feat: fase 1 — top_k 30→80, pool 100→300, sin truncado, max_tokens 16000

src/generator/generator.py

@@ -156,7 +156,7 @@ class OutputGenerator:
 
         # RAG: get most relevant context for this output type
         query = self._get_rag_query(output_type, topic)
-        context = await self.processor.rag_query(session_id, query, top_k=30)
+        context = await self.processor.rag_query(session_id, query, top_k=80)
 
         if not context:
             # Fallback: use raw top chunks
@@ -166,11 +166,6 @@ class OutputGenerator:
         if not context:
             raise ValueError("No processed content available. Run /process first.")
 
-        # Truncate context to avoid Ollama context limits
-        context_words = context.split()
-        if len(context_words) > 6000:
-            context = " ".join(context_words[:6000]) + "\n\n[... additional material truncated ...]"
-
         backend = "Claude Haiku" if settings.anthropic_api_key else "Ollama"
         if progress_callback:
             await progress_callback(f"✍️ Generando {output_type} con {backend}... (2-5 min)")
@@ -201,7 +196,7 @@ class OutputGenerator:
     async def _generate_with_claude(self, prompt: str, system: str, output_type: OutputType,
                                      session_id: int | None = None) -> str:
         import anthropic
-        max_tokens = 4096 if output_type == OutputType.THREAD else 8192
+        max_tokens = 4096 if output_type == OutputType.THREAD else 16000
         try:
             client = anthropic.AsyncAnthropic(api_key=settings.anthropic_api_key)
             msg = await client.messages.create(

src/processor/processor.py

@@ -285,7 +285,7 @@ class ContentProcessor:
         query_embedding = await self.ollama.embed(query)
 
         # Get top quality chunks
-        chunks = await self.db.get_top_chunks(session_id, limit=100)
+        chunks = await self.db.get_top_chunks(session_id, limit=300)
 
         if query_embedding and chunks:
             # Rank by embedding similarity