"""Collective Learning: aggregates patterns across all internal agents.

Subscribes to the shared memory bus, collects insights from all
Vainplex-internal agents, and builds an aggregated knowledge base
for pattern detection and cross-agent learning.

🚨 STRICT DATA ISOLATION: Only Vainplex-internal agents participate.
No customer data. No customer agent insights. Ever.
"""

from __future__ import annotations

import json
import logging
import os
from collections import defaultdict
from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Any

from cortex.intelligence.shared_memory import ALLOWED_AGENTS, Insight, SharedMemory

logger = logging.getLogger(__name__)


@dataclass
class AggregatedPattern:
    """A pattern detected across multiple agents."""

    topic: str
    description: str
    contributing_agents: list[str]
    confidence: float
    occurrence_count: int
    first_seen: str
    last_seen: str
    insights: list[Insight] = field(default_factory=list)


class CollectiveLearning:
    """Aggregates patterns from all internal agents into collective knowledge.

    Usage:
        collective = CollectiveLearning(shared_memory)
        await collective.start()
        patterns = collective.get_patterns()

    ⚠️ DATA ISOLATION: Only processes insights from ALLOWED_AGENTS.
    """

    def __init__(self, shared_memory: SharedMemory) -> None:
        self.shared_memory = shared_memory
        self._insights_by_topic: dict[str, list[Insight]] = defaultdict(list)
        self._patterns: list[AggregatedPattern] = []

    async def start(self) -> None:
        """Start listening for insights on all topics."""
        await self.shared_memory.subscribe(">", self._handle_insight)
        logger.info("Collective learning started — listening for insights")

    async def _handle_insight(self, insight: Insight) -> None:
        """Process an incoming insight."""
        # Double-check data isolation
        if insight.agent not in ALLOWED_AGENTS:
            logger.warning("Rejected insight from non-internal agent: %s", insight.agent)
            return

        self._insights_by_topic[insight.topic].append(insight)
        logger.debug(
            "Collected insight: %s from %s (topic: %s)",
            insight.content[:60], insight.agent, insight.topic,
        )

        # Re-analyze patterns when new data arrives
        self._detect_patterns()

    def _detect_patterns(self) -> None:
        """Analyze collected insights to find cross-agent patterns."""
        new_patterns: list[AggregatedPattern] = []

        for topic, insights in self._insights_by_topic.items():
            if len(insights) < 2:
                continue

            agents = list({i.agent for i in insights})
            if len(agents) < 2:
                # Single-agent observations aren't "collective" patterns
                continue

            timestamps = sorted(i.timestamp for i in insights)
            avg_confidence = sum(i.confidence for i in insights) / len(insights)

            pattern = AggregatedPattern(
                topic=topic,
                description=f"Cross-agent pattern on '{topic}' observed by {', '.join(agents)}",
                contributing_agents=agents,
                confidence=avg_confidence,
                occurrence_count=len(insights),
                first_seen=timestamps[0],
                last_seen=timestamps[-1],
                insights=insights,
            )
            new_patterns.append(pattern)

        self._patterns = new_patterns

    def get_patterns(
        self,
        topic: str | None = None,
        min_confidence: float = 0.0,
    ) -> list[AggregatedPattern]:
        """Retrieve detected collective patterns.

        Args:
            topic: Filter by topic (optional).
            min_confidence: Minimum confidence threshold.
        """
        patterns = self._patterns
        if topic:
            patterns = [p for p in patterns if p.topic == topic]
        if min_confidence > 0:
            patterns = [p for p in patterns if p.confidence >= min_confidence]
        return patterns

    def get_topic_summary(self) -> dict[str, Any]:
        """Get a summary of all topics and their insight counts."""
        return {
            topic: {
                "count": len(insights),
                "agents": list({i.agent for i in insights}),
                "latest": max(i.timestamp for i in insights) if insights else None,
            }
            for topic, insights in self._insights_by_topic.items()
        }

    def export_knowledge(self) -> str:
        """Export collective knowledge as JSON."""
        return json.dumps({
            "exported_at": datetime.now(timezone.utc).isoformat(),
            "allowed_agents": sorted(ALLOWED_AGENTS),
            "patterns": [
                {
                    "topic": p.topic,
                    "description": p.description,
                    "contributing_agents": p.contributing_agents,
                    "confidence": p.confidence,
                    "occurrence_count": p.occurrence_count,
                    "first_seen": p.first_seen,
                    "last_seen": p.last_seen,
                }
                for p in self._patterns
            ],
            "topics": self.get_topic_summary(),
        }, indent=2)