---
def _html_wrap(content: str, line_height: str = "2.0") -> str:
    """Wraps displaCy HTML in a consistent, scrollable div."""
    return f'<div style="overflow-x:auto; border: 1px solid #e6e9ef; border-radius: 0.25rem; padding: 1rem; line-height: {line_height};">{content}</div>'

# --- Helper for SVA ---
def _conjugate_to_person_number(verb_lemma: str, person: str, number: str) -> Optional[str]:
    """
    Return a present tense finite form for given person/number.
    person in {'1','2','3'}, number in {'sg','pl'}.
    """
    if not PATTERN_DE_AVAILABLE:
        return None
    try:
        alias = {"1sg":"1sg","2sg":"2sg","3sg":"3sg","1pl":"1pl","2pl":"2pl","3pl":"3pl"}[f"{person}{number}"]
        return conjugate(verb_lemma, alias)
    except Exception:
        return None

# ============================================================================
# 3. SPACY ANALYZER LOGIC
# ============================================================================
# --- Globals & Config for spaCy ---
SPACY_MODEL_INFO: Dict[str, Tuple[str, str, str]] = {
    "de": ("German", "de_core_news_md", "spacy"),
    "en": ("English", "en_core_web_md", "spacy"),
    "es": ("Spanish", "es_core_news_md", "spacy"),
    "grc-proiel-trf": ("Ancient Greek (PROIEL TRF)", "grc_proiel_trf", "grecy"),
    "grc-perseus-trf": ("Ancient Greek (Perseus TRF)", "grc_perseus_trf", "grecy"),
    "grc_ner_trf": ("Ancient Greek (NER TRF)", "grc_ner_trf", "grecy"),
    "grc-proiel-lg": ("Ancient Greek (PROIEL LG)", "grc_proiel_lg", "grecy"),
    "grc-perseus-lg": ("Ancient Greek (Perseus LG)", "grc_perseus_lg", "grecy"),
    "grc-proiel-sm": ("Ancient Greek (PROIEL SM)", "grc_proiel_sm", "grecy"),
    "grc-perseus-sm": ("Ancient Greek (Perseus SM)", "grc_perseus_sm", "grecy"),
}
SPACY_UI_TEXT = {
    "de": {
        "title": "# 🔍 Mehrsprachiger Morpho-Syntaktischer Analysator",
        "subtitle": "Analysieren Sie Texte auf Deutsch, Englisch, Spanisch und Altgriechisch",
        "ui_lang_label": "Benutzeroberflächensprache",
        "model_lang_label": "Textsprache für Analyse",
        "input_label": "Text eingeben",
        "input_placeholder": "Geben Sie hier Ihren Text ein...",
        "button_text": "Text analysieren",
        "button_processing_text": "Verarbeitung läuft...",
        "tab_graphic": "Grafische Darstellung",
        "tab_table": "Tabelle",
        "tab_json": "JSON",
        "tab_ner": "Entitäten",
        "html_label": "Abhängigkeitsparsing",
        "table_label": "Morphologische Analyse",
        "table_headers": ["Wort", "Lemma", "POS", "Tag", "Morphologie", "Abhängigkeit"],
        "json_label": "JSON-Ausgabe",
        "ner_label": "Benannte Entitäten",
        "error_message": "Fehler: "
    },
    "en": {
        "title": "# 🔍 Multilingual Morpho-Syntactic Analyzer",
        "subtitle": "Analyze texts in German, English, Spanish, and Ancient Greek",
        "ui_lang_label": "Interface Language",
        "model_lang_label": "Text Language for Analysis",
        "input_label": "Enter Text",
        "input_placeholder": "Enter your text here...",
        "button_text": "Analyze Text",
        "button_processing_text": "Processing...",
        "tab_graphic": "Graphic View",
        "tab_table": "Table",
        "tab_json": "JSON",
        "tab_ner": "Entities",
        "html_label": "Dependency Parsing",
        "table_label": "Morphological Analysis",
        "table_headers": ["Word", "Lemma", "POS", "Tag", "Morphology", "Dependency"],
        "json_label": "JSON Output",
        "ner_label": "Named Entities",
        "error_message": "Error: "
    },
    "es": {
        "title": "# 🔍 Analizador Morfo-Sintáctico Multilingüe",
        "subtitle": "Analice textos en alemán, inglés, español y griego antiguo",
        "ui_lang_label": "Idioma de la Interfaz",
        "model_lang_label": "Idioma del Texto para Análisis",
        "input_label": "Introducir Texto",
        "input_placeholder": "Ingrese su texto aquí...",
        "button_text": "Analizar Texto",
        "button_processing_text": "Procesando...",
        "tab_graphic": "Vista Gráfica",
        "tab_table": "Tabla",
        "tab_json": "JSON",
        "tab_ner": "Entidades",
        "html_label": "Análisis de Dependencias",
        "table_label": "Análisis Morfológico",
        "table_headers": ["Palabra", "Lema", "POS", "Etiqueta", "Morfología", "Dependencia"],
        "json_label": "Salida JSON",
        "ner_label": "Entidades Nombradas",
        "error_message": "Error: "
    }
}
SPACY_MODELS: Dict[str, Optional[spacy.Language]] = {}

# --- Dependency Installation ---
def spacy_install_spacy_transformers_once():
    """ Installs spacy-transformers, required for all _trf models. """
    marker_file = Path(".spacy_transformers_installed")
    if marker_file.exists():
        print("✓ spacy-transformers already installed (marker found)")
        return True
    
    print("Installing spacy-transformers (for _trf models)...")
    cmd = [sys.executable, "-m", "pip", "install", "spacy-transformers"]
    try:
        subprocess.run(cmd, capture_output=True, text=True, check=True, timeout=900)
        print("✓ Successfully installed spacy-transformers")
        marker_file.touch()
        return True
    except Exception as e:
        print(f"✗ FAILED to install spacy-transformers: {e}")
        if hasattr(e, 'stdout'): print(f"STDOUT: {e.stdout}")
        if hasattr(e, 'stderr'): print(f"STDERR: {e.stderr}")
        return False

def spacy_install_grecy_model_from_github(model_name: str) -> bool:
    """ Installs a greCy model from GitHub Release. """
    marker_file = Path(f".{model_name}_installed")
    if marker_file.exists():
        print(f"✓ {model_name} already installed (marker found)")
        return True
    print(f"Installing grecy model: {model_name}...")
    if model_name == "grc_proiel_trf":
        wheel_filename = "grc_proiel_trf-3.7.5-py3-none-any.whl"
    elif model_name in ["grc_perseus_trf", "grc_proiel_lg", "grc_perseus_lg",
                        "grc_proiel_sm", "grc_perseus_sm", "grc_ner_trf"]:
        wheel_filename = f"{model_name}-0.0.0-py3-none-any.whl"
    else:
        print(f"✗ Unknown grecy model: {model_name}")
        return False
    install_url = f"https://github.com/CrispStrobe/greCy/releases/download/v1.0-models/{wheel_filename}"
    cmd = [sys.executable, "-m", "pip", "install", install_url, "--no-deps"]
    print(f"Running: {' '.join(cmd)}")
    try:
        result = subprocess.run(cmd, capture_output=True, text=True, check=True, timeout=900)
        if result.stdout: print("STDOUT:", result.stdout)
        if result.stderr: print("STDERR:", result.stderr)
        print(f"✓ Successfully installed {model_name} from GitHub")
        marker_file.touch()
        return True
    except subprocess.CalledProcessError as e:
        print(f"✗ Installation subprocess FAILED with code {e.returncode}")
        print("STDOUT:", e.stdout)
        print("STDERR:", e.stderr)
        return False
    except Exception as e:
        print(f"✗ Installation exception: {e}")
        traceback.print_exc()
        return False

# --- Model Loading (Lazy Loading) ---
def spacy_load_spacy_model(model_name: str) -> Optional[spacy.Language]:
    """Load or install a standard spaCy model."""
    try:
        return spacy.load(model_name)
    except OSError:
        print(f"Installing {model_name}...")
        try:
            subprocess.check_call([sys.executable, "-m", "spacy", "download", model_name])
            return spacy.load(model_name)
        except Exception as e:
            print(f"✗ Failed to install {model_name}: {e}")
            if hasattr(e, 'stderr'): print(f"STDERR: {e.stderr}")
            return None

def spacy_load_grecy_model(model_name: str) -> Optional[spacy.Language]:
    """ Load a grecy model, installing from GitHub if needed. """
    if not spacy_install_grecy_model_from_github(model_name):
        print(f"✗ Cannot load {model_name} because installation failed.")
        return None
    try:
        print("Refreshing importlib to find new package...")
        importlib.invalidate_caches()
        try: importlib.reload(site)
        except Exception: pass
        print(f"Trying: spacy.load('{model_name}')")
        nlp = spacy.load(model_name)
        print(f"✓ Successfully loaded {model_name}")
        return nlp
    except Exception as e:
        print(f"✗ Model {model_name} is installed but FAILED to load.")
        print(f"   Error: {e}")
        traceback.print_exc()
        return None

def spacy_initialize_models():
    """ Pre-load standard models and ensure _trf dependencies are ready. """
    print("\n" + "="*70)
    print("INITIALIZING SPACY MODELS")
    print("="*70 + "\n")
    spacy_install_spacy_transformers_once()
    loaded_count = 0
    spacy_model_count = 0
    for lang_code, (lang_name, model_name, model_type) in SPACY_MODEL_INFO.items():
        if model_type == "spacy":
            spacy_model_count += 1
            print(f"Loading {lang_name} ({model_name})...")
            nlp = spacy_load_spacy_model(model_name)
            SPACY_MODELS[lang_code] = nlp
            if nlp:
                print(f"✓ {lang_name} ready\n")
                loaded_count += 1
            else:
                print(f"✗ {lang_name} FAILED\n")
        else:
            print(f"✓ {lang_name} ({model_name}) will be loaded on first use.\n")
            SPACY_MODELS[lang_code] = None
    print(f"Pre-loaded {loaded_count}/{spacy_model_count} standard models.")
    print("="*70 + "\n")

# --- Analysis Logic ---
def spacy_get_analysis(ui_lang: str, model_lang_key: str, text: str):
    """Analyze text and return results."""
    ui_config = SPACY_UI_TEXT.get(ui_lang.lower(), SPACY_UI_TEXT["en"])
    error_prefix = ui_config["error_message"]
    try:
        if not text.strip():
            return ([], [], "<p style='color: orange;'>No text provided.</p>", "<p>No text provided.</p>",
                    gr.Button(value=ui_config["button_text"], interactive=True))
        nlp = SPACY_MODELS.get(model_lang_key)
        if nlp is None:
            print(f"First use of {model_lang_key}. Loading model...")
            if model_lang_key not in SPACY_MODEL_INFO:
                 raise ValueError(f"Unknown model key: {model_lang_key}")
            _, model_name, model_type = SPACY_MODEL_INFO[model_lang_key]
            if model_type == "grecy":
                nlp = spacy_load_grecy_model(model_name)
            else:
                nlp = spacy_load_spacy_model(model_name)
            if nlp is None:
                SPACY_MODELS.pop(model_lang_key, None)
                err_msg = f"Model for {model_lang_key} ({model_name}) FAILED to load. Check logs."
                err_html = f"<p style='color: red;'>{err_msg}</p>"
                return ([], {"error": err_msg}, err_html, err_html,
                        gr.Button(value=ui_config["button_text"], interactive=True))
            else:
                SPACY_MODELS[model_lang_key] = nlp
                print(f"✓ {model_lang_key} is now loaded and cached.")
        doc = nlp(text)
        dataframe_output = []
        json_output = []
        for token in doc:
            lemma_str = token.lemma_
            morph_str = str(token.morph) if token.morph else ''
            dep_str = token.dep_ if doc.is_parsed else ''
            tag_str = token.tag_ or ''
            pos_str = token.pos_ or ''
            json_output.append({
                "word": token.text, "lemma": lemma_str, "pos": pos_str,
                "tag": tag_str, "morphology": morph_str, "dependency": dep_str,
                "is_stopword": token.is_stop
            })
            dataframe_output.append([token.text, lemma_str, pos_str, tag_str, morph_str, dep_str])
        html_dep_out = ""
        if "parser" in nlp.pipe_names and doc.is_parsed:
            try:
                options = {"compact": True, "bg": "#ffffff", "color": "#000000", "font": "Source Sans Pro"}
                html_svg = displacy.render(doc, style="dep", jupyter=False, options=options)
                html_dep_out = _html_wrap(html_svg, line_height="2.5")
            except Exception as e:
                html_dep_out = f"<p style='color: orange;'>Visualization error (DEP): {e}</p>"
        else:
            html_dep_out = "<p style='color: orange;'>Dependency parsing ('parser') not available or doc not parsed.</p>"
        html_ner_out = ""
        if "ner" in nlp.pipe_names:
            if doc.ents:
                try:
                    html_ner = displacy.render(doc, style="ent", jupyter=False)
                    html_ner_out = _html_wrap(html_ner, line_height="2.5")
                except Exception as e:
                    html_ner_out = f"<p style='color: orange;'>Visualization error (NER): {e}</p>"
            else:
                html_ner_out = "<p>No named entities found in this text.</p>"
        else:
            html_ner_out = "<p style='color: orange;'>Named Entity Recognition ('ner') not available for this model.</p>"
        return (dataframe_output, json_output, html_dep_out, html_ner_out,
                gr.Button(value=ui_config["button_text"], interactive=True))
    except Exception as e:
        traceback.print_exc()
        error_html = f"<div style='color: red; border: 1px solid red; padding: 10px; border-radius: 5px; background-color: #fff5f5;'><strong>{error_prefix}</strong> {str(e)}</div>"
        return ([], {"error": str(e)}, error_html, error_html,
                gr.Button(value=ui_config["button_text"], interactive=True))

# --- UI Update Logic ---
def spacy_update_ui(ui_lang: str):
    """Update UI language for the spaCy tab."""
    ui_config = SPACY_UI_TEXT.get(ui_lang.lower(), SPACY_UI_TEXT["en"])
    return [
        gr.update(value=ui_config["title"]),
        gr.update(value=ui_config["subtitle"]),
        gr.update(label=ui_config["ui_lang_label"]),
        gr.update(label=ui_config["model_lang_label"]),
        gr.update(label=ui_config["input_label"], placeholder=ui_config["input_placeholder"]),
        gr.update(value=ui_config["button_text"]),
        gr.update(label=ui_config["tab_graphic"]),
        gr.update(label=ui_config["tab_table"]),
        gr.update(label=ui_config["tab_json"]),
        gr.update(label=ui_config["tab_ner"]),
        gr.update(label=ui_config["html_label"]),
        gr.update(label=ui_config["table_label"], headers=ui_config["table_headers"]),
        gr.update(label=ui_config["json_label"]),
        gr.update(label=ui_config["ner_label"])
    ]

# ============================================================================
# 3b. IWNLP PIPELINE (NEW)
# ============================================================================
IWNLP_PIPELINE: Optional[spacy.Language] = None
IWNLP_LOCK = threading.Lock()

# Define paths for the data
DATA_DIR = "data"
LEMMATIZER_JSON_NAME = "IWNLP.Lemmatizer_20181001.json"
LEMMATIZER_JSON_PATH = os.path.join(DATA_DIR, LEMMATIZER_JSON_NAME)
LEMMATIZER_ZIP_URL = "https://dbs.cs.uni-duesseldorf.de/datasets/iwnlp/IWNLP.Lemmatizer_20181001.zip"
LEMMATIZER_ZIP_PATH = os.path.join(DATA_DIR, "IWNLP.Lemmatizer_20181001.zip")

def iwnlp_download_and_unzip_data():
    """
    Checks for IWNLP data file. Downloads and unzips if not present.
    """
    if os.path.exists(LEMMATIZER_JSON_PATH):
        print("✓ IWNLP data file already exists.")
        return True

    # --- File not found, must download and unzip ---
    try:
        os.makedirs(DATA_DIR, exist_ok=True)
        
        # 1. Download the ZIP file if it's not already here
        if not os.path.exists(LEMMATIZER_ZIP_PATH):
            print(f"IWNLP data not found. Downloading from {LEMMATIZER_ZIP_URL}...")
            with requests.get(LEMMATIZER_ZIP_URL, stream=True) as r:
                r.raise_for_status()
                with open(LEMMATIZER_ZIP_PATH, 'wb') as f:
                    for chunk in r.iter_content(chunk_size=8192): 
                        f.write(chunk)
            print("✓ IWNLP Download complete.")
        else:
            print("✓ IWNLP zip file already present.")

        # 2. Unzip the file
        print(f"Unzipping '{LEMMATIZER_ZIP_PATH}'...")
        with zipfile.ZipFile(LEMMATIZER_ZIP_PATH, 'r') as zip_ref:
            # Extract the specific file we need to the data directory
            zip_ref.extract(LEMMATIZER_JSON_NAME, path=DATA_DIR)
        print(f"✓ Unzip complete. File extracted to {LEMMATIZER_JSON_PATH}")
        
        if not os.path.exists(LEMMATIZER_JSON_PATH):
            raise Exception("Unzip appeared to succeed, but the .json file is still missing.")
            
        return True

    except Exception as e:
        print(f"✗ CRITICAL: Failed to download or unzip IWNLP data: {e}")
        traceback.print_exc()
        return False


def iwnlp_get_pipeline() -> Optional[spacy.Language]:
    """ Thread-safe function to get a single instance of the IWNLP pipeline. """
    global IWNLP_PIPELINE
    if not IWNLP_AVAILABLE:
        raise ImportError("spacy-iwnlp library is not installed.")
    
    if IWNLP_PIPELINE:
        return IWNLP_PIPELINE
        
    with IWNLP_LOCK:
        if IWNLP_PIPELINE:
            return IWNLP_PIPELINE
            
        try:
            print("Initializing spaCy-IWNLP pipeline...")
            
            # --- 1. Ensure data file exists ---
            if not iwnlp_download_and_unzip_data():
                return None # Failed to get data

            # --- 2. Load spaCy model ---
            print("Loading 'de_core_news_md' for IWNLP...")
            nlp_de = SPACY_MODELS.get("de")
            if not nlp_de:
                nlp_de = spacy_load_spacy_model("de_core_news_md")
                if nlp_de:
                    SPACY_MODELS["de"] = nlp_de
                else:
                    raise Exception("Failed to load 'de_core_news_md' for IWNLP.")
            
            # --- 3. Add IWNLP pipe ---
            if not nlp_de.has_pipe("iwnlp"):
                # This is the V3.0 initialization method
                nlp_de.add_pipe('iwnlp', config={'lemmatizer_path': LEMMATIZER_JSON_PATH})
                print("✓ IWNLP pipe added to 'de' model.")
            else:
                print("✓ IWNLP pipe already present.")
                
            IWNLP_PIPELINE = nlp_de
            return IWNLP_PIPELINE
            
        except Exception as e:
            print(f"CRITICAL ERROR: Failed to initialize IWNLP pipeline: {e}")
            traceback.print_exc()
            return None

# ============================================================================
# 4. LANGUAGETOOL LOGIC
# ============================================================================
# --- Globals for LanguageTool ---
LT_TOOL_INSTANCE: Optional[language_tool_python.LanguageTool] = None
LT_TOOL_LOCK = threading.Lock()
def lt_get_language_tool() -> Optional[language_tool_python.LanguageTool]:
    """ Thread-safe function to get a single instance of the LanguageTool. """
    global LT_TOOL_INSTANCE
    if not LT_AVAILABLE:
        raise ImportError("language-tool-python library is not installed.")
    if LT_TOOL_INSTANCE:
        return LT_TOOL_INSTANCE
    with LT_TOOL_LOCK:
        if LT_TOOL_INSTANCE:
            return LT_TOOL_INSTANCE
        try:
            print("Initializing LanguageTool for German (de-DE)...")
            tool = language_tool_python.LanguageTool('de-DE')
            try:
                tool.picky = True
            except Exception:
                pass
            _ = tool.check("Dies ist ein Test.")
            print("LanguageTool (local server) initialized successfully.")
            LT_TOOL_INSTANCE = tool
            return LT_TOOL_INSTANCE
        except Exception as e:
            print(f"CRITICAL ERROR: Failed to initialize LanguageTool: {e}")
            return None
# --- Grammar Checking Logic ---
def lt_check_grammar(text: str) -> List[Dict[str, Any]]:
    """ Checks a German text for grammar and spelling errors and returns a JSON list. """
    try:
        tool = lt_get_language_tool()
        if tool is None:
            return [{"error": "LanguageTool service failed to initialize."}]
        if not text or not text.strip():
            return [{"info": "No text provided to check."}]
        print(f"Checking text: {text}")
        matches = tool.check(text)
        if not matches:
            try:
                tool.picky = True
                matches = tool.check(text)
            except Exception:
                pass
        if not matches:
            return [{"info": "No errors found!", "status": "perfect"}]
        errors_list = []
        for match in matches:
            error = {
                "message": match.message,
                "rule_id": match.ruleId,
                "category": getattr(match.category, 'name', match.category),
                "incorrect_text": text[match.offset : match.offset + match.errorLength],
                "replacements": match.replacements,
                "offset": match.offset,
                "length": match.errorLength,
                "context": getattr(match, "context", None),
                "short_message": getattr(match, "shortMessage", None)
            }
            errors_list.append(error)
        print(f"Found {len(errors_list)} errors.")
        return errors_list
    except Exception as e:
        traceback.print_exc()
        return [{"error": f"An unexpected error occurred: {str(e)}"}]

# ============================================================================
# 5. ODENET THESAURUS LOGIC
# ============================================================================
# --- Globals & Classes for OdeNet ---
@dataclass
class OdeNetWorkItem:
    """Represents a lookup request."""
    word: str
    response_queue: queue.Queue
class OdeNetWorkerState(Enum):
    NOT_STARTED = 1
    INITIALIZING = 2
    READY = 3
    ERROR = 4
odenet_worker_state = OdeNetWorkerState.NOT_STARTED
odenet_worker_thread = None
odenet_work_queue = queue.Queue()
odenet_de_wn = None
# --- Worker Thread Logic ---
def odenet_download_wordnet_data():
    """Download WordNet data. Called once by worker thread."""
    if not WN_AVAILABLE:
        print("[OdeNet Worker] 'wn' library not available. Skipping download.")
        return False
    try:
        print("[OdeNet Worker] Downloading WordNet data...")
        try:
            wn.download('odenet:1.4')
        except Exception as e:
            print(f"[OdeNet Worker] Note: odenet download: {e}")
        try:
            wn.download('cili:1.0')
        except Exception as e:
            print(f"[OdeNet Worker] Note: cili download: {e}")
        print("[OdeNet Worker] ✓ WordNet data ready")
        return True
    except Exception as e:
        print(f"[OdeNet Worker] ✗ Failed to download WordNet data: {e}")
        return False
def odenet_worker_loop():
    """ Worker thread main loop. """
    global odenet_worker_state, odenet_de_wn
    if not WN_AVAILABLE:
        print("[OdeNet Worker] 'wn' library not available. Worker cannot start.")
        odenet_worker_state = OdeNetWorkerState.ERROR
        return
    try:
        print("[OdeNet Worker] Starting worker thread...")
        odenet_worker_state = OdeNetWorkerState.INITIALIZING
        if not odenet_download_wordnet_data():
            odenet_worker_state = OdeNetWorkerState.ERROR
            print("[OdeNet Worker] Failed to initialize")
            return
        print("[OdeNet Worker] Creating WordNet instance...")
        odenet_de_wn = wn.Wordnet('odenet:1.4')
        odenet_worker_state = OdeNetWorkerState.READY
        print("[OdeNet Worker] Ready to process requests")
        while True:
            try:
                item: OdeNetWorkItem = odenet_work_queue.get(timeout=1)
                try:
                    result = odenet_process_word_lookup(item.word)
                    item.response_queue.put(("success", result))
                except Exception as e:
                    traceback.print_exc()
                    item.response_queue.put(("error", str(e)))
                finally:
                    odenet_work_queue.task_done()
            except queue.Empty:
                continue
    except Exception as e:
        print(f"[OdeNet Worker] Fatal error: {e}")
        traceback.print_exc()
        odenet_worker_state = OdeNetWorkerState.ERROR
def odenet_process_word_lookup(word: str) -> Dict[str, Any]:
    """ Process a single word lookup. Runs in the worker thread. """
    global odenet_de_wn
    if not word or not word.strip():
        return {"info": "No word provided to check."}
    word = word.strip().lower()
    senses = odenet_de_wn.senses(word)
    if not senses:
        return {"info": f"The word '{word}' was not found in the thesaurus."}
    results: Dict[str, Any] = {"input_word": word, "senses": []}
    for sense in senses:
        synset = sense.synset()
        def get_lemmas(synsets, remove_self=False):
            lemmas: Set[str] = set()
            for s in synsets:
                for lemma in s.lemmas():
                    if not (remove_self and lemma == word):
                        lemmas.add(lemma)
            return sorted(list(lemmas))
        antonym_words: Set[str] = set()
        try:
            for ant_sense in sense.get_related('antonym'):
                antonym_words.add(ant_sense.word().lemma())
        except Exception:
            pass
        sense_info = {
            "pos": synset.pos,
            "definition": synset.definition() or "No definition available.",
            "synonyms": get_lemmas([synset], remove_self=True),
            "antonyms": sorted(list(antonym_words)),
            "hypernyms (is a type of)": get_lemmas(synset.hypernyms()),
            "hyponyms (examples are)": get_lemmas(synset.hyponyms()),
            "holonyms (is part of)": get_lemmas(synset.holonyms()),
            "meronyms (has parts)": get_lemmas(synset.meronyms()),
        }
        results["senses"].append(sense_info)
    print(f"[OdeNet Worker] Found {len(results['senses'])} senses for '{word}'")
    return results
def odenet_start_worker():
    """Start the worker thread if not already started."""
    global odenet_worker_thread, odenet_worker_state
    if odenet_worker_state != OdeNetWorkerState.NOT_STARTED:
        return
    if not WN_AVAILABLE:
        print("[OdeNet] 'wn' library not available. Worker will not be started.")
        odenet_worker_state = OdeNetWorkerState.ERROR
        return
    odenet_worker_thread = threading.Thread(target=odenet_worker_loop, daemon=True, name="OdeNetWorker")
    odenet_worker_thread.start()
    timeout = 30
    for _ in range(timeout * 10):
        if odenet_worker_state in (OdeNetWorkerState.READY, OdeNetWorkerState.ERROR):
            break
        threading.Event().wait(0.1)
    if odenet_worker_state != OdeNetWorkerState.READY:
        raise Exception("OdeNet Worker failed to initialize")
# --- Public API (Called by Gradio) ---
def odenet_get_thesaurus_info(word: str) -> Dict[str, Any]:
    """ Public API: Finds thesaurus info for a German word. Thread-safe. """
    if not WN_AVAILABLE:
        return {"error": "WordNet (wn) library is not available."}
    if odenet_worker_state != OdeNetWorkerState.READY:
        return {"error": "WordNet service is not ready yet. Please try again in a moment."}
    try:
        response_queue = queue.Queue()
        item = OdeNetWorkItem(word=word, response_queue=response_queue)
        odenet_work_queue.put(item)
        try:
            status, result = response_queue.get(timeout=30)
            if status == "success":
                return result
            else:
                return {"error": f"Lookup failed: {result}"}
        except queue.Empty:
            return {"error": "Request timed out"}
    except Exception as e:
        traceback.print_exc()
        return {"error": f"An unexpected error occurred: {str(e)}"}

# ============================================================================
# 6. PATTERN INFLECTION LOGIC
# ============================================================================
# --- Word Type Detection ---
def pattern_detect_word_type(word: str) -> Dict[str, Any]:
    """ Use pattern.de's parser as a hint. """
    if not PATTERN_DE_AVAILABLE:
        return {'pos': None, 'lemma': word, 'type': 'unknown'}
    if not word or not word.strip() or all(ch in ".,;:!?()[]{}-–—'.../\|" for ch in word):
        return {'pos': None, 'lemma': word, 'type': 'unknown'}
    word_norm = word.strip()
    log(f"Detecting type for: {word_norm}")
    parser_result = {'pos': None, 'lemma': word_norm, 'type': None}
    try:
        parsed = parse(word_norm, lemmata=True)
        for sentence in split(parsed):
            if hasattr(sentence, "words") and sentence.words:
                w = sentence.words[0]
                w_type = getattr(w, "type", None) or getattr(w, "pos", None)
                w_lemma = (getattr(w, "lemma", None) or word_norm)
                non_content_prefixes = ("DT","ART","IN","APPR","APPRART","APPO","APZR","PTK","PRP","PPER","PPOS","PDS","PIS","KOUI","KON","$,","$.")
                if w_type and any(w_type.startswith(p) for p in non_content_prefixes):
                    return {'pos': w_type, 'lemma': w_lemma, 'type': None}
                parser_result['pos'] = w_type or ""
                parser_result['lemma'] = w_lemma
                if w_type and w_type.startswith('NN'):
                    parser_result['type'] = 'noun'
                elif w_type and w_type.startswith('VB'):
                    parser_result['type'] = 'verb'
                elif w_type and w_type.startswith('JJ'):
                    parser_result['type'] = 'adjective'
                log(f"  Parser says: POS={w_type}, lemma={w_lemma}, type={parser_result['type']}")
    except Exception as e:
        log(f"  Parser failed: {e}")
    return parser_result

def pattern_is_good_analysis(analysis, analysis_type):
    """Check if an analysis has meaningful data."""
    if not analysis: return False
    
    if analysis_type == 'noun':
        # Check for declensions, either in the simple or ambiguous map
        return len(analysis.get('declension', {})) >= 4 or len(analysis.get('declension_by_gender', {})) > 0
        
    elif analysis_type == 'verb':
        present = analysis.get('conjugation', {}).get('Präsens', {})
        if len(present) < 4: return False
        unique_forms = set(present.values())
        if len(unique_forms) < 2: return False
        return True
        
    elif analysis_type == 'adjective':
        # **FIX: Better adjective validation**
        # Must have attributive forms
        if len(analysis.get('attributive', {})) == 0:
            log("    ✗ Not a good adjective: No attributive forms.")
            return False
        
        pred = analysis.get('predicative', '')
        comp = analysis.get('comparative', '')
        sup = analysis.get('superlative', '')
        
        if not pred:
            log("    ✗ Not a good adjective: No predicative form.")
            return False

        # Filter out nonsense: "lauf" -> "laufer", "laufst"
        # Real comparatives end in -er. Real superlatives end in -st or -est.
        # This allows "rasch" (rascher, raschst) but rejects "lauf" (laufer, laufst)
        if comp and not comp.endswith("er"):
            log(f"    ✗ Not a good adjective: Comparative '{comp}' doesn't end in -er.")
            return False
        if sup and not (sup.endswith("st") or sup.endswith("est")):
            log(f"    ✗ Not a good adjective: Superlative '{sup}' doesn't end in -st/-est.")
            return False
            
        return True
        
    return False


# --- Inflection Generators ---
def pattern_analyze_as_noun(word: str, hint_lemma: str = None, fixed_gender: int = None) -> Dict[str, Any]:
    """
    Comprehensive noun inflection analysis.
    Args:
        hint_lemma: A lemma suggestion to help Pattern.
        fixed_gender: A pattern.de constant (MALE, FEMALE, NEUTRAL) to FORCE a specific gender.
    """
    log(f"   Analyzing as noun (hint_lemma={hint_lemma}, fixed_gender={fixed_gender})")
    analysis = {}
    
    # 1. Determine Base Form
    singular = singularize(word)
    plural = pluralize(word)
    
    if plural != word and singular != word:
        base = word
    elif singular != word:
        base = singular
    elif hint_lemma and hint_lemma != word:
        base = hint_lemma
    else:
        base = word
    
    # 2. Determine Gender
    # If Wiktionary gave us a gender, USE IT. Ignore Pattern's internal dictionary.
    if fixed_gender is not None:
        genders = [fixed_gender]
        log(f"   [Pattern] Enforcing gender from DB: {fixed_gender}")
    else:
        # Fallback to auto-detection
        g = gender(base, pos=NOUN)
        if isinstance(g, tuple):
            genders = list(g)
        elif g is None:
            genders = [MALE] 
        else:
            genders = [g]
        
    analysis["base_form"] = base
    analysis["plural"] = pluralize(base)
    analysis["singular"] = base
    analysis["declension_by_gender"] = {}

    # 3. Generate Declensions
    for gen in genders:
        gender_str = {MALE: "Masculine", FEMALE: "Feminine", NEUTRAL: "Neuter"}.get(gen, "Unknown")
        gen_declension = {}
        
        for number, number_name in [(SINGULAR, "Singular"), (PLURAL, "Plural")]:
            word_form = base if number == SINGULAR else pluralize(base)
            word_form_cap = word_form.capitalize()
            gender_for_article = gen if number == SINGULAR else PLURAL
            
            for case, case_name in [(NOMINATIVE, "Nominativ"), (ACCUSATIVE, "Akkusativ"), 
                                    (DATIVE, "Dativ"), (GENITIVE, "Genitiv")]:
                try:
                    def_art = article(word_form, DEFINITE, gender_for_article, case)
                    indef_art = article(word_form, INDEFINITE, gender_for_article, case)
                    
                    indef_form = f"{indef_art} {word_form_cap}" if indef_art else word_form_cap
                    if number == PLURAL: indef_form = "—"
                    
                    # Fix for Pattern sometimes missing Genitive 's' suffix on Masculine/Neuter
                    noun_text = word_form_cap
                    if number == SINGULAR and case == GENITIVE and gen in [MALE, NEUTRAL] and not noun_text.endswith("s") and not noun_text.endswith("x") and not noun_text.endswith("z"):
                         # Simple heuristic fix: German Genitive usually adds 's' or 'es'
                         # Pattern handles this usually, but if we force gender on a word Pattern doesn't know, it might miss it.
                         # For safety, we trust Pattern's output, but if you find Pattern fails here, you inject logic here.
                         pass 

                    gen_declension[f"{case_name} {number_name}"] = {
                        "definite": f"{def_art} {noun_text}" if def_art else noun_text,
                        "indefinite": indef_form,
                        "bare": noun_text
                    }
                except Exception as e:
                    log(f"    Failed to get article for {gender_str}/{case_name} {number_name}: {e}")
        
        analysis["declension_by_gender"][gender_str] = gen_declension
    
    # Flatten for the main keys if only one gender exists
    if len(genders) == 1:
        first_gen_key = list(analysis["declension_by_gender"].keys())[0]
        analysis["declension"] = analysis["declension_by_gender"][first_gen_key]
        analysis["gender"] = first_gen_key

    return analysis

def pattern_analyze_as_verb(word: str, hint_lemma: str = None) -> Dict[str, Any]:
    """Comprehensive verb conjugation analysis."""
    log(f"  Analyzing as verb (hint_lemma={hint_lemma})")
    verb_lemma = lemma(word)
    log(f"    lemma({word}) = {verb_lemma}")

    # If the input word is already an infinitive (ends in 'en', 'n', 'ln'),
    # and pattern.de gives a weird lemma, trust the input word.
    # This fixes lemma('gießen') -> 'gaßen'
    is_infinitive_form = word.endswith("en") or word.endswith("ln") or word.endswith("rn")
    if is_infinitive_form and verb_lemma != word.lower():
        log(f"    Pattern.de lemma '{verb_lemma}' is suspicious for infinitive '{word}'. Trusting input word.")
        verb_lemma = word
        
    if not verb_lemma or verb_lemma == word:
        if hint_lemma and hint_lemma != word:
            verb_lemma = hint_lemma
            log(f"    Using hint lemma: {verb_lemma}")
        elif not verb_lemma:
            log(f"    No lemma found, trying base word")
            verb_lemma = word # e.g. "lauf"
    
    analysis = {"infinitive": verb_lemma}
    try:
        lex = lexeme(verb_lemma)
        if lex and len(lex) > 1:
            analysis["lexeme"] = lex
            log(f"    lexeme has {len(lex)} forms")
    except Exception as e:
        log(f"    Failed to get lexeme: {e}")
    analysis["conjugation"] = {}
    analysis["conjugation"]["Präsens"] = {}
    present_count = 0
    for alias, name in [("1sg", "ich"), ("2sg", "du"), ("3sg", "er/sie/es"),
                        ("1pl", "wir"), ("2pl", "ihr"), ("3pl", "sie/Sie")]:
        try:
            form = conjugate(verb_lemma, alias)
            if form:
                analysis["conjugation"]["Präsens"][name] = form
                present_count += 1
        except Exception as e:
            log(f"    Failed conjugate({verb_lemma}, {alias}): {e}")
    log(f"    Generated {present_count} present tense forms")
    if present_count < 4:
        # Try again with infinitive, e.g. if input was "lauf"
        try:
            verb_lemma = conjugate(word, INFINITIVE)
            log(f"    Retrying with infinitive '{verb_lemma}'")
            analysis["infinitive"] = verb_lemma
            present_count = 0
            for alias, name in [("1sg", "ich"), ("2sg", "du"), ("3sg", "er/sie/es"),
                                ("1pl", "wir"), ("2pl", "ihr"), ("3pl", "sie/Sie")]:
                form = conjugate(verb_lemma, alias)
                if form:
                    analysis["conjugation"]["Präsens"][name] = form
                    present_count += 1
            if present_count < 4:
                log(f"    Too few present forms, not a valid verb")
                return None
        except Exception as e:
             log(f"    Retry failed, not a valid verb: {e}")
             return None

    analysis["conjugation"]["Präteritum"] = {}
    for alias, name in [("1sgp", "ich"), ("2sgp", "du"), ("3sgp", "er/sie/es"),
                        ("1ppl", "wir"), ("2ppl", "ihr"), ("3ppl", "sie/Sie")]:
        try:
            form = conjugate(verb_lemma, alias)
            if form: analysis["conjugation"]["Präteritum"][name] = form
        except: pass
    analysis["participles"] = {}
    try:
        form = conjugate(verb_lemma, "part")
        if form: analysis["participles"]["Partizip Präsens"] = form
    except: pass
    try:
        form = conjugate(verb_lemma, "ppart")
        if form: analysis["participles"]["Partizip Perfekt"] = form
    except: pass
    analysis["conjugation"]["Imperativ"] = {}
    for alias, name in [("2sg!", "du"), ("2pl!", "ihr")]:
        try:
            form = conjugate(verb_lemma, alias)
            if form: analysis["conjugation"]["Imperativ"][name] = form
        except: pass
    analysis["conjugation"]["Konjunktiv I"] = {}
    for alias, name in [("1sg?", "ich"), ("2sg?", "du"), ("3sg?", "er/sie/es"),
                        ("1pl?", "wir"), ("2pl?", "ihr"), ("3pl?", "sie/Sie")]:
        try:
            form = conjugate(verb_lemma, alias)
            if form: analysis["conjugation"]["Konjunktiv I"][name] = form
        except: pass
    analysis["conjugation"]["Konjunktiv II"] = {}
    for alias, name in [("1sgp?", "ich"), ("2sgp?", "du"), ("3sgp?", "er/sie/es"),
                        ("1ppl?", "wir"), ("2ppl?", "ihr"), ("3ppl?", "sie/Sie")]:
        try:
            form = conjugate(verb_lemma, alias)
            if form: analysis["conjugation"]["Konjunktiv II"][name] = form
        except: pass
    return analysis

def pattern_analyze_as_adjective(word: str, hint_lemma: str = None) -> Dict[str, Any]:
    """Comprehensive adjective inflection analysis."""
    log(f"  Analyzing as adjective (hint_lemma={hint_lemma})")
    base = predicative(word)
    log(f"    predicative({word}) = {base}")
    if base == word.lower() and hint_lemma and hint_lemma != word:
        base = hint_lemma
        log(f"    Using hint lemma: {base}")

    analysis = {}
    analysis["predicative"] = base
    
    # *** FIX: Removed pos=ADJECTIVE, which was causing a crash ***
    try:
        analysis["comparative"] = comparative(base)
    except Exception as e:
        log(f"    Failed to get comparative: {e}")
        analysis["comparative"] = f"{base}er" # Fallback
        
    try:
        analysis["superlative"] = superlative(base)
    except Exception as e:
        log(f"    Failed to get superlative: {e}")
        analysis["superlative"] = f"{base}st" # Fallback
    
    log(f"    comparative = {analysis['comparative']}")
    log(f"    superlative = {analysis['superlative']}")
    
    analysis["attributive"] = {}
    attr_count = 0
    for article_type, article_name in [(None, "Strong"), (INDEFINITE, "Mixed"), (DEFINITE, "Weak")]:
        analysis["attributive"][article_name] = {}
        for gender, gender_name in [(MALE, "Masculine"), (FEMALE, "Feminine"),
                                    (NEUTRAL, "Neuter"), (PLURAL, "Plural")]:
            analysis["attributive"][article_name][gender_name] = {}
            for case, case_name in [(NOMINATIVE, "Nom"), (ACCUSATIVE, "Acc"),
                                    (DATIVE, "Dat"), (GENITIVE, "Gen")]:
                try:
                    attr_form = attributive(base, gender, case, article_type)
                    if article_type:
                        art = article("_", article_type, gender, case)
                        full_form = f"{art} {attr_form} [Noun]" if art else f"{attr_form} [Noun]"
                    else:
                        full_form = f"{attr_form} [Noun]"
                    analysis["attributive"][article_name][gender_name][case_name] = {
                        "form": attr_form, "example": full_form
                    }
                    attr_count += 1
                except Exception as e:
                    log(f"    Failed attributive for {article_name}/{gender_name}/{case_name}: {e}")
                    
    log(f"    Generated {attr_count} attributive forms")
    if attr_count == 0:
        return None
    return analysis
# --- Public API (Called by Gradio) ---
def pattern_get_all_inflections(word: str) -> Dict[str, Any]:
    """
    Generates ALL possible inflections for a German word.
    Analyzes the word as-is AND its lowercase version to catch
    ambiguities like "Lauf" (noun) vs "lauf" (verb).
    """
    if not PATTERN_DE_AVAILABLE:
        return {"error": "`PatternLite` library not available."}
    if not word or not word.strip():
        return {"info": "Please enter a word."}
    word = word.strip()
    word_lc = word.lower()
    log("="*70); log(f"ANALYZING: {word} (and {word_lc})"); log("="*70)
    
    # --- Analyze word as-is (e.g., "Lauf") ---
    detection_as_is = pattern_detect_word_type(word)
    analyses_as_is: Dict[str, Any] = {}
    try:
        log("\n--- Trying analysis for: " + word + " ---")
        noun_analysis_as_is = pattern_analyze_as_noun(word, detection_as_is['lemma'])
        if noun_analysis_as_is and pattern_is_good_analysis(noun_analysis_as_is, 'noun'):
            log("✓ Noun analysis is good")
            analyses_as_is["noun"] = noun_analysis_as_is
        verb_analysis_as_is = pattern_analyze_as_verb(word, detection_as_is['lemma'])
        if verb_analysis_as_is and pattern_is_good_analysis(verb_analysis_as_is, 'verb'):
            log("✓ Verb analysis is good")
            analyses_as_is["verb"] = verb_analysis_as_is
        adj_analysis_as_is = pattern_analyze_as_adjective(word, detection_as_is['lemma'])
        if adj_analysis_as_is and pattern_is_good_analysis(adj_analysis_as_is, 'adjective'):
            log("✓ Adjective analysis is good")
            analyses_as_is["adjective"] = adj_analysis_as_is
    except Exception as e:
        log(f"\nERROR during 'as-is' analysis: {e}")
        traceback.print_exc()
        return {"error": f"An unexpected error occurred during 'as-is' analysis: {str(e)}"}

    # --- Analyze lowercase version (e.g., "lauf") if different ---
    analyses_lc: Dict[str, Any] = {}
    if word != word_lc:
        detection_lc = pattern_detect_word_type(word_lc)
        try:
            log("\n--- Trying analysis for: " + word_lc + " ---")
            noun_analysis_lc = pattern_analyze_as_noun(word_lc, detection_lc['lemma'])
            if noun_analysis_lc and pattern_is_good_analysis(noun_analysis_lc, 'noun'):
                log("✓ Noun analysis (lc) is good")
                analyses_lc["noun"] = noun_analysis_lc
            verb_analysis_lc = pattern_analyze_as_verb(word_lc, detection_lc['lemma'])
            if verb_analysis_lc and pattern_is_good_analysis(verb_analysis_lc, 'verb'):
                log("✓ Verb analysis (lc) is good")
                analyses_lc["verb"] = verb_analysis_lc
            adj_analysis_lc = pattern_analyze_as_adjective(word_lc, detection_lc['lemma'])
            if adj_analysis_lc and pattern_is_good_analysis(adj_analysis_lc, 'adjective'):
                log("✓ Adjective analysis (lc) is good")
                analyses_lc["adjective"] = adj_analysis_lc
        except Exception as e:
            log(f"\nERROR during 'lowercase' analysis: {e}")
            traceback.print_exc()
            return {"error": f"An unexpected error occurred during 'lowercase' analysis: {str(e)}"}

    # --- Merge the results ---
    final_analyses = analyses_as_is.copy()
    for key, value in analyses_lc.items():
        if key not in final_analyses:
            final_analyses[key] = value
            
    results: Dict[str, Any] = {
        "input_word": word,
        "analyses": final_analyses
    }
    if not results["analyses"]:
        results["info"] = "Word could not be analyzed as noun, verb, or adjective."
    log(f"\nFinal merged result: {len(results['analyses'])} analysis/analyses")
    return results

def word_appears_in_inflections(word: str, inflections: Dict[str, Any], pos_type: str) -> bool:
    """
    Check if the input word appears in the inflection forms AND
    cross-validate the POS with OdeNet to reject artifacts.
    """
    import re
    word_lower = word.lower()
    word_cap = word.capitalize()

    # 1. Extract all actual inflection forms (not metadata)
    actual_forms = []
    if pos_type == 'noun':
        declension = inflections.get('declension', {})
        declension_by_gender = inflections.get('declension_by_gender', {})
        for case_data in declension.values():
            if isinstance(case_data, dict): actual_forms.append(case_data.get('bare', ''))
        for gender_data in declension_by_gender.values():
            if isinstance(gender_data, dict):
                for case_data in gender_data.values():
                    if isinstance(case_data, dict): actual_forms.append(case_data.get('bare', ''))
                    
    elif pos_type == 'verb':
        conjugation = inflections.get('conjugation', {})
        for tense_data in conjugation.values():
            if isinstance(tense_data, dict): actual_forms.extend(tense_data.values())
        participles = inflections.get('participles', {})
        actual_forms.extend(participles.values())
        actual_forms.extend(inflections.get('lexeme', []))
        actual_forms.append(inflections.get('infinitive', ''))

    elif pos_type == 'adjective':
        actual_forms.append(inflections.get('predicative', ''))
        actual_forms.append(inflections.get('comparative', ''))
        actual_forms.append(inflections.get('superlative', ''))
        attributive = inflections.get('attributive', {})
        for article_data in attributive.values():
            if isinstance(article_data, dict):
                for gender_data in article_data.values():
                    if isinstance(gender_data, dict):
                        for case_data in gender_data.values():
                            if isinstance(case_data, dict): actual_forms.append(case_data.get('form', ''))

    # 2. Clean forms and check for match
    cleaned_forms = set()
    for form in actual_forms:
        if not form or form == '—': continue
        # For simple forms (most verb forms, adjectives), use as-is
        # For complex forms (nouns with articles), extract words
        if ' ' in form or '[' in form:
            words = re.findall(r'\b[\wäöüÄÖÜß]+\b', form)
            cleaned_forms.update(w.lower() for w in words)
        else:
            cleaned_forms.add(form.lower())
            
    articles = {'der', 'die', 'das', 'den', 'dem', 'des', 'ein', 'eine', 'einen', 'einem', 'eines', 'einer'}
    cleaned_forms = {f for f in cleaned_forms if f not in articles}

    word_found_in_forms = False
    if pos_type == 'noun':
        # Nouns can be input as lowercase, but inflections are capitalized.
        # We check if the *lowercase* input word matches a *lowercase* form.
        if word_lower in cleaned_forms:
            word_found_in_forms = True
    else:
        # For verbs/adjectives, a lowercase match is sufficient
        if word_lower in cleaned_forms:
            word_found_in_forms = True
    
    if not word_found_in_forms:
        log(f"    ✗ Word '{word}' not found in any {pos_type} inflection forms.")
        return False

    log(f"    ✓ Word '{word}' was found in the {pos_type} inflection table.")
    
    # 3. Cross-validate POS with OdeNet to filter artifacts (e.g., 'heute' as 'heuen')
    if not WN_AVAILABLE:
        log("    ⚠️  OdeNet (WN_AVAILABLE=False) is not available to validate POS. Accepting pattern.de's analysis.")
        return True

    try:
        if pos_type == 'noun':
            pos_lemma = inflections.get("base_form", word_lower)
            expected_pos_tag = 'n'
        elif pos_type == 'verb':
            pos_lemma = inflections.get("infinitive", word_lower)
            expected_pos_tag = 'v'
        elif pos_type == 'adjective':
            pos_lemma = inflections.get("predicative", word_lower)
            expected_pos_tag = 'a'
        else:
            log(f"    ? Unknown pos_type '{pos_type}' for OdeNet check.")
            return True # Don't block unknown types
            
        log(f"    Validating {pos_type} (lemma: '{pos_lemma}') with OdeNet (expecting pos='{expected_pos_tag}')...")
        odenet_result = odenet_get_thesaurus_info(pos_lemma)
        senses = odenet_result.get('senses', [])
        pos_senses = [s for s in senses if s.get('pos') == expected_pos_tag]
        
        # If no senses for lemma, check input word as fallback
        if not pos_senses and pos_lemma.lower() != word.lower():
            log(f"    No '{expected_pos_tag}' senses for lemma '{pos_lemma}'. Checking input word '{word}'...")
            odenet_result = odenet_get_thesaurus_info(word)
            senses = odenet_result.get('senses', [])
            pos_senses = [s for s in senses if s.get('pos') == expected_pos_tag]

        if not pos_senses:
            log(f"    ✗ REJECTED: OdeNet has no '{expected_pos_tag}' senses for '{pos_lemma}' or '{word}'. This is likely a pattern.de artifact.")
            return False
        else:
            log(f"    ✓ VERIFIED: OdeNet found {len(pos_senses)} '{expected_pos_tag}' sense(s).")
            return True
            
    except Exception as e:
        log(f"    ⚠️  OdeNet validation check failed with error: {e}")
        return True # Fail open: If OdeNet fails, trust pattern.de

# ============================================================================
# 6b. CONCEPTNET HELPER LOGIC (V2 - ROBUST PARSER)
# ============================================================================
def get_conceptnet_client() -> Optional[Client]:
    """ Thread-safe function to get a single instance of the Gradio Client. """
    global CONCEPTNET_CLIENT
    if not GRADIO_CLIENT_AVAILABLE:
        return None
    
    if CONCEPTNET_CLIENT:
        return CONCEPTNET_CLIENT
    
    with CONCEPTNET_CLIENT_LOCK:
        if CONCEPTNET_CLIENT:
            return CONCEPTNET_CLIENT
        try:
            print("Initializing Gradio Client for ConceptNet...")
            client = Client("cstr/conceptnet_normalized")
            print("✓ Gradio Client for ConceptNet initialized.")
            CONCEPTNET_CLIENT = client
            return CONCEPTNET_CLIENT
        except Exception as e:
            print(f"✗ CRITICAL: Failed to initialize ConceptNet Gradio Client: {e}")
            traceback.print_exc()
            return None

def conceptnet_get_relations(word: str, language: str = 'de') -> Dict[str, Any]:
    """
    Fetches relations from the cstr/conceptnet_normalized Gradio API.
    
    This V2 version uses a robust regex parser to correctly handle the
    Markdown output and filter self-referential junk.
    """
    if not GRADIO_CLIENT_AVAILABLE:
        return {"error": "`gradio_client` library is not installed. Install with: pip install gradio_client"}
        
    if not word or not word.strip():
        return {"info": "No word provided."}
        
    word_lower = word.strip().lower()
    cache_key = (word_lower, language)
    
    # --- 1. Check Cache ---
    with CONCEPTNET_LOCK:
        if cache_key in CONCEPTNET_CACHE:
            log(f"ConceptNet: Found '{word_lower}' in cache.")
            return CONCEPTNET_CACHE[cache_key]
            
    log(f"ConceptNet: Fetching '{word_lower}' from Gradio API...")
    
    try:
        # --- 2. Call Gradio API ---
        client = get_conceptnet_client() # <-- USE HELPER
        if not client:
            return {"error": "ConceptNet Gradio Client is not available."}
        
        selected_relations = [
            "RelatedTo", "IsA", "PartOf", "HasA", "UsedFor",
            "CapableOf", "AtLocation", "Synonym", "Antonym",
            "Causes", "HasProperty", "MadeOf", "HasSubevent",
            "DerivedFrom", "SimilarTo", "Desires", "CausesDesire"
        ]
        
        result_markdown = client.predict(
            word=word_lower,
            lang=language,
            selected_relations=selected_relations,
            api_name="/get_semantic_profile"
        )
        
        # --- 3. Parse the Markdown Result (Robustly) ---
        relations_list = []
        if not isinstance(result_markdown, str):
            raise TypeError(f"ConceptNet API returned type {type(result_markdown)}, expected str.")

        lines = result_markdown.split('\n')
        current_relation = None
        
        # Regex to capture: "- <NODE1> <RELATION> → <NODE2> `[WEIGHT]`"
        # Groups:           (1: Node1) (2: Relation) (3: Node2) (4: Weight)
        line_pattern = None

        for line in lines:
            line = line.strip()
            if not line:
                continue

            # Check for relation headers (e.g., "## IsA")
            if line.startswith('## '):
                current_relation = line[3:].strip()
                if current_relation:
                    # Pre-compile the regex for this specific relation
                    line_pattern = re.compile(
                        r"-\s*(.+?)\s+(%s)\s+→\s+(.+?)\s+\`\[([\d.]+)\]\`" % re.escape(current_relation)
                    )
                continue

            # Parse relation entries
            if line.startswith('- ') and current_relation and line_pattern:
                match = line_pattern.search(line)
                
                if not match:
                    log(f"ConceptNet Parser: No match for line '{line}' with relation '{current_relation}'")
                    continue
                    
                try:
                    # Extract parts
                    node1 = match.group(1).strip().strip('*')
                    relation = match.group(2) # This is current_relation
                    node2 = match.group(3).strip().strip('*')
                    weight = float(match.group(4))
                    
                    other_node = None
                    direction = None
                    
                    # Determine direction and filter self-references
                    if node1.lower() == word_lower and node2.lower() != word_lower:
                        other_node = node2
                        direction = "->"
                    elif node2.lower() == word_lower and node1.lower() != word_lower:
                        other_node = node1
                        direction = "<-"
                    else:
                        # This filters "schnell Synonym → schnell"
                        continue 
                        
                    relations_list.append({
                        "relation": relation,
                        "direction": direction,
                        "other_node": other_node,
                        "other_lang": language, # We assume the other node is also in the same lang
                        "weight": weight,
                        "surface": f"{node1} {relation} {node2}"
                    })

                except Exception as e:
                    log(f"ConceptNet Parser: Error parsing line '{line}': {e}")
                    continue
        
        # --- 4. Finalize and Cache Result ---
        if not relations_list:
            final_result = {"info": f"No valid (non-self-referential) relations found for '{word_lower}'."}
        else:
            # Sort by weight, descending
            relations_list.sort(key=lambda x: x.get('weight', 0.0), reverse=True)
            final_result = {"relations": relations_list}
            
        with CONCEPTNET_LOCK:
            CONCEPTNET_CACHE[cache_key] = final_result
            
        log(f"ConceptNet: Returning {len(relations_list)} relations for '{word_lower}'")
        return final_result
        
    except Exception as e:
        error_msg = f"ConceptNet Gradio API request failed: {type(e).__name__} - {e}"
        log(f"ConceptNet API error for '{word_lower}': {e}")
        traceback.print_exc()
        return {"error": error_msg, "traceback": traceback.format_exc()}

# ============================================================================
# 6c. NEW: HANTA INITIALIZER & HELPERS
# ============================================================================

def hanta_get_tagger() -> Optional[HanoverTagger]:
    """ Thread-safe function to get a single instance of the HanTa Tagger. """
    global HANTA_TAGGER_INSTANCE
    if not HANTA_AVAILABLE:
        raise ImportError("HanTa library is not installed.")
    
    if HANTA_TAGGER_INSTANCE:
        return HANTA_TAGGER_INSTANCE
    
    with HANTA_TAGGER_LOCK:
        if HANTA_TAGGER_INSTANCE:
            return HANTA_TAGGER_INSTANCE
        
        try:
            print("Initializing HanTa Tagger (loading model)...")
            PACKAGE_DIR = os.path.dirname(HanTa.HanoverTagger.__file__)
            MODEL_PATH = os.path.join(PACKAGE_DIR, 'morphmodel_ger.pgz')
            
            if not os.path.exists(MODEL_PATH):
                 print(f"CRITICAL: HanTa model file 'morphmodel_ger.pgz' not found at {MODEL_PATH}")
                 raise FileNotFoundError("HanTa model file missing. Please ensure HanTa is correctly installed.")
            
            tagger = HanoverTagger(MODEL_PATH)
            _ = tagger.analyze("Test") # Warm-up call
            print("✓ HanTa Tagger initialized successfully.")
            HANTA_TAGGER_INSTANCE = tagger
            return HANTA_TAGGER_INSTANCE
        except Exception as e:
            print(f"CRITICAL ERROR: Failed to initialize HanTa Tagger: {e}")
            traceback.print_exc()
            return None

def _get_odenet_senses_by_pos(word: str) -> Dict[str, List[Dict[str, Any]]]:
    """
    (Helper) Fetches OdeNet senses for a word and groups them by POS.
    
    *** V18 FIX: OdeNet uses 'a' for BOTH Adjective and Adverb. ***
    """
    senses_by_pos: Dict[str, List[Dict]] = {
        "noun": [], "verb": [], "adjective": [], "adverb": []
    }
    if not WN_AVAILABLE:
        log(f"OdeNet check skipped for '{word}': WN_AVAILABLE=False")
        # If OdeNet is down, we can't validate, so we must return
        # non-empty lists to avoid incorrectly rejecting a POS.
        # This is a "fail-open" strategy.
        return {"noun": [{"info": "OdeNet unavailable"}],
                "verb": [{"info": "OdeNet unavailable"}],
                "adjective": [{"info": "OdeNet unavailable"}],
                "adverb": [{"info": "OdeNet unavailable"}]}
        
    try:
        all_senses = odenet_get_thesaurus_info(word).get("senses", [])
        for sense in all_senses:
            if "error" in sense: continue
            pos_tag = sense.get("pos")
            
            if pos_tag == 'n':
                senses_by_pos["noun"].append(sense)
            elif pos_tag == 'v':
                senses_by_pos["verb"].append(sense)
            
            # --- THIS IS THE CRITICAL FIX ---
            elif pos_tag == 'a':
                log(f"Found OdeNet 'a' tag (Adj/Adv) for sense: {sense.get('definition', '...')[:30]}")
                senses_by_pos["adjective"].append(sense)
                senses_by_pos["adverb"].append(sense)
            # --- END OF FIX ---
                
    except Exception as e:
        log(f"OdeNet helper check failed for '{word}': {e}")
        
    log(f"OdeNet senses for '{word}': "
        f"{len(senses_by_pos['noun'])}N, "
        f"{len(senses_by_pos['verb'])}V, "
        f"{len(senses_by_pos['adjective'])}Adj, "
        f"{len(senses_by_pos['adverb'])}Adv")
    return senses_by_pos

def _hanta_get_candidates(word: str, hanta_tagger: "HanoverTagger") -> Set[str]:
    """
    (Helper) Gets all possible HanTa STTS tags for a word,
    checking both lowercase and capitalized versions.
    """
    all_tags = set()
    try:
        # Check lowercase (for verbs, adjs, advs)
        tags_lower = hanta_tagger.tag_word(word.lower(), cutoff=20)
        all_tags.update(tag[0] for tag in tags_lower)
    except Exception as e:
        log(f"HanTa tag_word (lower) failed for '{word}': {e}")
    
    try:
        # Check capitalized (for nouns)
        tags_upper = hanta_tagger.tag_word(word.capitalize(), cutoff=20)
        all_tags.update(tag[0] for tag in tags_upper)
    except Exception as e:
        log(f"HanTa tag_word (upper) failed for '{word}': {e}")
    
    log(f"HanTa candidates for '{word}': {all_tags}")
    return all_tags

def _hanta_map_tags_to_pos(hanta_tags: Set[str]) -> Dict[str, Set[str]]:
    """
    (Helper) Maps STTS tags to simplified POS groups and injects the
    ADJ(D) -> ADV heuristic.
    """
    pos_groups = {"noun": set(), "verb": set(), "adjective": set(), "adverb": set()}
    has_adjd = False

    for tag in hanta_tags:
        # Nouns (NN), Proper Nouns (NE), Nominalized Inf. (NNI), Nom. Adj. (NNA)
        if tag.startswith("NN") or tag == "NE":
            pos_groups["noun"].add(tag)
        # Verbs (VV...), Auxiliaries (VA...), Modals (VM...)
        elif tag.startswith("VV") or tag.startswith("VA") or tag.startswith("VM"):
            pos_groups["verb"].add(tag)
        # Adjectives (Attributive ADJ(A), Predicative ADJ(D))
        elif tag.startswith("ADJ"):
            pos_groups["adjective"].add(tag)
            if tag == "ADJ(D)":
                has_adjd = True
        # Adverbs
        elif tag == "ADV":
            pos_groups["adverb"].add(tag)
    
    # --- The Core Heuristic ---
    # If HanTa found a predicative adjective (ADJD), it can *also* be used
    # as an adverb (e..g, "er singt schön" [ADV] vs. "er ist schön" [ADJD]).
    if has_adjd:
        log("Injecting ADV possibility based on ADJ(D) tag.")
        pos_groups["adverb"].add("ADV (from ADJD)")

    # Filter out empty groups
    return {k: v for k, v in pos_groups.items() if v}

def _hanta_get_lemma_for_pos(word: str, pos_group: str, hanta_tagger: "HanoverTagger") -> str:
    """
    (Helper) Gets the correct lemma for a given word and POS group
    using case-sensitive analysis.
    """
    lemma = ""
    try:
        if pos_group == "noun":
            # Nouns must be lemmatized from their capitalized form
            lemma = hanta_tagger.analyze(word.capitalize(), casesensitive=True)[0]
        elif pos_group == "verb":
            # Verbs must be lemmatized from their lowercase form
            lemma = hanta_tagger.analyze(word.lower(), casesensitive=True)[0]
        elif pos_group == "adjective":
            # Adjectives are lemmatized from their lowercase form
            lemma = hanta_tagger.analyze(word.lower(), casesensitive=True)[0]
        elif pos_group == "adverb":
            # Adverbs are also lemmatized from lowercase
            lemma = hanta_tagger.analyze(word.lower(), casesensitive=True)[0]
            
    except Exception as e:
        log(f"HanTa analyze failed for {word}/{pos_group}: {e}. Falling back.")
        
    # Fallback logic
    if not lemma:
        if pos_group == "noun":
            return word.capitalize()
        return word.lower()
        
    return lemma

def _build_semantics(lemma: str, odenet_senses: List[Dict], top_n: int) -> Dict[str, Any]:
    """
    (Helper) Builds the semantics block with OdeNet and ConceptNet.
    """
    conceptnet_relations = []
    if REQUESTS_AVAILABLE:
        try:
            conceptnet_result = conceptnet_get_relations(lemma, language='de')
            conceptnet_relations = conceptnet_result.get("relations", [])
        except Exception as e:
            conceptnet_relations = [{"error": str(e)}]
    
    if top_n > 0:
        odenet_senses = odenet_senses[:top_n]
        conceptnet_relations.sort(key=lambda x: x.get('weight', 0.0), reverse=True)
        conceptnet_relations = conceptnet_relations[:top_n]
        
    return {
        "lemma": lemma,
        "odenet_senses": odenet_senses,
        "conceptnet_relations": conceptnet_relations
    }

# ============================================================================
# 6d. WIKTIONARY DATABASE LOGIC (PRIMARY ENGINE)
# ============================================================================

def wiktionary_download_db() -> bool:
    """
    Downloads the Wiktionary DB from Hugging Face Hub if it doesn't exist.
    """
    global WIKTIONARY_AVAILABLE
    if os.path.exists(WIKTIONARY_DB_PATH):
        print(f"✓ Wiktionary DB '{WIKTIONARY_DB_PATH}' already exists.")
        WIKTIONARY_AVAILABLE = True
        return True

    print(f"Wiktionary DB not found. Downloading from '{WIKTIONARY_REPO_ID}'...")
    try:
        hf_hub_download(
            repo_id=WIKTIONARY_REPO_ID,
            filename=WIKTIONARY_DB_PATH,
            repo_type="dataset",
            local_dir=".",
            local_dir_use_symlinks=False
        )
        print(f"✓ Wiktionary DB downloaded successfully.")
        WIKTIONARY_AVAILABLE = True
        return True
    except Exception as e:
        print(f"✗ CRITICAL: Failed to download Wiktionary DB: {e}")
        traceback.print_exc()
        return False

def wiktionary_get_connection() -> Optional[sqlite3.Connection]:
    """
    Thread-safe function to get a single, read-only SQLite connection.
    """
    global WIKTIONARY_CONN, WIKTIONARY_AVAILABLE
    if not WIKTIONARY_AVAILABLE:
        log("Wiktionary DB is not available, cannot create connection.")
        return None

    if WIKTIONARY_CONN:
        return WIKTIONARY_CONN

    with WIKTIONARY_CONN_LOCK:
        if WIKTIONARY_CONN:
            return WIKTIONARY_CONN
        
        if not os.path.exists(WIKTIONARY_DB_PATH):
            log("Wiktionary DB file missing, connection failed.")
            WIKTIONARY_AVAILABLE = False
            return None
            
        try:
            log("Creating new read-only connection to Wiktionary DB...")
            # URI mode for read-only connection
            db_uri = f"file:{WIKTIONARY_DB_PATH}?mode=ro"
            conn = sqlite3.connect(db_uri, uri=True, check_same_thread=False)
            conn.row_factory = sqlite3.Row # Makes results dict-like
            
            # Test query
            _ = conn.execute("SELECT name FROM sqlite_master WHERE type='table' LIMIT 1").fetchone()
            
            print("✓ Wiktionary DB connection successful.")
            WIKTIONARY_CONN = conn
            return WIKTIONARY_CONN
        except Exception as e:
            print(f"✗ CRITICAL: Failed to connect to Wiktionary DB: {e}")
            traceback.print_exc()
            WIKTIONARY_AVAILABLE = False
            return None

def _wiktionary_map_pos_key(wikt_pos: Optional[str]) -> str:
    """Maps Wiktionary POS tags to our internal keys."""
    if not wikt_pos:
        return "unknown"
    if wikt_pos == "noun": return "noun"
    if wikt_pos == "verb": return "verb"
    if wikt_pos == "adj": return "adjective"
    if wikt_pos == "adv": return "adverb"
    return wikt_pos # E.g., "phrase", "abbrev"

def _wiktionary_build_report_for_entry(entry_id: int, conn: sqlite3.Connection) -> Dict[str, Any]:
    """
    (REVISED FOR FULL DB V3)
    Fetches ALL associated data for a single Wiktionary entry_id.
    This version correctly queries expressions/proverbs by entry_id.
    """
    report = {}
    
    # 1. Get Base Entry Info
    entry_data = conn.execute(
        "SELECT word, title, redirect, pos, pos_title, lang, etymology_text FROM entries WHERE id = ?", (entry_id,)
    ).fetchone()
    if not entry_data:
        return {"error": "Entry ID not found"}
    report.update(dict(entry_data))
    report["entry_id"] = entry_id
    report["lemma"] = entry_data["word"]

    # 2. Get Senses (with Glosses, Tags, Topics, Categories, and Examples)
    senses_q = conn.execute(
        """
        SELECT 
            s.id as sense_id, 
            s.sense_index,
            (SELECT GROUP_CONCAT(g.gloss_text, '; ') FROM glosses g WHERE g.sense_id = s.id) as glosses,
            (SELECT GROUP_CONCAT(t.tag, ', ') FROM sense_tags st JOIN tags t ON st.tag_id = t.id WHERE st.sense_id = s.id) as tags,
            (SELECT GROUP_CONCAT(top.topic, ', ') FROM sense_topics stop JOIN topics top ON stop.topic_id = top.id WHERE stop.sense_id = s.id) as topics
        FROM senses s
        WHERE s.entry_id = ?
        ORDER BY s.id
        """, (entry_id,)
    ).fetchall()
    
    senses_list = []
    for sense_row in senses_q:
        sense_dict = dict(sense_row)
        sense_id = sense_dict["sense_id"]
        
        # Get examples (linked to sense_id)
        examples_q = conn.execute(
            "SELECT text, ref, author, title, year, url FROM examples WHERE sense_id = ?", (sense_id,)
        ).fetchall()
        sense_dict["examples"] = [dict(ex) for ex in examples_q]
        
        senses_list.append(sense_dict)
    report["senses"] = senses_list

    # 3. Get Inflected Forms (with Tags and Topics)
    forms_q = conn.execute(
        """
        SELECT 
            f.form_text, 
            f.sense_index,
            (SELECT GROUP_CONCAT(t.tag, ', ') FROM form_tags ft JOIN tags t ON ft.tag_id = t.id WHERE ft.form_id = f.id) as tags,
            (SELECT GROUP_CONCAT(top.topic, ', ') FROM form_topics ftop JOIN topics top ON ftop.topic_id = top.id WHERE ftop.form_id = f.id) as topics
        FROM forms f
        WHERE f.entry_id = ?
        GROUP BY f.id ORDER BY f.id
        """, (entry_id,)
    ).fetchall()
    report["forms"] = [dict(f) for f in forms_q]

    # 4. Get Pronunciations (with Tags)
    sounds_q = conn.execute(
        """
        SELECT 
            s.ipa, s.audio, s.mp3_url, s.ogg_url, s.rhymes,
            (SELECT GROUP_CONCAT(t.tag, ', ') FROM sound_tags st JOIN tags t ON st.tag_id = t.id WHERE st.sound_id = s.id) as tags
        FROM sounds s
        WHERE s.entry_id = ?
        GROUP BY s.id
        """, (entry_id,)
    ).fetchall()
    report["sounds"] = [dict(s) for s in sounds_q]

    # 5. Get Synonyms (with Tags and Topics)
    syn_q = conn.execute(
        """
        SELECT 
            s.synonym_word, s.sense_index,
            (SELECT GROUP_CONCAT(t.tag, ', ') FROM synonym_tags st JOIN tags t ON st.tag_id = t.id WHERE st.synonym_id = s.id) as tags,
            (SELECT GROUP_CONCAT(top.topic, ', ') FROM synonym_topics stop JOIN topics top ON stop.topic_id = top.id WHERE stop.synonym_id = s.id) as topics
        FROM synonyms s
        WHERE s.entry_id = ?
        GROUP BY s.id
        """, (entry_id,)
    ).fetchall()
    report["synonyms"] = [dict(s) for s in syn_q]

    # 6. Get Antonyms (with Tags)
    ant_q = conn.execute(
        """
        SELECT 
            a.antonym_word, a.sense_index,
            (SELECT GROUP_CONCAT(t.tag, ', ') FROM antonym_tags at JOIN tags t ON at.tag_id = t.id WHERE at.antonym_id = a.id) as tags
        FROM antonyms a
        WHERE a.entry_id = ?
        GROUP BY a.id
        """, (entry_id,)
    ).fetchall()
    report["antonyms"] = [dict(a) for a in ant_q]

    # 7. Get Translations (with Tags)
    trans_q = conn.execute(
        """
        SELECT 
            tr.lang, tr.lang_code, tr.word, tr.sense_text, tr.roman,
            (SELECT GROUP_CONCAT(t.tag, ', ') FROM translation_tags tt JOIN tags t ON tt.tag_id = t.id WHERE tt.translation_id = tr.id) as tags
        FROM translations tr
        WHERE tr.entry_id = ?
        GROUP BY tr.id
        """, (entry_id,)
    ).fetchall()
    report["translations"] = [dict(tr) for tr in trans_q]
    
    # 8. Get Hyphenations
    hyphen_q = conn.execute(
        "SELECT hyphenation FROM hyphenations WHERE entry_id = ?", (entry_id,)
    ).fetchall()
    report["hyphenations"] = [h["hyphenation"] for h in hyphen_q]
    
    # 9. Get Derived and Related Terms
    derived_q = conn.execute(
        "SELECT derived_word, sense_index FROM derived_terms WHERE entry_id = ?", (entry_id,)
    ).fetchall()
    report["derived_terms"] = [dict(d) for d in derived_q]
    
    related_q = conn.execute(
        "SELECT related_word, sense_index, raw_tags_json FROM related_terms WHERE entry_id = ?", (entry_id,)
    ).fetchall()
    report["related_terms"] = [dict(r) for r in related_q]
    
    # 10. Get Entry-level Tags and Categories
    entry_tags_q = conn.execute(
        "SELECT t.tag FROM entry_tags et JOIN tags t ON et.tag_id = t.id WHERE et.entry_id = ?", (entry_id,)
    ).fetchall()
    report["entry_tags"] = [t["tag"] for t in entry_tags_q]
    
    entry_cats_q = conn.execute(
        "SELECT c.category FROM entry_categories ec JOIN categories c ON ec.category_id = c.id WHERE ec.entry_id = ?", (entry_id,)
    ).fetchall()
    report["entry_categories"] = [c["category"] for c in entry_cats_q]

    # --- 11. GET ALL NEW OMITTED FIELDS (linked to entry_id) ---
    
    notes_q = conn.execute("SELECT note FROM entry_notes WHERE entry_id = ?", (entry_id,)).fetchall()
    report["entry_notes"] = [n["note"] for n in notes_q]
    
    other_pos_q = conn.execute("SELECT pos_value FROM other_pos WHERE entry_id = ?", (entry_id,)).fetchall()
    report["other_pos"] = [p["pos_value"] for p in other_pos_q]
    
    raw_tags_q = conn.execute("SELECT raw_tag FROM entry_raw_tags WHERE entry_id = ?", (entry_id,)).fetchall()
    report["raw_tags"] = [t["raw_tag"] for t in raw_tags_q]
    
    desc_q = conn.execute("SELECT lang, word, roman FROM descendants WHERE entry_id = ?", (entry_id,)).fetchall()
    report["descendants"] = [dict(d) for d in desc_q]
    
    hyper_q = conn.execute("SELECT hypernym_word, sense_index FROM hypernyms WHERE entry_id = ?", (entry_id,)).fetchall()
    report["hypernyms"] = [dict(h) for h in hyper_q]
    
    hypo_q = conn.execute("SELECT hyponym_word, sense_index FROM hyponyms WHERE entry_id = ?", (entry_id,)).fetchall()
    report["hyponyms"] = [dict(h) for h in hypo_q]
    
    holo_q = conn.execute("SELECT holonym_word, sense_index FROM holonyms WHERE entry_id = ?", (entry_id,)).fetchall()
    report["holonyms"] = [dict(h) for h in holo_q]
    
    mero_q = conn.execute("SELECT meronym_word, sense_index FROM meronyms WHERE entry_id = ?", (entry_id,)).fetchall()
    report["meronyms"] = [dict(m) for m in mero_q]
    
    coord_q = conn.execute(
        """
        SELECT 
            ct.id, ct.coordinate_word, ct.sense_index,
            (SELECT GROUP_CONCAT(t.tag, ', ') FROM coordinate_term_tags ctt JOIN tags t ON ctt.tag_id = t.id WHERE ctt.coordinate_term_id = ct.id) as tags
        FROM coordinate_terms ct
        WHERE ct.entry_id = ?
        GROUP BY ct.id
        """, (entry_id,)
    ).fetchall()
    report["coordinate_terms"] = [dict(c) for c in coord_q]

    # --- FIXED: Query expressions and proverbs by entry_id ---
    expr_q = conn.execute(
        "SELECT expression, sense_index FROM expressions WHERE entry_id = ?", (entry_id,)
    ).fetchall()
    report["expressions"] = [dict(ex) for ex in expr_q]
        
    prov_q = conn.execute(
        "SELECT proverb, sense_index FROM proverbs WHERE entry_id = ?", (entry_id,)
    ).fetchall()
    report["proverbs"] = [dict(p) for p in prov_q]

    return report

def _wiktionary_find_all_entries(word: str, conn: sqlite3.Connection) -> List[Dict[str, Any]]:
    """
    (FIXED V24)
    Finds all entries related to a word.
    1. Finds direct lemma matches (e.g., input "Vertrag" -> finds "Vertrag" entry)
    2. Finds inflection matches (e.g., input "Häuser" -> finds "Haus" entry via `forms` table)
    3. Finds declined form matches (e.g., input "Verträge" -> finds "Verträge" entry,
       then finds "Vertrag" entry via `senses.form_of` table)
    Returns a list of full entry reports.
    """
    log(f"Wiktionary: Querying for '{word}'...")
    found_entry_ids: Set[int] = set()
    
    # 1. Check if the word is a lemma (base form)
    lemma_q = conn.execute(
        "SELECT id, pos_title FROM entries WHERE word = ? AND lang = 'Deutsch'", (word,)
    ).fetchall()
    
    parent_lemmas_to_find: Set[str] = set()
    
    for row in lemma_q:
        entry_id = row["id"]
        pos_title = row["pos_title"]
        found_entry_ids.add(entry_id)
        
        # --- THIS IS THE NEW LOGIC (STEP 3) ---
        if pos_title in ("Deklinierte Form", "Konjugierte Form", "Komparativ", "Superlativ"):
            log(f"Wiktionary: Word '{word}' is an inflected entry (ID {entry_id}). Looking for its parent lemma...")
            form_of_q = conn.execute(
                "SELECT form_of FROM senses WHERE entry_id = ?", (entry_id,)
            ).fetchall()
            
            for form_row in form_of_q:
                form_of_json = form_row["form_of"]
                if not form_of_json:
                    continue
                try:
                    # Parse the JSON string (e.g., '[{"word": "Vertrag"}]')
                    form_of_data = json.loads(form_of_json)
                    if isinstance(form_of_data, list) and form_of_data:
                        parent_lemma_word = form_of_data[0].get("word")
                        if parent_lemma_word:
                            parent_lemmas_to_find.add(parent_lemma_word)
                except json.JSONDecodeError:
                    log(f"Wiktionary: Failed to parse form_of JSON: {form_of_json}")
        # --- END OF NEW LOGIC ---

    # 2. Check if the word is an inflected form (in the `forms` table)
    form_q = conn.execute(
        """
        SELECT DISTINCT e.id
        FROM forms f
        JOIN entries e ON f.entry_id = e.id
        WHERE f.form_text = ? AND e.lang = 'Deutsch'
          AND f.id NOT IN (
            -- Exclude all form_ids that are tagged as 'variant' or 'auxiliary'
            SELECT ft.form_id
            FROM form_tags ft
            JOIN tags t ON ft.tag_id = t.id
            WHERE t.tag IN ('variant', 'auxiliary')
          )
        """, (word,)
    ).fetchall()
    for row in form_q:
        found_entry_ids.add(row["id"])
        
    # --- NEW: Add parent lemmas found in step 3 ---
    if parent_lemmas_to_find:
        log(f"Wiktionary: Found parent lemmas to add: {parent_lemmas_to_find}")
        for lemma_word in parent_lemmas_to_find:
            parent_id_q = conn.execute(
                "SELECT id FROM entries WHERE word = ? AND lang = 'Deutsch'", (lemma_word,)
            ).fetchall()
            for row in parent_id_q:
                found_entry_ids.add(row["id"])
    
    log(f"Wiktionary: Found {len(found_entry_ids)} unique matching entries.")
    
    # 3. Build a full report for each unique entry
    all_reports = []
    for entry_id in found_entry_ids:
        try:
            report = _wiktionary_build_report_for_entry(entry_id, conn)
            all_reports.append(report)
        except Exception as e:
            log(f"Wiktionary: Failed to build report for entry {entry_id}: {e}")
            
    return all_reports

def _wiktionary_format_semantics_block(
    wikt_report: Dict[str, Any], 
    pattern_block: Dict[str, Any], 
    top_n: int
) -> Dict[str, Any]:
    """
    (FIXED V24)
    Combines Wiktionary senses with OdeNet/ConceptNet senses,
    using the *correct* lemma.
    
    Priority:
    1. Wiktionary's lemma (from `wikt_report`)
    2. Pattern.de's lemma (from `pattern_block`)
    """
    
    pos_key = _wiktionary_map_pos_key(wikt_report.get("pos"))
    
    # --- THIS IS THE FIX ---
    # Prioritize Wiktionary's lemma first, as it's more reliable.
    semantic_lemma = wikt_report.get("lemma")
    
    # If Wiktionary's lemma is missing or bad, try pattern.de's
    if not semantic_lemma:
        if pos_key == "verb":
            semantic_lemma = pattern_block.get("infinitive")
        elif pos_key == "noun":
            semantic_lemma = pattern_block.get("base_form")
        elif pos_key == "adjective":
            semantic_lemma = pattern_block.get("predicative")
    
    # Final fallback
    if not semantic_lemma:
         semantic_lemma = wikt_report.get("word", "") # Use the original word as last resort
    
    log(f"[DEBUG] Wiktionary Semantics: Building block for lemma='{semantic_lemma}', pos='{pos_key}'")
    # --- END OF FIX ---

    # 1. Get Wiktionary senses (from the original report)
    wiktionary_senses = []
    for sense in wikt_report.get("senses", []):
        wiktionary_senses.append({
            "definition": sense.get("glosses"), # <-- Corrected from gloss_text
            "source": "wiktionary"
        })

    # 2. Get OdeNet senses for the *semantic_lemma*
    odenet_senses = []
    if WN_AVAILABLE:
        try:
            senses_by_pos = _get_odenet_senses_by_pos(semantic_lemma)
            odenet_senses_raw = senses_by_pos.get(pos_key, [])
            
            # Filter out placeholder
            if odenet_senses_raw and "info" not in odenet_senses_raw[0]:
                odenet_senses = odenet_senses_raw
        except Exception as e:
            log(f"[DEBUG] OdeNet lookup failed for {semantic_lemma} ({pos_key}): {e}")

    # 3. Get ConceptNet relations for the *semantic_lemma*
    conceptnet_relations = []
    if REQUESTS_AVAILABLE:
        try:
            conceptnet_result = conceptnet_get_relations(semantic_lemma, language='de')
            conceptnet_relations = conceptnet_result.get("relations", [])
        except Exception as e:
            conceptnet_relations = [{"error": str(e)}]
            
    # 4. Apply top_n limit
    if top_n > 0:
        wiktionary_senses = wiktionary_senses[:top_n]
        odenet_senses = odenet_senses[:top_n]
        conceptnet_relations.sort(key=lambda x: x.get('weight', 0.0), reverse=True)
        conceptnet_relations = conceptnet_relations[:top_n]

    return {
        "lemma": semantic_lemma, # Return the *correct* lemma for this path
        "wiktionary_senses": wiktionary_senses,
        "odenet_senses": odenet_senses,
        "conceptnet_relations": conceptnet_relations,
        "wiktionary_synonyms": wikt_report.get("synonyms", []),
        "wiktionary_antonyms": wikt_report.get("antonyms", []),
        "wiktionary_translations": wikt_report.get("translations", []),
        "wiktionary_derived_terms": wikt_report.get("derived_terms", []),
        "wiktionary_related_terms": wikt_report.get("related_terms", [])
    }

def _analyze_word_with_wiktionary(word: str, top_n: int) -> Dict[str, Any]:
    """
    (PRIMARY ENGINE) Analyzes a word using the Wiktionary DB as Ground Truth,
    filling in missing gaps with Pattern.de generation.
    """
    print(f"\n[Wiktionary Engine] Starting analysis for: {word}")
    
    final_result: Dict[str, Any] = {
        "input_word": word,
        "analysis": {}
    }
    
    conn = wiktionary_get_connection()
    if not conn:
        log("[Wiktionary Engine] No DB connection available.")
        return {} 
        
    # --- 1. GET SPACY/IWNLP HINT FOR PRIORITIZATION ---
    spacy_pos_hint = None
    spacy_lemma_hint = None
    if IWNLP_AVAILABLE:
        try:
            iwnlp = iwnlp_get_pipeline()
            if iwnlp:
                doc = iwnlp(word)
                token = doc[0]
                # Map spaCy POS to our internal keys
                spacy_pos_raw = token.pos_.lower()
                if spacy_pos_raw == "adj": spacy_pos_hint = "adjective"
                elif spacy_pos_raw == "adv": spacy_pos_hint = "adverb"
                elif spacy_pos_raw == "verb": spacy_pos_hint = "verb"
                elif spacy_pos_raw == "noun": spacy_pos_hint = "noun"
                else: spacy_pos_hint = spacy_pos_raw
                    
                spacy_lemma_hint = token.lemma_
                log(f"[DEBUG] Priority Hint: spaCy POS='{spacy_pos_hint}', Lemma='{spacy_lemma_hint}'")
        except Exception as e:
            log(f"[DEBUG] Priority Hint failed: {e}")

    # --- 2. FIND ALL WIKTIONARY ENTRIES ---
    try:
        wiktionary_reports = _wiktionary_find_all_entries(word, conn)
    except Exception as e:
        log(f"[DEBUG] Wiktionary query failed: {e}")
        return {} 

    if not wiktionary_reports:
        log(f"[DEBUG] No Wiktionary entries found for '{word}'.")
        return {} 

    # --- 3. PRIORITIZE/SORT THE WIKTIONARY ENTRIES ---
    def get_priority_score(report):
        wikt_pos = _wiktionary_map_pos_key(report.get("pos"))
        wikt_lemma = report.get("lemma")
        # Priority 1: Exact POS match with spaCy hint
        if spacy_pos_hint and wikt_pos == spacy_pos_hint:
            if spacy_lemma_hint and wikt_lemma == spacy_lemma_hint: return 1
            return 2
        # Priority 2: Input word is the lemma
        if wikt_lemma and wikt_lemma.lower() == word.lower(): return 3
        return 4

    wiktionary_reports.sort(key=get_priority_score)
    log(f"[DEBUG] Sorted {len(wiktionary_reports)} entries: {[r.get('lemma') + ' (' + r.get('pos') + ')' for r in wiktionary_reports]}")

    # --- 4. PROCESS ENTRIES (HYBRID STRATEGY) ---
    word_lower = word.lower()
    
    for wikt_report in wiktionary_reports:
        pos_key = _wiktionary_map_pos_key(wikt_report.get("pos"))
        lemma = wikt_report.get("lemma", word)
        pos_title = wikt_report.get("pos_title", "")
        
        log(f"\n--- Processing Entry: {lemma} ({pos_key}) ---")

        # --- A. Raw Wiktionary Forms (Ground Truth) ---
        wikt_forms_list = wikt_report.get("forms", [])
        inflections_wikt_block = {
            "base_form": lemma,
            "forms_list": wikt_forms_list,
            "source": "wiktionary"
        }
        
        # --- B. Generate Base Pattern Template (The Scaffold) ---
        # We ALWAYS generate this if Pattern is available, to provide the table structure.
        pattern_block = {}
        
        if PATTERN_DE_AVAILABLE:
            try:
                log(f"[DEBUG] Generating Pattern.de base template for '{lemma}' ({pos_key})...")
                if pos_key == "noun" or "Substantiv" in pos_title:
                    # Gender-Aware Generation
                    wikt_tags = wikt_report.get("entry_tags", [])
                    forced_gender = _map_wikt_gender_to_pattern(wikt_tags)
                    if forced_gender:
                        log(f"[DEBUG] Context: Forcing Pattern gender to {forced_gender} based on Wiktionary tags.")
                    else:
                        log(f"[DEBUG] Context: No gender tags in Wiktionary. Letting Pattern auto-detect.")
                        
                    pattern_block = pattern_analyze_as_noun(lemma, fixed_gender=forced_gender)
                    
                elif pos_key == "verb" or "Verb" in pos_title or "Konjugierte Form" in pos_title:
                    use_word = word if "Konjugierte Form" in pos_title else lemma
                    pattern_block = pattern_analyze_as_verb(use_word)
                        
                elif pos_key == "adjective" or "Adjektiv" in pos_title or "Deklinierte Form" in pos_title:
                    use_word = word if "Deklinierte Form" in pos_title else lemma
                    pattern_block = pattern_analyze_as_adjective(use_word)
                         
                elif pos_key == "adverb":
                    pattern_block = {"base_form": lemma, "info": "Adverbs are non-inflecting."}
            except Exception as e:
                log(f"[ERROR] Pattern.de generation failed: {e}")
                pattern_block = {"error": f"Pattern.de failed: {e}"}

        # --- C. THE HYBRID MERGE: Overwrite Pattern data with Wiktionary Truth ---
        # logic: If Wiktionary has a form for a specific slot, use it. 
        # If not, keep the Pattern generated form (thereby filling the gap).
        
        if pattern_block and "error" not in pattern_block and wikt_forms_list:
            log(f"[DEBUG] Starting Hybrid Merge (Wiktionary forms: {len(wikt_forms_list)})...")
            
            overwrites_count = 0
            
            for wikt_form in wikt_forms_list:
                text = wikt_form.get("form_text")
                tags = wikt_form.get("tags")
                if not text or not tags: continue
                
                # Map Wikt tags to the address inside pattern_block
                path_keys = _map_wikt_form_to_pattern_keys(pos_key, tags)
                
                if path_keys:
                    # Navigate to the slot in pattern_block
                    target = pattern_block
                    
                    # Special handling for Noun structure (declension_by_gender)
                    if pos_key == "noun" and "declension_by_gender" in pattern_block:
                        # We apply the overwrite to ALL genders present in the pattern block
                        # (Usually only 1 if we forced it, but maybe more if ambiguous)
                        for gender_key in pattern_block["declension_by_gender"]:
                            # path_keys[0] is e.g. "Nominativ Singular"
                            slot_key = path_keys[0]
                            target_dict = pattern_block["declension_by_gender"][gender_key]
                            
                            if slot_key in target_dict:
                                # Noun slots have subkeys: 'bare', 'definite', 'indefinite'
                                # Wiktionary usually gives the form with article "der See" or without "Seen"
                                # We try to be smart about updating 'bare' vs 'definite'
                                
                                current_bare = target_dict[slot_key].get('bare', '')
                                
                                # Simple clean: remove articles to get bare
                                clean_text = re.sub(r"^(der|die|das|den|dem|des|ein|eine|einen|einem|einer|eines)\s+", "", text, flags=re.IGNORECASE).strip()
                                
                                if clean_text != current_bare:
                                    log(f"[DEBUG] Merge: Overwriting {gender_key} -> {slot_key} | Old: '{current_bare}' -> New: '{clean_text}' (Source: Wiktionary)")
                                    target_dict[slot_key]['bare'] = clean_text
                                    # Also update full forms if possible
                                    if "definite" in target_dict[slot_key]:
                                        # We can reconstruct definite if we know the article, but let's just trust the bare text update
                                        # because the HTML renderer often rebuilds the article. 
                                        # However, let's update 'definite' if the wikt text looks like it has an article
                                        if " " in text:
                                             target_dict[slot_key]['definite'] = text
                                    overwrites_count += 1

                    # Handling for Verbs/Adjectives (Nested Dicts)
                    else:
                        # Navigate deep
                        valid_path = True
                        for key in path_keys[:-1]:
                            if key in target:
                                target = target[key]
                            else:
                                valid_path = False
                                break
                        
                        if valid_path:
                            last_key = path_keys[-1]
                            if last_key in target and target[last_key] != text:
                                log(f"[DEBUG] Merge: Overwriting {path_keys} | Old: '{target[last_key]}' -> New: '{text}' (Source: Wiktionary)")
                                target[last_key] = text
                                overwrites_count += 1
                                
            log(f"[DEBUG] Merge complete. {overwrites_count} slots updated with Ground Truth.")
            # Mark the block as hybrid so UI can verify validity
            pattern_block["is_hybrid"] = True

        # --- D. Build Semantics Block ---
        # Use lemma from Wiktionary (Ground Truth)
        semantics_lemma = lemma
        semantics_block = _wiktionary_format_semantics_block(wikt_report, pattern_block, top_n)
        
        # --- E. Assemble Final Report ---
        pos_entry_report = {
            "inflections_wiktionary": inflections_wikt_block,
            "inflections_pattern": pattern_block, # This is now the Hybrid Block
            "semantics_combined": semantics_block,
            "wiktionary_metadata": {
                "pos_title": pos_title,
                "etymology": wikt_report.get("etymology_text"),
                "pronunciation": wikt_report.get("sounds"),
                "hyphenation": wikt_report.get("hyphenations"),
                "examples": [ex for s in wikt_report.get("senses", []) for ex in s.get("examples", [])],
                "entry_tags": wikt_report.get("entry_tags"),
                "entry_categories": wikt_report.get("entry_categories"),
                # New fields
                "entry_notes": wikt_report.get("entry_notes"),
                "other_pos": wikt_report.get("other_pos"),
                "raw_tags": wikt_report.get("raw_tags"),
                "descendants": wikt_report.get("descendants"),
                "hypernyms": wikt_report.get("hypernyms"),
                "hyponyms": wikt_report.get("hyponyms"),
                "holonyms": wikt_report.get("holonyms"),
                "meronyms": wikt_report.get("meronyms"),
                "coordinate_terms": wikt_report.get("coordinate_terms"),
                "expressions": wikt_report.get("expressions"),
                "proverbs": wikt_report.get("proverbs")
            }
        }
        
        # --- F. Validation Filter ---
        is_valid = False
        is_inflected_entry = "Konjugierte Form" in pos_title or "Deklinierte Form" in pos_title

        # Check 1: Lemma Match
        if lemma.lower() == word_lower:
            is_valid = True
            log(f"[DEBUG] Validate: Accepted '{lemma}' (Lemma Match)")
        
        # Check 2: Form Match
        if not is_valid and not is_inflected_entry:
            # Look in Ground Truth (Wiktionary)
            for form_entry in wikt_forms_list:
                form_text = form_entry.get("form_text", "")
                clean_form = re.sub(r"\(.*\)", "", form_text).strip() # Remove parens
                clean_form = re.sub(r"^(der|die|das|ein|eine|...)\s+", "", clean_form, flags=re.IGNORECASE).strip() # Remove articles
                if word_lower in clean_form.lower():
                    is_valid = True
                    log(f"[DEBUG] Validate: Accepted '{lemma}' (Found in Wiktionary forms)")
                    break
            
            # Look in Pattern Generation (if Wikt failed)
            if not is_valid and pattern_block:
                 if word_appears_in_inflections(word, pattern_block, pos_key):
                     is_valid = True
                     log(f"[DEBUG] Validate: Accepted '{lemma}' (Found in Pattern forms)")

        if is_valid:
            if pos_key not in final_result["analysis"]:
                final_result["analysis"][pos_key] = []
            final_result["analysis"][pos_key].append(pos_entry_report)
        else:
             log(f"[DEBUG] Validate: Dropped '{lemma}' ({pos_key}) - No match found.")

    final_result["info"] = f"Analysis from Wiktionary (Hybrid Engine). Found {len(wiktionary_reports)} entries."
    return final_result

# ============================================================================
# 6e. SHARED SEMANTIC HELPER 
# ============================================================================

def _build_semantics_block_for_lemma(lemma: str, pos_key: str, top_n: int) -> Dict[str, Any]:
    """
    (REUSABLE HELPER)
    Fetches OdeNet and ConceptNet data for a given lemma and POS.
    """
    log(f"[DEBUG] Building semantics for lemma='{lemma}', pos='{pos_key}'")
    
    # 1. Get OdeNet senses for this lemma + POS
    odenet_senses = []
    if WN_AVAILABLE:
        try:
            senses_by_pos = _get_odenet_senses_by_pos(lemma)
            odenet_senses_raw = senses_by_pos.get(pos_key, [])
            
            # Filter out placeholder
            if odenet_senses_raw and "info" not in odenet_senses_raw[0]:
                odenet_senses = odenet_senses_raw
        except Exception as e:
            log(f"[DEBUG] OdeNet lookup failed for {lemma} ({pos_key}): {e}")

    # 2. Get ConceptNet relations for this lemma
    conceptnet_relations = []
    if REQUESTS_AVAILABLE:
        try:
            conceptnet_result = conceptnet_get_relations(lemma, language='de')
            conceptnet_relations = conceptnet_result.get("relations", [])
        except Exception as e:
            conceptnet_relations = [{"error": str(e)}]
            
    # 3. Apply top_n limit
    if top_n > 0:
        odenet_senses = odenet_senses[:top_n]
        conceptnet_relations.sort(key=lambda x: x.get('weight', 0.0), reverse=True)
        conceptnet_relations = conceptnet_relations[:top_n]

    return {
        "lemma": lemma,
        "wiktionary_senses": [], # This block is for non-Wiktionary engines
        "odenet_senses": odenet_senses,
        "conceptnet_relations": conceptnet_relations,
        "wiktionary_synonyms": [],
        "wiktionary_antonyms": []
    }

# ============================================================================
# 6f. DWDSMOR ENGINE (NEW FALLBACK 1)
# ============================================================================

def dwdsmor_get_lemmatizer() -> Optional[Any]: # Return type is 'sfst.Transducer'
    """
    Thread-safe function to get a single instance of the DWDSmor analyzer.
    It will automatically download/cache the 'open' automata from Hugging Face Hub.
    """
    global DWDSMOR_LEMMATIZER
    if not DWDSMOR_AVAILABLE:
        raise ImportError("dwdsmor library is not installed.")

    if DWDSMOR_LEMMATIZER:
        return DWDSMOR_LEMMATIZER

    with DWDSMOR_LEMMATIZER_LOCK:
        if DWDSMOR_LEMMATIZER:
            return DWDSMOR_LEMMATIZER
        
        try:
            print("Initializing DWDSmor lemmatizer (loading automata)...")
            
            # --- THIS IS THE FIX ---
            # Use the correct API from dwdsmor's own tools (analysis.py)
            # This will find and download the HF repo automatically
            from dwdsmor import automaton
            automata = automaton.automata() 
            analyzer = automata.analyzer("lemma") # Use the 'lemma' automaton
            # --- END OF FIX ---

            # Force the traversal to actually run by converting to a list.
            print("[DEBUG] DWDSmor: Running warm-up call...")
            _ = list(analyzer.analyze("Test", join_tags=True)) 

            print("✓ DWDSmor lemmatizer initialized successfully.")
            DWDSMOR_LEMMATIZER = analyzer
            return DWDSMOR_LEMMATIZER
        except Exception as e:
            print(f"✗ CRITICAL: Failed to initialize DWDSmor: {e}")
            traceback.print_exc()
            return None

def _dwdsmor_map_pos_key(dwdsmor_pos: str) -> str:
    """Maps DWDSmor POS tags to our internal keys."""
    if dwdsmor_pos == "V": return "verb"
    if dwdsmor_pos == "NN": return "noun"
    if dwdsmor_pos == "NPROP": return "noun" # Proper Noun
    if dwdsmor_pos == "ADJ": return "adjective"
    if dwdsmor_pos == "ADV": return "adverb"
    return dwdsmor_pos.lower() # Fallback for others

def _analyze_word_with_dwdsmor(word: str, top_n: int) -> Dict[str, Any]:
    """
    (FALLBACK ENGINE 1) Analyzes a single word using DWDSmor + Pattern + Semantics.
    Returns {} on failure.
    """
    if not DWDSMOR_AVAILABLE:
        return {} # Signal failure
        
    print(f"\n[Word Encyclopedia] Running V21 (DWDSmor) engine for: \"{word}\"")
    final_result: Dict[str, Any] = {
        "input_word": word,
        "analysis": {}
    }
    
    try:
        analyzer = dwdsmor_get_lemmatizer()
        if not analyzer:
            raise Exception("DWDSmor lemmatizer failed to initialize.")

        analyses = list(analyzer.analyze(word, join_tags=True))
        
        if not analyses:
            return {} # No results
        
        log(f"[DEBUG] DWDSmor: Found {len(analyses)} potential analyses.")
        
        processed_lemmas_pos: Set[Tuple[str, str]] = set()

        for analysis in analyses:
            
            # --- THIS IS THE FIX ---
            # The 'Traversal' object from analyzer.analyze() uses:
            # .analysis -> for the lemma string (e.g., "Haus")
            # .pos -> for the POS tag (e.g., "NN")
            # .spec -> for the full analysis string
            if not analysis.analysis or not analysis.pos:
                continue

            lemma = analysis.analysis  # Use .analysis, not .lemma
            pos_key = _dwdsmor_map_pos_key(analysis.pos)
            # --- END OF FIX ---
            
            if (lemma, pos_key) in processed_lemmas_pos:
                continue
            processed_lemmas_pos.add((lemma, pos_key))

            log(f"--- Analyzing DWDSmor path: lemma='{lemma}', pos='{pos_key}' ---")

            # --- 1. Get Inflections (Pattern) ---
            pattern_block = {}
            if PATTERN_DE_AVAILABLE:
                try:
                    if pos_key == "noun":
                        pattern_block = pattern_analyze_as_noun(lemma)
                    elif pos_key == "verb":
                        pattern_block = pattern_analyze_as_verb(lemma)
                    elif pos_key == "adjective":
                        pattern_block = pattern_analyze_as_adjective(lemma)
                    elif pos_key == "adverb":
                        pattern_block = {"base_form": lemma, "info": "Adverbs are non-inflecting."}
                except Exception as e:
                    pattern_block = {"error": f"Pattern.de analysis for {pos_key}('{lemma}') failed: {e}"}

            # --- 2. Build Semantics Block ---
            semantics_block = _build_semantics_block_for_lemma(lemma, pos_key, top_n)

            # --- 3. Build Final Report Block ---
            pos_entry_report = {
                "dwdsmor_analysis": {
                    "lemma": lemma,
                    "pos": analysis.pos,
                    "analysis_string": analysis.spec, # .spec is the full string
                    "source": "dwdsmor"
                },
                "inflections_pattern": pattern_block,
                "semantics_combined": semantics_block
            }
            
            if pos_key not in final_result["analysis"]:
                final_result["analysis"][pos_key] = []
            final_result["analysis"][pos_key].append(pos_entry_report)

        if not final_result["analysis"]:
            return {} # No valid paths found
            
        final_result["info"] = "Analysis performed by DWDSmor-led engine."
        return final_result

    except Exception as e:
        print(f"[Word Encyclopedia] DWDSmor Engine FAILED: {e}")
        traceback.print_exc()
        return {} # Signal failure
    
# ============================================================================
# 7. CONSOLIDATED ANALYZER LOGIC
# ============================================================================

# --- 7a. Comprehensive (Contextual) Analyzer ---

def comprehensive_german_analysis(text: str, top_n_value: Optional[float] = 0) -> Dict[str, Any]:
    """
    (CONTEXTUAL) Combines NLP tools for a deep analysis of German text.
    
    Reads the list-based, multi-engine output 
    from `analyze_word_encyclopedia` and combines all senses for ranking.
    """
    
    try:
        if not text or not text.strip():
            return {"info": "Please enter text to analyze."}
        top_n = int(top_n_value) if top_n_value is not None else 0
        
        print(f"\n[Comprehensive Analysis] Starting analysis for: \"{text}\" (top_n={top_n})")
        results: Dict[str, Any] = {"input_text": text}
        nlp_de = None
        context_doc = None

        # --- 1. LanguageTool Grammar Check ---
        print("[Comprehensive Analysis] Running LanguageTool...")
        if LT_AVAILABLE:
            try:
                results["grammar_check"] = lt_check_grammar(text)
            except Exception as e:
                results["grammar_check"] = {"error": f"LanguageTool failed: {e}"}
        else:
            results["grammar_check"] = {"error": "LanguageTool not available."}

        # --- 2. spaCy Morpho-Syntactic Backbone ---
        print("[Comprehensive Analysis] Running spaCy...")
        spacy_json_output = []
        try:
            _, spacy_json, _, _, _ = spacy_get_analysis("en", "de", text)
            if isinstance(spacy_json, list):
                spacy_json_output = spacy_json
                results["spacy_analysis"] = spacy_json_output
                nlp_de = SPACY_MODELS.get("de")
                if nlp_de:
                    context_doc = nlp_de(text)
                    if not context_doc.has_vector or context_doc.vector_norm == 0:
                        print("[Comprehensive Analysis] WARNING: Context sentence has no vector.")
                        context_doc = None
            else:
                 results["spacy_analysis"] = spacy_json
        except Exception as e:
            results["spacy_analysis"] = {"error": f"spaCy analysis failed: {e}"}

        # --- 2b. Heuristic SVA check ---
        try:
            if isinstance(results.get("grammar_check"), list) and any(d.get("status") == "perfect" for d in results["grammar_check"]):
                subj_num = None
                verb_num = None
                verb_token = None
                subj_token = None
                for tok in spacy_json_output:
                    if tok.get("dependency") in {"sb", "nsubj"}:
                        m = tok.get("morphology","")
                        if "Number=Sing" in m:
                            subj_num = "Sing"
                            subj_token = tok
                    spacy_pos_up = (tok.get("pos") or "").upper()
                    if (spacy_pos_up in {"VERB", "AUX"}) and ("VerbForm=Fin" in tok.get("morphology","")):
                        verb_token = tok
                        m = tok.get("morphology","")
                        if "Number=Plur" in m:
                            verb_num = "Plur"
                if subj_num == "Sing" and verb_num == "Plur":
                    corrected_sentence_sg = None
                    corrected_sentence_pl = None
                    replacements = []
                    v_lemma = verb_token.get("lemma") if verb_token else None
                    v_word = verb_token.get("word") if verb_token else None
                    v_3sg = _conjugate_to_person_number(v_lemma, "3", "sg") if v_lemma else None
                    if v_3sg and v_word:
                        corrected_sentence_sg = text.replace(v_word, v_3sg, 1)
                        replacements.append(corrected_sentence_sg)
                    subj_word = subj_token.get("word") if subj_token else None
                    subj_pl = None
                    if subj_word and PATTERN_DE_AVAILABLE:
                        try: subj_pl = pluralize(subj_word)
                        except Exception: subj_pl = None
                    if subj_word and subj_pl and subj_pl != subj_word:
                        corrected_sentence_pl = text.replace(subj_word, subj_pl, 1)
                        replacements.append(corrected_sentence_pl)
                    sva = {
                        "message": "Möglicher Kongruenzfehler: Singular-Subjekt mit pluralischer Verbform.",
                        "rule_id": "HEURISTIC_SUBJ_VERB_AGREEMENT",
                        "category": "Grammar",
                        "incorrect_text": f"{verb_token.get('word')}" if verb_token else "",
                        "replacements": replacements, "offset": None, "length": None,
                        "context": None, "short_message": "Subjekt–Verb-Kongruenz"
                    }
                    results["grammar_check"] = [sva]
        except Exception as e:
            print(f"SVA Heuristic failed: {e}")
            pass

        # --- 3. Lemma-by-Lemma Deep Dive (V19 LOGIC) ---
        print("[Comprehensive Analysis] Running Lemma Deep Dive...")
        FUNCTION_POS = {"DET","ADP","AUX","PUNCT","SCONJ","CCONJ","PART","PRON","NUM","SYM","X", "SPACE"}
        lemma_deep_dive: Dict[str, Any] = {}
        processed_lemmas: Set[str] = set()

        if not spacy_json_output:
            print("[Comprehensive Analysis] No spaCy tokens to analyze. Skipping deep dive.")
        else:
            for token in spacy_json_output:
                lemma = token.get("lemma")
                pos = (token.get("pos") or "").upper()
                
                if not lemma or lemma == "--" or pos in FUNCTION_POS or lemma in processed_lemmas:
                    continue
                processed_lemmas.add(lemma)
                print(f"[Deep Dive] Analyzing lemma: '{lemma}' (from token '{token.get('word')}')")
                
                # --- 3a. Get Validated Grammatical & Semantic Analysis ---
                # We call our new, multi-engine dispatcher.
                lemma_report: Dict[str, Any] = {}
                inflection_analysis = {}
                semantic_analysis = {}
                
                try:
                    # We pass top_n=0 to get ALL semantic possibilities for ranking
                    encyclopedia_data = analyze_word_encyclopedia(lemma, 0)
                    
                    # The "analysis" key contains {"noun": [ ... ], "verb": [ ... ], ...}
                    word_analysis = encyclopedia_data.get("analysis", {})
                    
                    # *** THIS IS THE KEY CHANGE ***
                    # Iterate over the POS keys and the *list* of entries for each
                    for pos_key, entry_list in word_analysis.items():
                        if not entry_list:
                            continue
                        
                        # For context, we only rank the *first* (most likely) entry
                        # provided by the encyclopedia for that POS.
                        data = entry_list[0] 
                        
                        # Store all inflection blocks AND metadata
                        inflection_analysis[f"{pos_key}_wiktionary"] = data.get("inflections_wiktionary")
                        inflection_analysis[f"{pos_key}_pattern"] = data.get("inflections_pattern")

                        # --- Capture Metadata ---
                        inflection_analysis[f"{pos_key}_metadata"] = data.get("wiktionary_metadata")
                        
                        # --- Combine ALL senses (Wiktionary, OdeNet) for ranking ---
                        all_senses_for_pos = []
                        semantics_block = data.get("semantics_combined", {})
                        
                        # Add Wiktionary senses
                        wikt_senses = semantics_block.get("wiktionary_senses", [])
                        for s in wikt_senses:
                            s["source"] = "wiktionary"
                            all_senses_for_pos.append(s)

                        # Add OdeNet senses
                        odenet_senses = semantics_block.get("odenet_senses", [])
                        for s in odenet_senses:
                            s["source"] = "odenet"
                            all_senses_for_pos.append(s)

                        semantic_analysis[f"{pos_key}_senses"] = all_senses_for_pos
                        
                        # Add ConceptNet relations (store separately, as they are not "senses")
                        if "conceptnet_relations" not in semantic_analysis:
                            semantic_analysis["conceptnet_relations"] = []
                        semantic_analysis["conceptnet_relations"].extend(
                            semantics_block.get("conceptnet_relations", [])
                        )

                    lemma_report["inflection_analysis"] = inflection_analysis
                    
                except Exception as e:
                    lemma_report["inflection_analysis"] = {"error": f"V19 Analyzer failed: {e}", "traceback": traceback.format_exc()}


                # --- 3b. Contextual Re-ranking (Unchanged) ---
                # re-rank the semantic data we gathered in step 3a.
                
                # OdeNet Senses (now combined with Wiktionary senses)
                for key in semantic_analysis:
                    if key.endswith("_senses") and nlp_de:
                        ranked_senses = []
                        for sense in semantic_analysis[key]:
                            if "error" in sense: continue
                            definition = sense.get("definition", "")
                            relevance = 0.0
                            if definition and context_doc:
                                try:
                                    def_doc = nlp_de(definition)
                                    if def_doc.has_vector and def_doc.vector_norm > 0:
                                        relevance = context_doc.similarity(def_doc)
                                except Exception:
                                    relevance = 0.0
                            sense["relevance_score"] = float(relevance)
                            ranked_senses.append(sense)
                        
                        ranked_senses.sort(key=lambda x: x.get('relevance_score', 0.0), reverse=True)
                        if top_n > 0:
                            ranked_senses = ranked_senses[:top_n]
                        semantic_analysis[key] = ranked_senses

                # ConceptNet Relations
                if "conceptnet_relations" in semantic_analysis and nlp_de:
                    ranked_relations = []
                    for rel in semantic_analysis["conceptnet_relations"]:
                        if "error" in rel: continue
                        text_to_score = rel.get('surface') or rel.get('other_node', '')
                        relevance = 0.0
                        if text_to_score and context_doc:
                             try:
                                 rel_doc = nlp_de(text_to_score)
                                 if rel_doc.has_vector and rel_doc.vector_norm > 0:
                                     relevance = context_doc.similarity(rel_doc)
                             except Exception:
                                 relevance = 0.0
                        rel["relevance_score"] = float(relevance)
                        ranked_relations.append(rel)
                    
                    ranked_relations.sort(key=lambda x: x.get('relevance_score', 0.0), reverse=True)
                    if top_n > 0:
                        ranked_relations = ranked_relations[:top_n]
                    semantic_analysis["conceptnet_relations"] = ranked_relations

                lemma_report["semantic_analysis"] = semantic_analysis
                lemma_deep_dive[lemma] = lemma_report
                
        results["lemma_deep_dive"] = lemma_deep_dive
        print("[Comprehensive Analysis] Analysis complete.")
        return results
    
    except Exception as e:
        print(f"[Comprehensive Analysis] FATAL ERROR: {e}")
        traceback.print_exc()
        return {
            "error": f"Analysis failed: {str(e)}",
            "traceback": traceback.format_exc(),
            "input_text": text
        }

# --- 7b. NEW: Word Encyclopedia (Non-Contextual) Analyzer ---
def _analyze_word_with_hanta(word: str, top_n_value: Optional[float] = 0) -> Dict[str, Any]:
    """
    (FALLBACK ENGINE 2) Analyzes a single word using HanTa + OdeNet + Pattern.
    This was the V18 engine. Returns {} on failure.
    """
    if not HANTA_AVAILABLE:
        return {} # Signal failure
        
    top_n = int(top_n_value) if top_n_value is not None else 0
    print(f"\n[Word Encyclopedia] Running V18 (HanTa) fallback for: \"{word}\"")
    final_result: Dict[str, Any] = {
        "input_word": word,
        "analysis": {}
    }
    word_lower = word.lower() # For validation
    
    try:
        hanta_tagger = hanta_get_tagger()
        if not hanta_tagger:
            raise Exception("HanTa Tagger failed to initialize.")

        hanta_tags = _hanta_get_candidates(word, hanta_tagger)
        if not hanta_tags:
            return {}

        pos_groups_map = _hanta_map_tags_to_pos(hanta_tags)
        log(f"Found {len(pos_groups_map)} possible POS group(s): {list(pos_groups_map.keys())}")

        for pos_group, specific_tags in pos_groups_map.items():
            print(f"--- Analyzing as: {pos_group.upper()} ---")
            
            lemma = _hanta_get_lemma_for_pos(word, pos_group, hanta_tagger)
            log(f"Lemma for {pos_group} is: '{lemma}'")
            
            all_odenet_senses = _get_odenet_senses_by_pos(lemma)
            pos_odenet_senses = all_odenet_senses.get(pos_group, [])
            
            if not pos_odenet_senses:
                log(f"✗ REJECTED {pos_group}: OdeNet is available but has no '{pos_group}' senses for lemma '{lemma}'.")
                continue
            
            if pos_odenet_senses and "info" in pos_odenet_senses[0]:
                log(f"✓ VERIFIED {pos_group}: OdeNet is unavailable, proceeding without validation.")
                pos_odenet_senses = []
            else:
                log(f"✓ VERIFIED {pos_group}: OdeNet found {len(pos_odenet_senses)} sense(s).")
            
            # --- 1. Get Inflections (Pattern) ---
            inflection_report = {}
            if not PATTERN_DE_AVAILABLE:
                inflection_report = {"info": "pattern.de library not available. No inflections generated."}
            else:
                try:
                    if pos_group == "noun":
                        inflection_report = pattern_analyze_as_noun(lemma)
                    elif pos_group == "verb":
                        inflection_report = pattern_analyze_as_verb(lemma)
                    elif pos_group == "adjective":
                        inflection_report = pattern_analyze_as_adjective(lemma)
                    elif pos_group == "adverb":
                        inflection_report = {"base_form": lemma, "info": "Adverbs are non-inflecting."}
                    
                    if not pattern_is_good_analysis(inflection_report, pos_group) and pos_group != "adverb":
                        log(f"⚠️ Warning: pattern.de generated a poor inflection table for {lemma} ({pos_group}).")
                        inflection_report["warning"] = "Inflection table from pattern.de seems incomplete or invalid."
                except Exception as e:
                    log(f"pattern.de inflection failed for {lemma} ({pos_group}): {e}")
                    inflection_report = {"error": f"pattern.de failed: {e}", "traceback": traceback.format_exc()}

            # --- 2. Build Semantics Block ---
            semantics_block = _build_semantics_block_for_lemma(lemma, pos_group, top_n)

            # --- 3. Build Final Report Block ---
            pos_entry_report = {
                "hanta_analysis": {
                    "detected_tags": sorted(list(specific_tags)),
                    "lemma": lemma,
                    "morphemes": [
                         hanta_tagger.analyze(word.capitalize() if pos_group == 'noun' else word.lower(), taglevel=3)
                    ]
                },
                "inflections_pattern": inflection_report,
                "semantics_combined": semantics_block
            }
            
            # --- 4. *** VALIDATION FILTER *** ---
            is_valid = False
            if lemma.lower() == word_lower:
                is_valid = True
                log(f"[DEBUG] HanTa: KEEPING entry '{lemma}' ({pos_group}) because input word matches lemma.")
            
            if not is_valid:
                # Check pattern.de's lexeme (for verbs)
                for form in inflection_report.get("lexeme", []):
                    if form.lower() == word_lower:
                        is_valid = True
                        log(f"[DEBUG] HanTa: KEEPING entry '{lemma}' ({pos_group}) because input word found in pattern.de lexeme.")
                        break
            
            if not is_valid:
                # Check pattern.de's participles (for "abgeschnitten")
                for part_form in inflection_report.get("participles", {}).values():
                    if part_form.lower() == word_lower:
                        is_valid = True
                        log(f"[DEBUG] HanTa: KEEPING entry '{lemma}' ({pos_group}) because input word found in pattern.de participles.")
                        break
            
            if not is_valid and pos_group == "adjective":
                 # Check adjective forms
                 if word_lower == inflection_report.get("predicative", "").lower() or \
                    word_lower == inflection_report.get("comparative", "").lower() or \
                    word_lower == inflection_report.get("superlative", "").lower():
                     is_valid = True
                     log(f"[DEBUG] HanTa: KEEPING entry '{lemma}' ({pos_group}) because input word matches adj comparison form.")

            if not is_valid and pos_group == "noun":
                 # Check noun forms
                 if word_lower == inflection_report.get("singular", "").lower() or \
                    word_lower == inflection_report.get("plural", "").lower():
                     is_valid = True
                     log(f"[DEBUG] HanTa: KEEPING entry '{lemma}' ({pos_group}) because input word matches noun singular/plural.")
            
            if not is_valid and pos_group == "adverb":
                is_valid = True # Adverbs are non-inflecting, always keep.

            if is_valid:
                if pos_group not in final_result["analysis"]:
                    final_result["analysis"][pos_group] = []
                final_result["analysis"][pos_group].append(pos_entry_report)
            else:
                 log(f"[DEBUG] HanTa: DROPPING entry '{lemma}' ({pos_group}) because input word '{word}' was not found in its valid forms.")
            # --- END OF VALIDATION ---

        if not final_result["analysis"]:
            return {} # No results
            
        final_result["info"] = "Analysis performed by HanTa-led fallback engine."
        return final_result

    except Exception as e:
        print(f"[Word Encyclopedia] HanTa FALLBACK Engine FAILED: {e}")
        traceback.print_exc()
        return {} # Signal failure
    
def _analyze_word_with_iwnlp(word: str, top_n_value: Optional[float] = 0) -> Dict[str, Any]:
    """
    (FALLBACK ENGINE 3) Analyzes a single word using IWNLP + OdeNet + Pattern.
    This is the full V16/V18 logic, restored and with the new validation filter.
    Returns {} on failure.
    """
    if not word or not word.strip():
        return {} # Use empty dict for "info"

    if not IWNLP_AVAILABLE:
        return {} # Signal failure

    top_n = int(top_n_value) if top_n_value is not None else 0
        
    print(f"\n[Word Encyclopedia] Running IWNLP-fallback analysis for: \"{word}\" (top_n={top_n})")
    
    final_result: Dict[str, Any] = {
        "input_word": word,
        "analysis": {}
    }
    word_lower = word.lower() # For validation
    
    # --- Helper: Get OdeNet senses ---
    def _get_odenet_senses_by_pos_internal(w):
        """
        (Internal helper for IWNLP fallback)
        OdeNet uses 'a' for BOTH Adjective and Adverb.
        """
        senses_by_pos: Dict[str, List[Dict]] = {
            "noun": [], "verb": [], "adjective": [], "adverb": []
        }
        if not WN_AVAILABLE:
            log(f"[IWNLP Fallback] OdeNet check skipped for '{w}': WN_AVAILABLE=False")
            # Fail-open strategy
            return {"noun": [{"info": "OdeNet unavailable"}],
                    "verb": [{"info": "OdeNet unavailable"}],
                    "adjective": [{"info": "OdeNet unavailable"}],
                    "adverb": [{"info": "OdeNet unavailable"}]}
        try:
            all_senses = odenet_get_thesaurus_info(w).get("senses", [])
            for sense in all_senses:
                if "error" in sense: continue
                pos_tag = sense.get("pos")
                if pos_tag == 'n':
                    senses_by_pos["noun"].append(sense)
                elif pos_tag == 'v':
                    senses_by_pos["verb"].append(sense)
                elif pos_tag == 'a':
                    log(f"[IWNLP Fallback] Found OdeNet 'a' tag (Adj/Adv) for sense: {sense.get('definition', '...')[:30]}")
                    senses_by_pos["adjective"].append(sense)
                    senses_by_pos["adverb"].append(sense)
        except Exception as e:
            print(f"[Word Encyclopedia] OdeNet check failed: {e}")
        return senses_by_pos

    # --- 1. GET ALL LEMMA CANDIDATES & SPACY POS ---
    try:
        iwnlp = iwnlp_get_pipeline()
        if not iwnlp:
            return {} # Signal failure
            
        doc = iwnlp(word)
        token = doc[0]
        
        spacy_pos = token.pos_ # e.g., "NOUN" for "Lauf", "ADV" for "heute"
        spacy_lemma = token.lemma_
        iwnlp_lemmas_list = token._.iwnlp_lemmas or []
        
        all_lemmas = set(iwnlp_lemmas_list)
        all_lemmas.add(spacy_lemma)
        all_lemmas.add(word) # Add the word itself
        
        print(f"[Word Encyclopedia] spaCy POS: {spacy_pos}")
        print(f"[Word Encyclopedia] All lemmas to check: {all_lemmas}")
        
    except Exception as e:
        traceback.print_exc()
        return {} # Signal failure

    # --- 2. CHECK INFLECTING POSSIBILITIES FOR EACH LEMMA ---
    valid_analyses: Dict[str, Dict[str, Any]] = {}
    for lemma in all_lemmas:
        if not lemma: continue
        
        odenet_senses_by_pos = _get_odenet_senses_by_pos_internal(lemma)

        # --- Check NOUN ---
        if 'noun' not in valid_analyses:
            noun_inflections = {}
            is_good_noun = False
            if not PATTERN_DE_AVAILABLE:
                noun_inflections = {"info": "pattern.de not available."}
                is_good_noun = True 
            else:
                try:
                    noun_inflections = pattern_analyze_as_noun(lemma.capitalize())
                    if pattern_is_good_analysis(noun_inflections, "noun"):
                        is_good_noun = True
                except Exception as e:
                    noun_inflections = {"error": f"pattern.de failed: {e}"}

            if is_good_noun:
                odenet_senses = odenet_senses_by_pos.get('noun', [])
                if not odenet_senses and lemma.lower() == word.lower():
                     odenet_senses = _get_odenet_senses_by_pos_internal(lemma.capitalize()).get('noun', [])
                if odenet_senses:
                    if "info" not in odenet_senses[0] or not WN_AVAILABLE:
                        log(f"  ✓ [IWNLP Fallback] Valid NOUN found: {lemma}")
                        valid_analyses['noun'] = {
                            "lemma": noun_inflections.get("base_form", lemma),
                            "inflections": noun_inflections,
                            "odenet_senses": [] if "info" in odenet_senses[0] else odenet_senses
                        }

        # --- Check VERB ---
        if 'verb' not in valid_analyses:
            verb_inflections = {}
            is_good_verb = False
            if not PATTERN_DE_AVAILABLE:
                verb_inflections = {"info": "pattern.de not available."}
                is_good_verb = True
            else:
                try:
                    verb_inflections = pattern_analyze_as_verb(lemma)
                    if pattern_is_good_analysis(verb_inflections, "verb"):
                        is_good_verb = True
                except Exception as e:
                    verb_inflections = {"error": f"pattern.de failed: {e}"}

            if is_good_verb:
                odenet_senses = odenet_senses_by_pos.get('verb', [])
                if odenet_senses:
                    if "info" not in odenet_senses[0] or not WN_AVAILABLE:
                        log(f"  ✓ [IWNLP Fallback] Valid VERB found: {lemma}")
                        valid_analyses['verb'] = {
                            "lemma": verb_inflections.get("infinitive", lemma),
                            "inflections": verb_inflections,
                            "odenet_senses": [] if "info" in odenet_senses[0] else odenet_senses
                        }

        # --- Check ADJECTIVE ---
        if 'adjective' not in valid_analyses:
            adj_inflections = {}
            is_good_adj = False
            if not PATTERN_DE_AVAILABLE:
                adj_inflections = {"info": "pattern.de not available."}
                is_good_adj = True
            else:
                try:
                    adj_inflections = pattern_analyze_as_adjective(lemma)
                    if pattern_is_good_analysis(adj_inflections, "adjective"):
                        is_good_adj = True
                except Exception as e:
                    adj_inflections = {"error": f"pattern.de failed: {e}"}
            
            if is_good_adj:
                odenet_senses = odenet_senses_by_pos.get('adjective', [])
                if odenet_senses:
                    if "info" not in odenet_senses[0] or not WN_AVAILABLE:
                        log(f"  ✓ [IWNLP Fallback] Valid ADJECTIVE found: {lemma}")
                        valid_analyses['adjective'] = {
                            "lemma": adj_inflections.get("predicative", lemma),
                            "inflections": adj_inflections,
                            "odenet_senses": [] if "info" in odenet_senses[0] else odenet_senses
                        }

    # --- 3. CHECK NON-INFLECTING POS (ADVERB) ---
    if spacy_pos == "ADV":
        odenet_senses = _get_odenet_senses_by_pos_internal(word).get('adverb', [])
        if odenet_senses:
            if "info" not in odenet_senses[0] or not WN_AVAILABLE:
                log(f"  ✓ [IWNLP Fallback] Valid ADVERB found: {word}")
                valid_analyses['adverb'] = {
                    "lemma": word,
                    "inflections": {"base_form": word},
                    "odenet_senses": [] if "info" in odenet_senses[0] else odenet_senses
                }
            
    # --- 4. CHECK OTHER FUNCTION WORDS (e.g. "mein" -> DET) ---
    FUNCTION_POS = {"DET", "PRON", "ADP", "AUX", "CCONJ", "SCONJ", "PART", "PUNCT", "SYM"}
    if spacy_pos in FUNCTION_POS and not valid_analyses:
        pos_key = spacy_pos.lower()
        print(f"  ✓ Valid Function Word found: {word} (POS: {spacy_pos})")
        valid_analyses[pos_key] = {
            "lemma": spacy_lemma,
            "inflections": {"base_form": spacy_lemma},
            "odenet_senses": [], 
            "spacy_analysis": { 
                "word": token.text, "lemma": token.lemma_,
                "pos_UPOS": token.pos_, "pos_TAG": token.tag_,
                "morphology": str(token.morph)
            }
        }

    # --- 5. BUILD FINAL REPORT (V21 MODIFIED + VALIDATION) ---
    for pos_key, analysis_data in valid_analyses.items():
        lemma = analysis_data["lemma"]
        inflection_block = analysis_data["inflections"]
        
        # --- E. VALIDATION FILTER ---
        is_valid = False
        if lemma.lower() == word_lower:
            is_valid = True
            log(f"[DEBUG] IWNLP: KEEPING entry '{lemma}' ({pos_key}) because input word matches lemma.")
        
        if not is_valid:
            # Check pattern.de's lexeme (for verbs)
            for form in inflection_block.get("lexeme", []):
                if form.lower() == word_lower:
                    is_valid = True
                    log(f"[DEBUG] IWNLP: KEEPING entry '{lemma}' ({pos_key}) because input word found in pattern.de lexeme.")
                    break
        
        if not is_valid:
            # Check pattern.de's participles (for "abgeschnitten")
            for part_form in inflection_block.get("participles", {}).values():
                if part_form.lower() == word_lower:
                    is_valid = True
                    log(f"[DEBUG] IWNLP: KEEPING entry '{lemma}' ({pos_key}) because input word found in pattern.de participles.")
                    break
        
        if not is_valid and pos_key == "adjective":
             # Check adjective forms
             if word_lower == inflection_block.get("predicative", "").lower() or \
                word_lower == inflection_block.get("comparative", "").lower() or \
                word_lower == inflection_block.get("superlative", "").lower():
                 is_valid = True
                 log(f"[DEBUG] IWNLP: KEEPING entry '{lemma}' ({pos_key}) because input word matches adj comparison form.")

        if not is_valid and pos_key == "noun":
             # Check noun forms
             if word_lower == inflection_block.get("singular", "").lower() or \
                word_lower == inflection_block.get("plural", "").lower():
                 is_valid = True
                 log(f"[DEBUG] IWNLP: KEEPING entry '{lemma}' ({pos_key}) because input word matches noun singular/plural.")
        
        if not is_valid and (pos_key == "adverb" or "spacy_analysis" in analysis_data):
            is_valid = True # Adverbs and Function Words are non-inflecting, always keep.
            log(f"[DEBUG] IWNLP: KEEPING entry '{lemma}' ({pos_key}) because it is a non-inflecting word (ADV/FUNC).")

        if is_valid:
            pos_report = {
                "inflections_pattern": inflection_block,
                # Use the new global helper
                "semantics_combined": _build_semantics_block_for_lemma(
                    lemma,
                    pos_key,
                    top_n
                )
            }
            if "spacy_analysis" in analysis_data:
                 pos_report["spacy_analysis"] = analysis_data["spacy_analysis"]
            
            if pos_key not in final_result["analysis"]:
                 final_result["analysis"][pos_key] = []
            final_result["analysis"][pos_key].append(pos_report)
        else:
             log(f"[DEBUG] IWNLP: DROPPING entry '{lemma}' ({pos_key}) because input word '{word}' was not found in its valid forms.")
        # --- END VALIDATION ---

    if not final_result["analysis"]:
       return {} # No results

    final_result["info"] = "Analysis performed by IWNLP-based fallback engine."
    return final_result


# --- 7b. Word Encyclopedia (Non-Contextual) Analyzer ---

# --- PUBLIC DISPATCHER FUNCTION ---
# --- THIS IS THE NEW PUBLIC DISPATCHER FUNCTION ---
def analyze_word_encyclopedia(word: str, top_n_value: Optional[float] = 0, engine_choice: str = "wiktionary") -> Dict[str, Any]:
    """
    (PUBLIC DISPATCHER V22) Analyzes a single word using the selected engine
    as a starting point, then automatically falls back if no results are found.
    
    Chain: Wiktionary -> DWDSmor -> HanTa -> IWNLP
    """
    if not word or not word.strip():
        return {"info": "Please enter a word."}
    
    word = word.strip()
    top_n = int(top_n_value) if top_n_value is not None else 0
    result = {}
    info_log = [] # To track which engines failed

    log(f"\n[Word Encyclopedia] User selected engine: '{engine_choice}' for word: '{word}'")

    try:
        # --- 1. Try Wiktionary ---
        if engine_choice == "wiktionary":
            log(f"[DEBUG] V22 Dispatcher: Trying Wiktionary (Primary) for '{word}'...")
            result = _analyze_word_with_wiktionary(word, top_n)
            if result and result.get("analysis"):
                return result # Success
            info_log.append("Wiktionary found no results.")
            log(f"[DEBUG] V22 Dispatcher: Wiktionary found no results. Falling back to DWDSmor...")
        
        # --- 2. Try DWDSmor (NEW) ---
        if engine_choice == "dwdsmor" or (engine_choice == "wiktionary" and not result.get("analysis")):
            log(f"[DEBUG] V22 Dispatcher: Trying DWDSmor (Fallback 1) for '{word}'...")
            result = _analyze_word_with_dwdsmor(word, top_n)
            if result and result.get("analysis"):
                result["info"] = f"Analysis from DWDSmor (Fallback 1). {(' '.join(info_log))}"
                return result # Success
            info_log.append("DWDSmor found no results.")
            log(f"[DEBUG] V22 Dispatcher: DWDSmor found no results. Falling back to HanTa...")

        # --- 3. Try HanTa ---
        if engine_choice == "hanta" or (not result.get("analysis")):
            log(f"[DEBUG] V22 Dispatcher: Trying HanTa (Fallback 2) for '{word}'...")
            result = _analyze_word_with_hanta(word, top_n)
            if result and result.get("analysis"):
                result["info"] = f"Analysis from HanTa (Fallback 2). {(' '.join(info_log))}"
                return result # Success
            info_log.append("HanTa found no results.")
            log(f"[DEBUG] V22 Dispatcher: HanTa found no results. Falling back to IWNLP...")

        # --- 4. Try IWNLP ---
        if engine_choice == "iwnlp" or (not result.get("analysis")):
            log(f"[DEBUG] V22 Dispatcher: Trying IWNLP (Fallback 3) for '{word}'...")
            result = _analyze_word_with_iwnlp(word, top_n)
            if result and result.get("analysis"):
                result["info"] = f"Analysis from IWNLP (Fallback 3). {(' '.join(info_log))}"
                return result # Success
            info_log.append("IWNLP found no results.")
            
    except Exception as e:
        log(f"--- Dispatcher FAILED for engine {engine_choice}: {e} ---")
        traceback.print_exc()
        return {
            "input_word": word,
            "error": f"An engine failed during analysis.",
            "traceback": traceback.format_exc()
        }

    # --- No engines found anything ---
    log(f"[DEBUG] V22 Dispatcher: All engines failed to find results for '{word}'.")
    return {
        "input_word": word,
        "info": f"No analysis found. All engines failed. ({' '.join(info_log)})"
    }


# ============================================================================
# 7.5 VISUALIZATION & HTML HELPERS (DE)
# ============================================================================

HTML_CSS = """
<style>
    /* Card Container - High Contrast */
    .ling-card { 
        font-family: 'Segoe UI', Roboto, Helvetica, Arial, sans-serif; 
        border: 1px solid #d1d5db; 
        border-radius: 8px; 
        padding: 20px; 
        margin-bottom: 20px; 
        background: #ffffff; 
        box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06); 
    }

    /* Header Section */
    .ling-header { 
        display: flex; 
        align-items: center; 
        margin-bottom: 15px; 
        border-bottom: 2px solid #e5e7eb; 
        padding-bottom: 10px; 
    }
    .ling-lemma { 
        font-size: 1.8em; 
        font-weight: 800; 
        color: #111827; 
        margin-right: 12px; 
    }
    .ling-pos { 
        font-size: 0.85em; 
        text-transform: uppercase; 
        font-weight: 700; 
        padding: 4px 10px; 
        border-radius: 6px; 
        color: #fff; 
        letter-spacing: 0.05em;
    }
    
    /* POS Colors */
    .pos-noun { background-color: #2563eb; }       /* Blue */
    .pos-verb { background-color: #059669; }       /* Green */
    .pos-adj { background-color: #d97706; }        /* Amber */
    .pos-adv { background-color: #7c3aed; }        /* Purple */
    .pos-name { background-color: #db2777; }       /* Pink */
    .pos-other { background-color: #4b5563; }      /* Gray */

    /* Section Headers */
    .ling-section { margin-top: 15px; }
    .ling-subtitle { 
        font-size: 0.85em; 
        font-weight: 700; 
        color: #374151; 
        text-transform: uppercase; 
        margin-bottom: 8px; 
        border-left: 4px solid #3b82f6;
        padding-left: 8px;
    }
    
    /* Tables */
    .inflection-table { width: 100%; font-size: 0.95em; border-collapse: collapse; margin-bottom: 10px; }
    .inflection-table td { padding: 6px 10px; border-bottom: 1px solid #e5e7eb; color: #1f2937; }
    .inflection-label { color: #6b7280; font-weight: 600; width: 35%; background: #f9fafb; }
    
    /* Senses */
    .sense-item { margin-bottom: 8px; line-height: 1.5; font-size: 1em; color: #1f2937; }
    .source-badge { 
        display: inline-block; font-size: 0.75em; font-weight: bold; padding: 2px 6px; 
        border-radius: 4px; border: 1px solid; margin-right: 8px; vertical-align: middle; 
    }
    .src-wikt { background: #fff1f2; color: #9f1239; border-color: #fda4af; }
    .src-oewn { background: #eff6ff; color: #1e40af; border-color: #93c5fd; }
    
    /* Relations Chips */
    .rel-chip { 
        display: inline-block; 
        background: #f3f4f6; 
        color: #1f2937;
        padding: 4px 10px; 
        border-radius: 15px; 
        font-size: 0.9em; 
        margin: 3px; 
        border: 1px solid #d1d5db; 
        font-weight: 500;
    }
    .rel-type { color: #6b7280; font-size: 0.8em; margin-right: 4px; font-weight: 700; text-transform: lowercase;}
    
    /* Collapsible */
    .kg-details > summary { 
        cursor: pointer; color: #2563eb; font-size: 0.9em; font-weight: 600; 
        margin-top: 10px; padding: 6px; border-radius: 4px; width: fit-content;
    }
    .kg-details > summary:hover { text-decoration: underline; background: #eff6ff; }
    .kg-content { margin-top: 10px; padding: 10px; background: #f9fafb; border-radius: 8px; border: 1px solid #e5e7eb; }
    
    /* Alert Banners */
    .grammar-alert { padding: 10px; border-radius: 6px; margin-bottom: 15px; border: 1px solid; }
    .alert-green { background: #ecfdf5; color: #065f46; border-color: #a7f3d0; }
    .alert-red { background: #fef2f2; color: #991b1b; border-color: #fecaca; }
</style>
"""

def _format_word_analysis_html(data: Dict[str, Any]) -> str:
    """ 
    Generates HTML for a single word analysis (German version).
    Renders the 'inflections_pattern' block, which contains the 
    Hybrid (Wiktionary-verified) data from the backend.
    """
    if not data or "analysis" not in data:
        return f"{HTML_CSS}<div class='ling-card'>Keine Daten verfügbar. {data.get('info', '')}</div>"
    
    html = HTML_CSS
    analysis = data["analysis"]
    
    # Iterate over POS categories (noun, verb, etc.)
    for pos_key, entries in analysis.items():
        if not entries: continue
        
        # We usually display the best candidate, but if there are multiple distinct entries 
        # (like "der See" vs "die See"), the backend groups them in the list. 
        # We should ideally render ALL entries in the list to show the homonyms.
        # This loop handles that.
        for entry in entries:
            
            # Data Extraction
            inf_wikt = entry.get("inflections_wiktionary") or {}
            inf_pat = entry.get("inflections_pattern") or {}
            sem_comb = entry.get("semantics_combined") or {}
            meta = entry.get("wiktionary_metadata") or {}
            
            lemma = inf_wikt.get("base_form") or \
                    inf_pat.get("base_form") or \
                    sem_comb.get("lemma") or \
                    data.get("input_word") or "?"

            # --- POS Display Logic ---
            display_pos = pos_key.upper()
            css_class = "pos-other"
            
            if pos_key == 'noun': 
                css_class = "pos-noun"
                display_pos = "SUBSTANTIV"
                # Append Gender to POS badge if available
                if "gender" in inf_pat:
                    gender_map = {"Masculine": "M", "Feminine": "F", "Neuter": "N"}
                    g_short = gender_map.get(inf_pat['gender'], "?")
                    display_pos += f" ({g_short})"
                    
            elif pos_key == 'verb': 
                css_class = "pos-verb"
                display_pos = "VERB"
            elif pos_key == 'adj' or pos_key == 'adjective': 
                css_class = "pos-adj"
                display_pos = "ADJEKTIV"
            elif pos_key == 'adv' or pos_key == 'adverb': 
                css_class = "pos-adv"
                display_pos = "ADVERB"

            # --- CARD START ---
            html += f"""
            <div class="ling-card">
                <div class="ling-header">
                    <span class="ling-lemma">{lemma}</span>
                    <span class="ling-pos {css_class}">{display_pos}</span>
            """
            
            # Add small title if available (e.g., "Konjugierte Form")
            if meta.get("pos_title"):
                 html += f"<span style='margin-left:10px; color:#6b7280; font-size:0.85em;'>{meta['pos_title']}</span>"
            
            html += "</div>" # End Header
            
            # --- SOURCE BADGE LOGIC ---
            # Determine credibility of the data
            is_hybrid = inf_pat.get("is_hybrid", False)
            wikt_forms_count = len(inf_wikt.get("forms_list", []))
            
            badge_style = "float:right; font-weight:bold; font-size:0.75em; padding:2px 6px; border-radius:4px;"
            
            if is_hybrid:
                source_html = f"<span style='{badge_style} background:#ecfdf5; color:#065f46; border:1px solid #a7f3d0;'>Quelle: Wiktionary (Verifiziert)</span>"
            elif wikt_forms_count > 0:
                source_html = f"<span style='{badge_style} background:#ecfdf5; color:#065f46; border:1px solid #a7f3d0;'>Quelle: Wiktionary (DB)</span>"
            elif inf_pat and "error" not in inf_pat:
                source_html = f"<span style='{badge_style} background:#fffbeb; color:#92400e; border:1px solid #fcd34d;'>Quelle: Pattern (Generiert)</span>"
            else:
                source_html = ""

            # --- INFLECTIONS SECTION ---
            html += f"<div class='ling-section'>{source_html}<div class='ling-subtitle'>Morphologie & Flexion</div>"
            html += "<table class='inflection-table'>"
            
            # We render the table based on 'inf_pat' because the backend has already merged 
            # the Wiktionary truths into this structure.
            
            if pos_key == 'noun':
                decl = inf_pat.get('declension')
                # Fallback if declension is nested in gender key
                if not decl and inf_pat.get('declension_by_gender'):
                    # If we have a specific gender from the analysis, try to grab that specific table
                    target_gender = inf_pat.get("gender")
                    if target_gender and target_gender in inf_pat['declension_by_gender']:
                        decl = inf_pat['declension_by_gender'][target_gender]
                    else:
                        # Fallback: take the first available
                        first_gender = list(inf_pat['declension_by_gender'].keys())[0]
                        decl = inf_pat['declension_by_gender'][first_gender]
                
                if decl:
                    # Noun Table Rows
                    nom_sg = decl.get('Nominativ Singular', {}).get('definite', '-')
                    nom_pl = decl.get('Nominativ Plural', {}).get('definite', '-')
                    gen_sg = decl.get('Genitiv Singular', {}).get('definite', '-')
                    dat_pl = decl.get('Dativ Plural', {}).get('definite', '-')
                    
                    html += f"<tr><td class='inflection-label'>Nom. Singular</td><td>{nom_sg}</td></tr>"
                    html += f"<tr><td class='inflection-label'>Nom. Plural</td><td>{nom_pl}</td></tr>"
                    html += f"<tr><td class='inflection-label'>Gen. Singular</td><td>{gen_sg}</td></tr>"
                    html += f"<tr><td class='inflection-label'>Dat. Plural</td><td>{dat_pl}</td></tr>"
                else:
                    html += f"<tr><td colspan='2'><i>Keine Flexionsdaten verfügbar.</i></td></tr>"
            
            elif pos_key == 'verb':
                cj = inf_pat.get('conjugation') or {}
                pres = cj.get('Präsens') or {}
                past = cj.get('Präteritum') or {}
                parts = inf_pat.get('participles') or {}
                
                html += f"<tr><td class='inflection-label'>Infinitiv</td><td>{inf_pat.get('infinitive', lemma)}</td></tr>"
                html += f"<tr><td class='inflection-label'>3. Pers. Sg. (er/sie)</td><td>{pres.get('er/sie/es', '-')}</td></tr>"
                html += f"<tr><td class='inflection-label'>Präteritum (ich)</td><td>{past.get('ich', '-')}</td></tr>"
                html += f"<tr><td class='inflection-label'>Partizip II</td><td>{parts.get('Partizip Perfekt', '-')}</td></tr>"
                html += f"<tr><td class='inflection-label'>Konjunktiv II (ich)</td><td>{cj.get('Konjunktiv II', {}).get('ich', '-')}</td></tr>"

            elif pos_key in ['adjective', 'adj']:
                html += f"<tr><td class='inflection-label'>Positiv</td><td>{inf_pat.get('predicative', lemma)}</td></tr>"
                html += f"<tr><td class='inflection-label'>Komparativ</td><td>{inf_pat.get('comparative', '-')}</td></tr>"
                html += f"<tr><td class='inflection-label'>Superlativ</td><td>{inf_pat.get('superlative', '-')}</td></tr>"
            
            elif pos_key in ['adverb', 'adv']:
                 html += f"<tr><td class='inflection-label'>Form</td><td>{lemma} (unveränderlich)</td></tr>"

            html += "</table>"

            # --- RAW FORMS FOOTER (The "Evidence") ---
            # Display the raw forms list from DB if available, as this proves the ground truth
            forms_list = inf_wikt.get("forms_list") or []
            if forms_list:
                # Deduplicate and flatten
                unique_forms = sorted(list(set([f.get('form_text') for f in forms_list if f.get('form_text')])))
                # Limit display to avoid wall of text
                display_forms = ", ".join(unique_forms[:12])
                if len(unique_forms) > 12: display_forms += f", ... ({len(unique_forms)-12} weitere)"
                
                html += f"<div style='font-size:0.8em; color:#6b7280; margin-top:5px;'>"
                html += f"<strong>Beobachtete Formen (DB):</strong> {display_forms}</div>"
            
            html += "</div>"
            
            # --- SEMANTICS SECTION ---
            html += "<div class='ling-section'><div class='ling-subtitle'>Bedeutungen & Definitionen</div>"
            
            wikt_senses = sem_comb.get("wiktionary_senses") or []
            ode_senses = sem_comb.get("odenet_senses") or []
            
            if not wikt_senses and not ode_senses:
                 html += "<div class='sense-item'><i>Keine Definitionen gefunden.</i></div>"

            # Render Wiktionary Senses
            for s in wikt_senses[:3]: 
                gloss_raw = s.get("definition") or "" 
                gloss = str(gloss_raw).replace(";", "<br>")
                if gloss:
                    html += f"<div class='sense-item'><span class='source-badge src-wikt'>Wikt</span> {gloss}</div>"
            
            # Render OdeNet Senses
            for s in ode_senses[:3]:
                defi = s.get("definition") or ""
                if defi:
                    html += f"<div class='sense-item'><span class='source-badge src-oewn'>OdeNet</span> {defi}</div>"
                
            html += "</div>"
            
            # --- RELATIONS SECTION ---
            rels = sem_comb.get("conceptnet_relations") or []
            if rels:
                html += "<div class='ling-section'><div class='ling-subtitle'>Wissensgraph (Kontext)</div>"
                
                top_n_rels = 6
                visible_rels = rels[:top_n_rels]
                hidden_rels = rels[top_n_rels:]
                
                def render_rel(r):
                    rel_name = r.get("relation", "Rel")
                    target = r.get("other_node") or "?"
                    if target == "?" and "surface" in r:
                          parts = str(r["surface"]).split()
                          if len(parts) > 2: target = parts[-1] 
                    return f"<span class='rel-chip'><span class='rel-type'>{rel_name}:</span> {target}</span>"

                html += "<div>"
                for r in visible_rels:
                    html += render_rel(r)
                html += "</div>"

                if hidden_rels:
                    html += f"""
                    <details class='kg-details'>
                        <summary>Zeige {len(hidden_rels)} weitere Relationen</summary>
                        <div class='kg-content'>
                    """
                    for r in hidden_rels:
                        html += render_rel(r)
                    html += "</div></details>"
                
                html += "</div>"
            
            html += "</div>" # End Card (div.ling-card)

    return html

def _map_wikt_form_to_pattern_keys(pos_key: str, tags_str: str) -> Optional[List[str]]:
    """
    Parses a Wiktionary tag string and returns the corresponding path keys 
    for the Pattern.de dictionary structure.
    """
    if not tags_str: return None
    t = tags_str.lower()
    
    if pos_key == "noun":
        # Pattern Structure: [Gender] -> "Nominativ Singular" -> "bare"/"definite"
        case = ""
        if "nominative" in t: case = "Nominativ"
        elif "genitive" in t: case = "Genitiv"
        elif "dative" in t: case = "Dativ"
        elif "accusative" in t: case = "Akkusativ"
        
        number = ""
        if "singular" in t: number = "Singular"
        elif "plural" in t: number = "Plural"
        
        if case and number:
            return [f"{case} {number}"]
            
    elif pos_key == "verb":
        # Pattern Structure: "conjugation" -> "Präsens" -> "ich"
        tense = ""
        if "present" in t: tense = "Präsens"
        elif "past" in t or "preterite" in t: tense = "Präteritum"
        elif "subjunctive i" in t: tense = "Konjunktiv I"
        elif "subjunctive ii" in t: tense = "Konjunktiv II"
        elif "imperative" in t: tense = "Imperativ"
        
        person_key = ""
        if "participle" in t:
            if "past" in t or "perfect" in t: return ["participles", "Partizip Perfekt"]
            if "present" in t: return ["participles", "Partizip Präsens"]
        
        if "singular" in t:
            if "1" in t: person_key = "ich" if tense != "Imperativ" else "du" # 1sg usually not imp, but handling safety
            elif "2" in t: person_key = "du"
            elif "3" in t: person_key = "er/sie/es"
        elif "plural" in t:
            if "1" in t: person_key = "wir"
            elif "2" in t: person_key = "ihr"
            elif "3" in t: person_key = "sie/Sie"
            
        if tense and person_key:
            return ["conjugation", tense, person_key]
            
    elif pos_key == "adjective":
        # Pattern Structure: "comparative", "superlative"
        if "comparative" in t and "predicative" in t: return ["comparative"]
        if "superlative" in t and "predicative" in t: return ["superlative"]
        if "positive" in t and "predicative" in t: return ["predicative"]
        
    return None

def _map_wikt_gender_to_pattern(tags_list: List[str]) -> Optional[int]:
    """ 
    Maps Wiktionary tag strings (e.g., 'masculine') to pattern.de constants. 
    Returns None if no specific gender is found.
    """
    if not tags_list: 
        return None
    
    # Flatten and normalize tags
    # Wiktionary often provides tags like "masculine", "feminine", "neuter"
    tags_lower = [str(t).lower() for t in tags_list]
    
    if "masculine" in tags_lower or "m" in tags_lower:
        return MALE
    if "feminine" in tags_lower or "f" in tags_lower:
        return FEMALE
    if "neuter" in tags_lower or "n" in tags_lower:
        return NEUTRAL
        
    return None

def _format_comprehensive_html(data: Dict[str, Any]) -> str:
    """ Generates HTML for the comprehensive sentence analysis. """
    if "error" in data:
        return f"<div style='color:red'>{data['error']}</div>"
        
    html = HTML_CSS
    
    # 1. Grammar Check Banner
    gc = data.get("grammar_check", [])
    if isinstance(gc, list) and len(gc) == 1 and gc[0].get("status") == "perfect":
        html += "<div class='grammar-alert alert-green'><strong>✓ Grammatikprüfung:</strong> Keine offensichtlichen Fehler gefunden.</div>"
    elif isinstance(gc, list) and gc:
        html += "<div class='grammar-alert alert-red'><strong>⚠ Grammatik-Hinweise:</strong><br>"
        for err in gc:
            msg = err.get("message", "Fehler")
            bad = err.get("incorrect_text", "")
            html += f"• {msg} (in: '<em>{bad}</em>')<br>"
        html += "</div>"
        
    # 2. Lemma Deep Dive Accordion
    deep_dive = data.get("lemma_deep_dive", {})
    if not deep_dive:
        html += "<p>Keine Tiefenanalyse verfügbar.</p>"
    else:
        html += "<h3>Wort-für-Wort Analyse</h3>"
        for lemma, details in deep_dive.items():
            html += f"<details><summary>{lemma}</summary>"
            
            inflections = details.get("inflection_analysis", {})
            semantics = details.get("semantic_analysis", {})
            
            # Guess the POS keys present
            all_keys = set([k.split('_')[0] for k in inflections.keys() if '_wiktionary' in k]) # Filter to avoid capturing 'metadata' as a pos key
            
            reconstructed_data = {"analysis": {}}
            
            for pos in all_keys:
                entry = {
                    "inflections_wiktionary": inflections.get(f"{pos}_wiktionary"),
                    "inflections_pattern": inflections.get(f"{pos}_pattern"),
                    # --- FIX: Inject Metadata back ---
                    "wiktionary_metadata": inflections.get(f"{pos}_metadata"),
                    "semantics_combined": {
                        "lemma": lemma,
                        "wiktionary_senses": [s for s in semantics.get(f"{pos}_senses", []) if s.get('source') == 'wiktionary'],
                        "odenet_senses": [s for s in semantics.get(f"{pos}_senses", []) if s.get('source') == 'odenet'],
                        "conceptnet_relations": semantics.get("conceptnet_relations", [])
                    }
                }
                reconstructed_data["analysis"][pos] = [entry]
                
            html += _format_word_analysis_html(reconstructed_data)
            html += "</details>"
            
    return html

# ============================================================================
# 8. GRADIO UI CREATION
# ============================================================================
def create_spacy_tab():
    """Creates the UI for the spaCy tab."""
    config = SPACY_UI_TEXT["en"]
    model_choices = list(SPACY_MODEL_INFO.keys())
    with gr.Row():
        ui_lang_radio = gr.Radio(["DE", "EN", "ES"], label=config["ui_lang_label"], value="EN")
        model_lang_radio = gr.Radio(
            choices=[(SPACY_MODEL_INFO[k][0], k) for k in model_choices],
            label=config["model_lang_label"],
            value=model_choices[0]
        )
    markdown_title = gr.Markdown(config["title"])
    markdown_subtitle = gr.Markdown(config["subtitle"])
    text_input = gr.Textbox(label=config["input_label"], placeholder=config["input_placeholder"], lines=5)
    analyze_button = gr.Button(config["button_text"], variant="primary")
    with gr.Tabs():
        with gr.Tab(config["tab_graphic"]) as tab_graphic:
            html_dep_out = gr.HTML(label=config["html_label"])
        with gr.Tab(config["tab_ner"]) as tab_ner:
            html_ner_out = gr.HTML(label=config["ner_label"])
        with gr.Tab(config["tab_table"]) as tab_table:
            df_out = gr.DataFrame(label=config["table_label"], headers=config["table_headers"], interactive=False)
        with gr.Tab(config["tab_json"]) as tab_json:
            json_out = gr.JSON(label=config["json_label"])
    analyze_button.click(fn=spacy_get_analysis,
                         inputs=[ui_lang_radio, model_lang_radio, text_input],
                         outputs=[df_out, json_out, html_dep_out, html_ner_out, analyze_button],
                         api_name="get_morphology")
    ui_lang_radio.change(fn=spacy_update_ui,
                         inputs=ui_lang_radio,
                         outputs=[markdown_title, markdown_subtitle, ui_lang_radio, model_lang_radio,
                                  text_input, analyze_button, tab_graphic, tab_table, tab_json, tab_ner,
                                  html_dep_out, df_out, json_out, html_ner_out])

def create_languagetool_tab():
    """Creates the UI for the LanguageTool tab."""
    gr.Markdown("# 🇩🇪 German Grammar & Spelling Checker")
    gr.Markdown("Powered by `language-tool-python`. This service checks German text for grammatical errors and spelling mistakes.")
    with gr.Column():
        text_input = gr.Textbox(
            label="German Text to Check",
            placeholder="e.g., Ich sehe dem Mann. Das ist ein Huas.",
            lines=5
        )
        check_button = gr.Button("Check Text", variant="primary")
    output = gr.JSON(label="Detected Errors (JSON)")
    check_button.click(
        fn=lt_check_grammar,
        inputs=[text_input],
        outputs=[output],
        api_name="check_grammar"
    )
    gr.Examples(
        [["Das ist ein Huas."], ["Ich sehe dem Mann."],
         ["Die Katze schlafen auf dem Tisch."], ["Er fragt ob er gehen kann."]],
        inputs=[text_input], outputs=[output], fn=lt_check_grammar,
        cache_examples=False
    )

def create_odenet_tab():
    """Creates the UI for the OdeNet tab."""
    gr.Markdown("# 🇩🇪 German Thesaurus (WordNet) Service")
    gr.Markdown("Powered by `wn` and `OdeNet (odenet:1.4)`. Finds synonyms, antonyms, and other semantic relations for German words.")
    with gr.Column():
        word_input = gr.Textbox(
            label="German Word",
            placeholder="e.g., Haus, schnell, gut, Katze"
        )
        check_button = gr.Button("Find Relations", variant="primary")
    output = gr.JSON(label="Thesaurus Information (JSON)")
    check_button.click(
        fn=odenet_get_thesaurus_info,
        inputs=[word_input],
        outputs=[output],
        api_name="get_thesaurus"
    )
    gr.Examples(
        [["Hund"], ["gut"], ["laufen"], ["Haus"], ["schnell"]],
        inputs=[word_input], outputs=[output], fn=odenet_get_thesaurus_info,
        cache_examples=False
    )

def create_pattern_tab():
    """Creates the UI for the Pattern.de tab."""
    gr.Markdown("# 🇩🇪 Complete German Word Inflection System")
    gr.Markdown("Powered by `PatternLite`. Generates complete inflection tables (declension, conjugation) for German words. Robustly handles ambiguity (e.g., 'Lauf' vs 'lauf').")
    with gr.Column():
        word_input = gr.Textbox(
            label="German Word",
            placeholder="z.B. Haus, gehen, schön, besser, lief, Lauf, See"
        )
        generate_button = gr.Button("Generate All Forms", variant="primary")
    output = gr.JSON(label="Complete Inflection Analysis")
    generate_button.click(
        fn=pattern_get_all_inflections,
        inputs=[word_input],
        outputs=[output],
        api_name="get_all_inflections"
    )
    gr.Examples(
        [["Haus"], ["gehen"], ["schön"], ["besser"], ["ging"], ["schnellem"], ["Katze"], ["Lauf"], ["See"]],
        inputs=[word_input], outputs=[output], fn=pattern_get_all_inflections,
        cache_examples=False
    )

def create_conceptnet_tab():
    """--- NEW: Creates the UI for the ConceptNet tab ---"""
    gr.Markdown("# 🌍 ConceptNet Knowledge Graph (Direct API)")
    gr.Markdown("Powered by `api.conceptnet.io`. Fetches semantic relations for a word in any language.")
    with gr.Row():
        word_input = gr.Textbox(
            label="Word or Phrase",
            placeholder="e.g., Baum, tree, Katze"
        )
        lang_input = gr.Textbox(
            label="Language Code",
            placeholder="de",
            value="de"
        )
    check_button = gr.Button("Find Relations", variant="primary")
    output = gr.JSON(label="ConceptNet Relations (JSON)")

    check_button.click(
        fn=conceptnet_get_relations,
        inputs=[word_input, lang_input],
        outputs=[output],
        api_name="get_conceptnet"
    )
    gr.Examples(
        [["Baum", "de"], ["tree", "en"], ["Katze", "de"], ["gato", "es"]],
        inputs=[word_input, lang_input], outputs=[output], fn=conceptnet_get_relations,
        cache_examples=False
    )

def create_combined_tab():
    """Creates the UI for the CONTEXTUAL Comprehensive Analyzer tab."""
    gr.Markdown("# 🚀 Umfassende Analyse (Kontextuell)")
    gr.Markdown("Dieses Tool bietet eine tiefe, **lemma-basierte** Analyse *im Kontext*. Es integriert alle Tools und nutzt den **ganzen Satz**, um Bedeutungen nach Relevanz zu sortieren.")
    with gr.Column():
        text_input = gr.Textbox(
            label="Deutscher Text",
            placeholder="z.B., Die schnelle Katze springt über den faulen Hund.",
            lines=5
        )
        top_n_number = gr.Number(
            label="Limit semantische Bedeutungen pro POS (0 für alle)",
            value=0, step=1, minimum=0, interactive=True
        )
        analyze_button = gr.Button("Umfassende Analyse starten", variant="primary")
    
    status_output = gr.Markdown(value="", visible=True)
    # --- NEW: Visual Output ---
    html_output = gr.HTML(label="Visueller Bericht")
    json_output = gr.JSON(label="Rohdaten (JSON)")
    
    def run_analysis_with_status_visual(text, top_n):
        try:
            status = "🔄 Analyse läuft..."
            yield status, "", {} 
            
            result = comprehensive_german_analysis(text, top_n)
            
            # Generate HTML
            html = _format_comprehensive_html(result)
            
            status = f"✅ Analyse abgeschlossen! {len(result.get('lemma_deep_dive', {}))} Lemmata analysiert."
            yield status, html, result
            
        except Exception as e:
            error_status = f"❌ Fehler: {str(e)}"
            yield error_status, f"<div style='color:red'>{str(e)}</div>", {"error": str(e), "traceback": traceback.format_exc()}
    
    analyze_button.click(
        fn=run_analysis_with_status_visual,
        inputs=[text_input, top_n_number],
        outputs=[status_output, html_output, json_output],
        api_name="comprehensive_analysis"
    )
    
    gr.Examples(
        [["Die Katze schlafen auf dem Tisch.", 3], 
         ["Das ist ein Huas.", 0], 
         ["Ich laufe schnell.", 3],
         ["Der Gärtner pflanzt einen Baum.", 5]],
        inputs=[text_input, top_n_number],
        outputs=[status_output, html_output, json_output], 
        fn=run_analysis_with_status_visual,
        cache_examples=False
    )

def create_word_encyclopedia_tab():
    """--- UI for the NON-CONTEXTUAL Word Analyzer tab ---"""
    gr.Markdown("# 📖 Wort-Enzyklopädie (Nicht-Kontextuell)")
    gr.Markdown("Analysiert ein **einzelnes Wort** auf alle grammatikalischen und semantischen Formen.")
    
    with gr.Column():
        word_input = gr.Textbox(
            label="Einzelnes deutsches Wort",
            placeholder="z.B., Lauf, See, schnell, heute"
        )
        
        with gr.Row():
            top_n_number = gr.Number(
                label="Limit semantische Bedeutungen pro POS (0 für alle)",
                value=0, step=1, minimum=0, interactive=True
            )
            
            engine_radio = gr.Radio(
                label="Wähle Analyse-Engine (Automatischer Fallback)",
                choices=[
                    ("Wiktionary (Standard)", "wiktionary"),
                    ("DWDSmor (Neu)", "dwdsmor"), 
                    ("HanTa (Fallback 2)", "hanta"),
                    ("IWNLP (Fallback 3)", "iwnlp")
                ],
                value="wiktionary",
                interactive=True
            )

        analyze_button = gr.Button("Wort analysieren", variant="primary")
    
    # --- NEW: Visual Output ---
    html_output = gr.HTML(label="Visueller Bericht")
    json_output = gr.JSON(label="Analyse Rohdaten (JSON)")
    
    def run_word_visual(word, top_n, engine):
        data = analyze_word_encyclopedia(word, top_n, engine)
        html = _format_word_analysis_html(data)
        return html, data
    
    analyze_button.click(
        fn=run_word_visual,
        inputs=[word_input, top_n_number, engine_radio], 
        outputs=[html_output, json_output],
        api_name="analyze_word"
    )
    
    gr.Examples(
        [["Lauf", 3, "wiktionary"], 
         ["See", 0, "wiktionary"], 
         ["schnell", 3, "wiktionary"],
         ["gebildet", 0, "dwdsmor"]],
        inputs=[word_input, top_n_number, engine_radio],
        outputs=[html_output, json_output], 
        fn=run_word_visual,
        cache_examples=False
    )

def create_wiktionary_tab():
    """Creates the UI for the standalone Wiktionary lookup tab."""
    gr.Markdown("# 📙 Wiktionary Lookup (Raw Engine)")
    gr.Markdown("Directly query the Wiktionary (Primary) engine. This shows the raw, combined data from the database, Pattern.de, and semantic sources.")
    with gr.Column():
        word_input = gr.Textbox(
            label="Single German Word",
            placeholder="e.g., Haus, gehe, heute"
        )
        analyze_button = gr.Button("Lookup Word in Wiktionary", variant="primary")
    
    output = gr.JSON(label="Wiktionary Engine Analysis (JSON)")
    
    # Call the internal engine function directly, hardcoding top_n=0
    analyze_button.click(
        fn=lambda word: _analyze_word_with_wiktionary(word, 0),
        inputs=[word_input],
        outputs=[output],
        api_name="wiktionary_lookup"
    )
    gr.Examples(
        [["Haus"], ["gehe"], ["heute"], ["Lauf"]],
        inputs=[word_input], outputs=[output], fn=lambda word: _analyze_word_with_wiktionary(word, 0),
        cache_examples=False
    )

def create_dwdsmor_tab():
    """Creates the UI for the standalone DWDSmor lookup tab."""
    gr.Markdown("# 🏛️ DWDSmor Morphology (Raw Engine)")
    gr.Markdown("Directly query the `dwdsmor` FST-based engine. This is a high-precision morphological analyzer.")
    
    def dwdsmor_raw_analysis(word):
        """Wrapper to get raw DWDSmor analysis as JSON."""
        if not DWDSMOR_AVAILABLE:
            return {"error": "DWDSmor library not installed."}
        try:
            analyzer = dwdsmor_get_lemmatizer()
            if not analyzer:
                return {"error": "DWDSmor lemmatizer failed to initialize."}
            
            # --- THIS IS THE FIX ---
            # The analyzer.analyze() returns a Traversal object, which is iterable
            analyses = list(analyzer.analyze(word, join_tags=True))
            # --- END OF FIX ---
            
            if not analyses:
                return {"info": f"No analysis found for '{word}'."}
            
            # Convert Traversal objects to plain dicts for JSON output
            results = []
            for analysis in analyses:
                results.append({
                    "lemma": analysis.analysis, # In this object, .analysis is the lemma
                    "pos": analysis.pos,
                    "analysis_string": analysis.spec, # .spec is the full string
                    "tags": analysis.tags
                })
            return {"input_word": word, "analyses": results}
        except Exception as e:
            return {"error": str(e), "traceback": traceback.format_exc()}

    with gr.Column():
        word_input = gr.Textbox(
            label="Single German Word",
            placeholder="e.g., gebildet, schnell, Häuser"
        )
        analyze_button = gr.Button("Analyze Word with DWDSmor", variant="primary")
    
    output = gr.JSON(label="DWDSmor Raw Analysis (JSON)")
    
    analyze_button.click(
        fn=dwdsmor_raw_analysis,
        inputs=[word_input],
        outputs=[output],
        api_name="dwdsmor_lookup"
    )
    gr.Examples(
        [["gebildet"], ["schnell"], ["Häuser"], ["gehe"]],
        inputs=[word_input], outputs=[output], fn=dwdsmor_raw_analysis,
        cache_examples=False
    )

def create_hanta_tab():
    """Creates the UI for the standalone HanTa Engine tab."""
    gr.Markdown("# 🤖 HanTa Lookup (Raw Engine)")
    gr.Markdown("Directly query the HanTa (Fallback 1) engine. This shows the raw, combined data from HanTa, Pattern.de, and semantic sources.")
    with gr.Column():
        word_input = gr.Textbox(
            label="Single German Word",
            placeholder="e.g., Haus, gehe, heute"
        )
        analyze_button = gr.Button("Lookup Word with HanTa", variant="primary")
    
    output = gr.JSON(label="HanTa Engine Analysis (JSON)")
    
    # Call the internal engine function directly, hardcoding top_n=0
    analyze_button.click(
        fn=lambda word: _analyze_word_with_hanta(word, 0),
        inputs=[word_input],
        outputs=[output],
        api_name="hanta_lookup"
    )
    gr.Examples(
        [["Haus"], ["gehe"], ["heute"], ["Lauf"]],
        inputs=[word_input], outputs=[output], fn=lambda word: _analyze_word_with_hanta(word, 0),
        cache_examples=False
    )

def create_iwnlp_tab():
    """Creates the UI for the standalone IWNLP Engine tab."""
    gr.Markdown("# 🔬 IWNLP-spaCy Lookup (Raw Engine)")
    gr.Markdown("Directly query the IWNLP-spaCy (Fallback 2) engine. This shows the raw, combined data from spaCy, IWNLP, Pattern.de, and semantic sources.")
    with gr.Column():
        word_input = gr.Textbox(
            label="Single German Word",
            placeholder="e.g., Haus, gehe, heute"
        )
        analyze_button = gr.Button("Lookup Word with IWNLP", variant="primary")
    
    output = gr.JSON(label="IWNLP Engine Analysis (JSON)")
    
    # Call the internal engine function directly, hardcoding top_n=0
    analyze_button.click(
        fn=lambda word: _analyze_word_with_iwnlp(word, 0),
        inputs=[word_input],
        outputs=[output],
        api_name="iwnlp_lookup"
    )
    gr.Examples(
        [["Haus"], ["gehe"], ["heute"], ["Lauf"]],
        inputs=[word_input], outputs=[output], fn=lambda word: _analyze_word_with_iwnlp(word, 0),
        cache_examples=False
    )

# --- Main UI Builder ---
def create_consolidated_interface():
    """Builds the final Gradio app with all tabs."""
    with gr.Blocks(title="Consolidated Linguistics Hub", theme=gr.themes.Soft()) as demo:
        gr.Markdown("# 🏛️ Consolidated Linguistics Hub")
        gr.Markdown("A suite of advanced tools for German linguistics, providing both contextual and non-contextual analysis.")
        
        with gr.Tabs():
            # --- Main Tools ---
            with gr.Tab("📖 Word Encyclopedia (DE)"):
                create_word_encyclopedia_tab()
                
            with gr.Tab("🚀 Comprehensive Analyzer (DE)"):
                create_combined_tab()

            with gr.Tab("🔬 spaCy Analyzer (Multi-lingual)"):
                create_spacy_tab()
            
            with gr.Tab("✅ Grammar Check (DE)"):
                create_languagetool_tab()
            
            # --- Standalone Engine Tabs (NEW) ---
            with gr.Tab("📙 Engine: Wiktionary (DE)"):
                create_wiktionary_tab()

            with gr.Tab("🤖 Engine: HanTa (DE)"):
                create_hanta_tab()

            with gr.Tab("🔬 Engine: IWNLP-spaCy (DE)"):
                create_iwnlp_tab()

            with gr.Tab("🏛️ Engine: DWDSmor (DE)"):
                create_dwdsmor_tab()
                
            # --- Standalone Component Tabs ---
            with gr.Tab("📚 Component: Inflections (DE)"):
                create_pattern_tab()

            with gr.Tab("📖 Component: Thesaurus (DE)"):
                create_odenet_tab()
            
            with gr.Tab("🌐 Component: ConceptNet (Direct)"):
                create_conceptnet_tab()
    
    return demo

# ============================================================================
# 9. MAIN EXECUTION BLOCK
# ============================================================================

if __name__ == "__main__":
    print("\n" + "="*70)
    print("CONSOLIDATED LINGUISTICS HUB (STARTING)")
    print("="*70 + "\n")

    # --- 1. Initialize spaCy Models ---
    print("--- Initializing spaCy Models ---")
    spacy_initialize_models()
    print("--- spaCy Done ---\n")
    
    # --- 2. Initialize OdeNet Worker ---
    print("--- Initializing OdeNet Worker ---")
    if WN_AVAILABLE:
        try:
            odenet_start_worker()
            print("✓ OdeNet worker is starting/ready.")
        except Exception as e:
            print(f"✗ FAILED to start OdeNet worker: {e}")
            print("  'Thesaurus' and 'Comprehensive' tabs may fail.")
    else:
        print("INFO: OdeNet ('wn') library not available, skipping worker.")
    print("--- OdeNet Done ---\n")

    # --- 3. Initialize Wiktionary ---
    print("--- Initializing Wiktionary DB ---")
    try:
        if not wiktionary_download_db():
             print("✗ WARNING: Failed to download Wiktionary DB. Primary engine is disabled.")
        else:
             # Try to pre-warm the connection
             _ = wiktionary_get_connection()
    except Exception as e:
        print(f"✗ FAILED to initialize Wiktionary: {e}")
    print("--- Wiktionary Done ---\n")

    # --- Initialize DWDSmor ---
    print("--- Initializing DWDSmor Lemmatizer ---")
    if DWDSMOR_AVAILABLE:
        try:
            dwdsmor_get_lemmatizer() # Call the function to load the model
        except Exception as e:
            print(f"✗ FAILED to start DWDSmor: {e}")
            print("  'Word Encyclopedia' DWDSmor engine will fail.")
    else:
        print("INFO: DWDSmor library not available, skipping lemmatizer.")
    print("--- DWDSmor Done ---\n")

    # --- 4. Initialize HanTa Tagger ---
    print("--- Initializing HanTa Tagger ---")
    if HANTA_AVAILABLE:
        try:
            hanta_get_tagger() # Call the function to load the model
        except Exception as e:
            print(f"✗ FAILED to start HanTa tagger: {e}")
            print("  'Word Encyclopedia' tab will fail.")
    else:
        print("INFO: HanTa library not available, skipping tagger.")
    print("--- HanTa Done ---\n")

    # --- 54. Check LanguageTool ---
    print("--- Checking LanguageTool ---")
    if not LT_AVAILABLE:
        print("WARNING: language-tool-python not available. 'Grammar' tab will fail.")
    else:
        print("✓ LanguageTool library is available (will lazy-load on first use).")
    print("--- LanguageTool Done ---\n")

    # --- 6. Check Pattern.de ---
    print("--- Checking Pattern.de ---")
    if not PATTERN_DE_AVAILABLE:
        print("WARNING: pattern.de library not available. 'Inflections' tab will fail.")
    else:
        print("✓ Pattern.de library is available.")
    print("--- Pattern.de Done ---\n")
    
    # --- 7. Check Requests (for ConceptNet) ---
    print("--- Checking Requests (for ConceptNet) ---")
    if not REQUESTS_AVAILABLE:
        print("WARNING: requests library not available. 'ConceptNet' features will fail.")
    else:
        print("✓ Requests library is available.")
    print("--- Requests Done ---\n")

    # --- 8. Initialize ConceptNet Client ---
    print("--- Initializing ConceptNet Client ---")
    if GRADIO_CLIENT_AVAILABLE:
        try:
            get_conceptnet_client() # Call the function to load the client
        except Exception as e:
            print(f"✗ FAILED to start ConceptNet Client: {e}")
    else:
        print("INFO: gradio_client not available, skipping ConceptNet client.")
    print("--- ConceptNet Client Done ---\n")

    print("="*70)
    print("All services initialized. Launching Gradio Hub...")
    print("="*70 + "\n")
    
    # --- 9. Launch Gradio ---
    demo = create_consolidated_interface()
    demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)