aiagent/backend/app/services/insight_analyzer.py

"""
洞察分析器 — 跨项目统计分析，生成最佳实践建议
"""
from __future__ import annotations

import logging
from typing import Dict, List, Any

from sqlalchemy.orm import Session
from app.models.company import CompanyProject

logger = logging.getLogger(__name__)


def analyze_company(company_id: str, db: Session) -> Dict[str, Any]:
    """分析公司所有项目，生成洞察报告。"""
    projects = (
        db.query(CompanyProject)
        .filter(CompanyProject.company_id == company_id)
        .order_by(CompanyProject.created_at.desc())
        .all()
    )

    if not projects:
        return {
            "dept_stats": [],
            "collaboration_pairs": [],
            "avg_rounds": 0,
            "avg_cost": 0,
            "failure_reasons": [],
            "recommendations": ["还没有项目数据，执行一些项目后这里会显示分析结果"],
        }

    total = len(projects)
    completed = [p for p in projects if p.status == "completed"]
    failed = [p for p in projects if p.status == "failed"]
    success_rate = round(len(completed) / total * 100, 1) if total > 0 else 0

    # Department statistics
    dept_stats: Dict[str, Dict] = {}
    for p in projects:
        ceo_plan = p.ceo_plan or {}
        dept_plans = ceo_plan.get("departments", [])
        review_scores = p.review_scores or []

        for dp in dept_plans:
            name = dp.get("department_name", "unknown")
            if name not in dept_stats:
                dept_stats[name] = {"name": name, "execution_count": 0, "success_count": 0, "avg_score": 0, "total_score": 0, "score_count": 0}
            dept_stats[name]["execution_count"] += 1

        # Match scores
        for s in review_scores:
            name = s.get("name", "")
            if name in dept_stats:
                dept_stats[name]["total_score"] += s.get("score", 0)
                dept_stats[name]["score_count"] += 1
                if s.get("pass"):
                    dept_stats[name]["success_count"] += 1

    # Calculate averages
    for k, v in dept_stats.items():
        if v["score_count"] > 0:
            v["avg_score"] = round(v["total_score"] / v["score_count"], 1)
        v["success_rate"] = round(v["success_count"] / v["execution_count"] * 100, 1) if v["execution_count"] > 0 else 0

    # Collaboration pairs (dependencies)
    dep_pairs: Dict[str, int] = {}
    for p in projects:
        ceo_plan = p.ceo_plan or {}
        dept_plans = ceo_plan.get("departments", [])
        for dp in dept_plans:
            deps = dp.get("dependencies", [])
            if isinstance(deps, str):
                deps = [deps] if deps else []
            for d in deps:
                pair_key = f"{d}→{dp.get('department_name', '')}"
                dep_pairs[pair_key] = dep_pairs.get(pair_key, 0) + 1

    top_pairs = sorted(dep_pairs.items(), key=lambda x: x[1], reverse=True)[:5]
    collaboration_pairs = [{"pair": k, "count": v} for k, v in top_pairs]

    # Rounds
    avg_rounds = round(sum(int(p.round_count or "1") for p in projects) / total, 1) if total > 0 else 0

    # Failure reasons (from review scores)
    failure_reasons = []
    for p in projects:
        scores = p.review_scores or []
        for s in scores:
            if not s.get("pass", True):
                failure_reasons.append({
                    "project": p.name,
                    "department": s.get("name", ""),
                    "score": s.get("score", 0),
                    "feedback": s.get("feedback", "")[:200],
                })

    # Recommendations
    recommendations = []
    if success_rate < 50:
        recommendations.append("成功率偏低，建议检查项目目标是否与公司能力匹配")
    if avg_rounds > 1.5:
        recommendations.append(f"平均迭代 {avg_rounds} 轮，建议优化首轮规划质量，减少返工")
    low_depts = [k for k, v in dept_stats.items() if v.get("avg_score", 0) < 6 and v["score_count"] > 0]
    if low_depts:
        recommendations.append(f"以下部门评分偏低，建议优化其 Agent 配置或成员组成: {', '.join(low_depts)}")
    if not recommendations:
        recommendations.append("整体表现良好，继续保持当前的项目执行模式")

    return {
        "total_projects": total,
        "completed_projects": len(completed),
        "failed_projects": len(failed),
        "success_rate": success_rate,
        "dept_stats": list(dept_stats.values()),
        "collaboration_pairs": collaboration_pairs,
        "avg_rounds": avg_rounds,
        "avg_cost": 0,  # Token cost tracking requires execution log integration
        "failure_reasons": failure_reasons[:10],
        "recommendations": recommendations,
    }