Update src/engine.py

src/engine.py

@@ -3,40 +3,56 @@ import nltk
 from nltk.sentiment.vader import SentimentIntensityAnalyzer
 from spacy.cli import download
 
+
 class DeepPragmaEngine:
+    """
+    Auxiliary lexical/syntactic analyzer.
+
+    NOTE: these heuristics are NOT the primary classifier for sarcasm/hate —
+    that responsibility belongs to GroqClient.classify(), since irony relies
+    on pragmatic context that word-level features can't reliably capture.
+    This class only surfaces supporting signals (sentiment, a naive
+    syntactic "attack" pattern) to log or display alongside the LLM's
+    decision, not to gate it.
+    """
+
+    HATE_TERMS = ["stupid", "useless", "disease", "parasite"]
+    IRONY_MARKERS = ["always", "oh", "sure"]
+
     def __init__(self):
-        # Configuración de NLTK
         try:
-            nltk.data.find('sentiment/vader_lexicon.zip')
+            nltk.data.find("sentiment/vader_lexicon.zip")
         except LookupError:
-            nltk.download('vader_lexicon')
-            
-        # Configuración de SpaCy con autodescarga
+            nltk.download("vader_lexicon")
+
         model_name = "en_core_web_sm"
         try:
             self.nlp = spacy.load(model_name)
         except OSError:
-            print(f"Modelo {model_name} no encontrado. Descargando...")
+            print(f"Model {model_name} not found. Downloading...")
             download(model_name)
             self.nlp = spacy.load(model_name)
-            
+
         self.sia = SentimentIntensityAnalyzer()
-        self.terminos_odio = ["stupid", "useless", "disease", "parasite"]
 
-    def analizar(self, texto):
-        doc = self.nlp(texto)
-        score = self.sia.polarity_scores(texto)
-        
-        ataque = False
+    def get_signals(self, text: str) -> dict:
+        """Compute auxiliary lexical/syntactic signals (informational only)."""
+        doc = self.nlp(text)
+        sentiment = self.sia.polarity_scores(text)
+
+        attack_pattern = False
         for token in doc:
             if token.dep_ == "nsubj" and token.head.lemma_ == "be":
                 for child in token.head.children:
-                    if child.dep_ in ["acomp", "attr"] and child.text.lower() in self.terminos_odio:
-                        ataque = True
-        
-        es_ironico = (score['pos'] > 0.3) and any(w in texto.lower() for w in ["always", "oh", "sure"])
-        
-        if ataque and es_ironico: return "Sarcastic Hate Speech"
-        if ataque: return "Direct Hate Speech"
-        if es_ironico: return "Sarcastic/Ironic (Non-Hateful)"
-        return "Neutral"
+                    if child.dep_ in ("acomp", "attr") and child.text.lower() in self.HATE_TERMS:
+                        attack_pattern = True
+
+        irony_hint = (sentiment["pos"] > 0.3) and any(
+            marker in text.lower() for marker in self.IRONY_MARKERS
+        )
+
+        return {
+            "sentiment": sentiment,
+            "attack_pattern": attack_pattern,
+            "irony_hint": irony_hint,
+        }