# -*- coding: utf-8 -*- """Project and strategy audit helpers.""" from __future__ import annotations from collections import Counter from pathlib import Path from alphasift.config import Config from alphasift.filter import requires_daily_features from alphasift.strategy import load_all_strategies def audit_project(strategies_dir: Path | None = None) -> dict[str, object]: """Return a structured audit of current strategy coverage and known gaps.""" if strategies_dir is None: strategies_dir = Config.from_env().strategies_dir strategies = load_all_strategies(strategies_dir) strategy_items = list(strategies.values()) categories = Counter(strategy.category for strategy in strategy_items) profile_fields = [ "scoring_profile", "risk_profile", "portfolio_profile", "scorecard_profile", ] profile_coverage = { field: { "configured": sum(1 for strategy in strategy_items if getattr(strategy.screening, field)), "missing": [ strategy.name for strategy in strategy_items if not getattr(strategy.screening, field) ], } for field in profile_fields } strategy_findings = [] for strategy in strategy_items: strategy_findings.extend(_audit_strategy(strategy)) project_gaps = _known_project_gaps() strengths = _current_strengths() return { "project": "alphasift", "positioning": "全市场候选发现与 LLM 横向排序引擎", "strategy_count": len(strategy_items), "categories": dict(sorted(categories.items())), "profile_coverage": profile_coverage, "strengths": strengths, "strategy_findings": strategy_findings, "project_gaps": project_gaps, "next_priorities": _next_priorities(strategy_findings, project_gaps), } def _audit_strategy(strategy) -> list[dict[str, str]]: findings: list[dict[str, str]] = [] screening = strategy.screening hard_filters = screening.hard_filters uses_daily = requires_daily_features(hard_filters) tags = {str(tag).lower() for tag in strategy.tags} is_shape_strategy = strategy.category in {"trend", "pattern"} or bool( tags & {"breakout", "pullback", "daily_k"} ) if is_shape_strategy and not uses_daily: findings.append({ "severity": "warn", "strategy": strategy.name, "area": "l1_shape_validation", "message": "趋势/形态策略尚未声明日 K 硬条件，可能只是在快照上近似形态。", "recommendation": "补充 require_ma_bullish、price_above_ma20、change_60d 或更严格的形态特征。", }) if not screening.ranking_hints.strip(): findings.append({ "severity": "warn", "strategy": strategy.name, "area": "llm_ranking", "message": "缺少 ranking_hints，LLM 无法获得策略级取舍偏好。", "recommendation": "补充候选排序优先级、风险排除条件和组合分散要求。", }) for field in ("scoring_profile", "risk_profile", "portfolio_profile", "scorecard_profile"): if not getattr(screening, field): findings.append({ "severity": "info", "strategy": strategy.name, "area": "configurability", "message": f"未配置 {field}，将使用默认规则。", "recommendation": "如该策略风格与默认基线不同，应在 YAML 中显式配置。", }) if screening.max_output > 10: findings.append({ "severity": "info", "strategy": strategy.name, "area": "output", "message": "max_output 较大，L3/LLM 输出可能变成列表而不是精选候选。", "recommendation": "保持最终输出精简，更多候选可通过保存 run 和后续评估管理。", }) return findings def _current_strengths() -> list[dict[str, str]]: return [ { "area": "positioning", "message": "定位在全市场候选发现，位于 DSA 单股深度分析和通知助手上游。", }, { "area": "llm", "message": "LLM 只在候选池内做结构化横向排序，支持按 code 注入文件线索或抓取 Top K 新闻/公告/资金流，并输出 thesis、sector、theme、risk、watch_items 和 invalidators。", }, { "area": "risk", "message": "风险层和组合分散层独立于 alpha 排序，扣分字段可审计。", }, { "area": "configurability", "message": "策略 YAML 可覆盖评分曲线、风险阈值、组合风险桶和 scorecard 规则。", }, { "area": "evaluation", "message": "支持保存运行、单次 T+N 最新快照评估、可选日 K 路径最大回撤/最大浮盈，以及按单只和等权组合维度聚合最近 runs。", }, { "area": "daily_shape", "message": "日 K 增强已支持 20 日突破幅度、区间振幅、20 日量能比、实体强度、MA20 回踩距离和平台持续天数。", }, { "area": "data", "message": "行业/概念锚点支持本地映射文件、可选 AkShare 板块反查、板块热度分、主题热度摘要，以及 history JSONL 滚动趋势、持续性、降温、状态回填和异常热度值过滤。", }, { "area": "shape_validation", "message": "T+N 评估会对突破、MA20 回踩和平台整理候选打形态后验标签，用于发现突破延续或失败回落。", }, ] def _known_project_gaps() -> list[dict[str, str]]: return [ { "severity": "gap", "area": "data", "message": "行业/概念和板块热度已能进入 LLM 上下文与 theme_heat 因子，industry-cache 会写 metadata/history sidecar 并回填滚动趋势、持续性和降温信号，但数据源仍偏本地/AkShare。", "recommendation": "补多源映射、板块层级口径归一和更完整的数据质量报告。", }, { "severity": "gap", "area": "news_context", "message": "LLM 上下文支持候选级文件注入、Top K 抓取、缓存、基础去重、来源置信度、来源权重分、压缩摘要、公告类别和粗粒度事件/负面风险标签；策略可配置事件偏好和来源权重，但事件权重尚未进入可学习评估闭环。", "recommendation": "将事件类型偏好纳入 T+N 归因统计，补更细公告分类和来源质量评估。", }, { "severity": "gap", "area": "backtest", "message": "当前 T+N 批量评估覆盖多维聚合、交易成本扣减、可选日 K 路径指标和等权组合摘要，但不是复权、持仓约束和调仓约束完整回测。", "recommendation": "增加持仓约束和组合层面的逐日权益曲线；完整研究后端可接 Qlib/Backtrader。", }, { "severity": "gap", "area": "market_scope", "message": "当前主链只支持 A 股，尚未抽象港股/美股字段映射和交易日历。", "recommendation": "在字段标准化稳定后再扩展多市场，不要先把 L1 规则泛化过度。", }, { "severity": "gap", "area": "shape_validation", "message": "突破、回踩等形态策略已有平台、20 日形态特征、T+N 后验标签和可选日 K 路径指标，但还没有前高压力密集度和盘中失效条件。", "recommendation": "补充前高压力密集度和盘中止损约束。", }, ] def _next_priorities(strategy_findings: list[dict[str, str]], project_gaps: list[dict[str, str]]) -> list[str]: priorities = [ "给行业/概念 history JSONL 增加板块层级口径归一和更完整的数据质量报告。", "将候选上下文事件类型偏好纳入 T+N 归因统计，补更细公告分类和来源质量评估。", "给批量 T+N 评估补持仓约束和组合层面的逐日权益曲线。", ] if any(item["area"] == "l1_shape_validation" for item in strategy_findings): priorities.insert(0, "补强趋势/形态策略的日 K 结构验证，尤其是放量突破类策略。") if any(item["area"] == "backtest" for item in project_gaps): priorities.append("后续再考虑接入 backtrader/qlib 这类完整回测框架，而不是在主链里硬塞。") return priorities