# learning_hub/curator.py
import json
import asyncio
from typing import List, Dict, Any, TYPE_CHECKING
from .schemas import Delta

if TYPE_CHECKING:
    from LLM import LLMService
    from .memory_store import MemoryStore

class Curator:
    def __init__(self, llm_service: 'LLMService', memory_store: 'MemoryStore'):
        self.llm_service = llm_service
        self.memory_store = memory_store
        
        # (This is a configuration parameter from Point 6, not a placeholder)
        self.distill_threshold: int = 50 
        self.distilled_rules_key: str = "learning_distilled_rules.json"
        print("✅ Learning Hub Module: Curator (Distiller) loaded")

    async def check_and_distill_domain(self, domain: str):
        """
        Checks if a domain needs distillation and runs it if the threshold is met.
        (Implements Point 6 - Distillation trigger)
        """
        try:
            deltas_list = await self.memory_store._load_deltas_from_r2(domain)
            
            # 1. Filter for approved Deltas only for distillation
            approved_deltas = [d for d in deltas_list if d.get('approved', False)]
            
            if len(approved_deltas) >= self.distill_threshold:
                print(f"ℹ️ [Curator] Distillation threshold reached for {domain} ({len(approved_deltas)} approved deltas). Starting...")
                await self.distill_deltas(domain, approved_deltas)
            else:
                print(f"ℹ️ [Curator] {domain} has {len(approved_deltas)}/{self.distill_threshold} approved deltas. Distillation not yet required.")
                
        except Exception as e:
            print(f"❌ [Curator] Failed to check distillation for {domain}: {e}")

    async def distill_deltas(self, domain: str, deltas_to_distill: List[Dict]):
        """
        Runs the LLM distillation process to merge and summarize Deltas.
        (Implements Point 4 - Curator (distillation job))
        """
        try:
            # 1. Create the distillation prompt (Now in English)
            prompt = self._create_distillation_prompt(domain, deltas_to_distill)
            
            # 2. Call the LLM
            response_text = await self.llm_service._call_llm(prompt)
            
            if not response_text:
                raise ValueError("Distiller LLM call returned no response.")

            # 3. Parse the response
            distilled_json = self.llm_service._parse_llm_response_enhanced(
                response_text, 
                fallback_strategy="distillation", 
                symbol=domain
            )
            
            if not distilled_json or "distilled_rules" not in distilled_json:
                raise ValueError(f"Failed to parse Distiller LLM response: {response_text}")

            distilled_rules_text_list = distilled_json.get("distilled_rules", [])
            if not isinstance(distilled_rules_text_list, list):
                 raise ValueError(f"Distiller LLM returned 'distilled_rules' not as a list.")

            # 4. Save the new distilled rules
            await self._save_distilled_rules(domain, distilled_rules_text_list, deltas_to_distill)
            
            # 5. Archive (delete) the old approved deltas that were just distilled
            all_deltas = await self.memory_store._load_deltas_from_r2(domain)
            approved_ids_to_archive = {d['id'] for d in deltas_to_distill}
            
            # Keep only non-approved (in-review) deltas, or deltas that weren't part of this batch
            remaining_deltas = [
                d for d in all_deltas 
                if not (d.get('approved', False) and d.get('id') in approved_ids_to_archive)
            ]
            
            await self.memory_store._save_deltas_to_r2(domain, remaining_deltas)
            
            print(f"✅ [Curator] Distillation complete for {domain}. Created {len(distilled_rules_text_list)} new rules. Archived {len(approved_ids_to_archive)} old deltas.")

        except Exception as e:
            print(f"❌ [Curator] Distillation process failed for {domain}: {e}")
            
    async def _save_distilled_rules(self, domain: str, new_rules_text: List[str], evidence_deltas: List[Dict]):
        """Saves the new distilled rules as high-priority Deltas."""
        
        # We save them back into the main delta file as high-priority,
        # so they get picked up by the get_active_context() function.
        
        deltas_list = await self.memory_store._load_deltas_from_r2(domain)
        evidence_ids = [d.get('id', 'N/A') for d in evidence_deltas]
        
        for rule_text in new_rules_text:
            if not rule_text: continue # Skip empty strings
            
            distilled_delta = Delta(
                text=rule_text,
                domain=domain,
                priority="high", # Distilled rules get high priority
                score=0.95,      # High confidence score
                evidence_refs=evidence_ids, # References all the deltas it summarized
                created_by="curator_v1 (distilled)",
                approved=True,   # Automatically approved
                usage_count=0
            )
            deltas_list.append(distilled_delta.model_dump())
            
        await self.memory_store._save_deltas_to_r2(domain, deltas_list)

    def _create_distillation_prompt(self, domain: str, deltas: List[Dict]) -> str:
        """
        Creates the (English-only) prompt for the LLM to act as a Distiller/Curator.
        (Implements Point 4 - Curator prompt)
        """
        
        deltas_text = "\n".join([f"- {d.get('text')} (Score: {d.get('score', 0.5):.2f})" for d in deltas])
        
        prompt = f"""
SYSTEM: You are an expert "Curator" AI. Your job is to read a list of "Deltas" (learning rules) for crypto trading, identify recurring patterns, and merge them into 3-5 concise, powerful "Golden Rules".

DOMAIN: {domain}

RAW DELTAS TO ANALYZE ({len(deltas)} rules):
{deltas_text}
--- END OF DELTAS ---

TASK:
1.  Analyze the "RAW DELTAS" above.
2.  Find overlaps, repetitions, and contradictions.
3.  Generate 3 to 5 new "Distilled Rules" that summarize the core wisdom of these deltas.
4.  Each new rule must be concise (max 25 words) and actionable.

OUTPUT FORMAT (JSON Only):
{{
  "justification": "A brief explanation of the patterns you found and how you merged them.",
  "distilled_rules": [
    "The first golden rule (e.g., 'Always use ATR trailing stops for breakout strategies.')",
    "The second golden rule (e.g., 'If RSI is overbought on 1H, avoid breakout entries.')",
    "..."
  ]
}}
"""
        return prompt