diff --git a/.env.example b/.env.example index 2188c63be..faba6ed66 100644 --- a/.env.example +++ b/.env.example @@ -110,6 +110,9 @@ SEARXNG_BASE_URLS= # =================================== # 新闻时效与分析筛选配置 # =================================== +# 新闻策略窗口档位：ultra_short(1天) / short(3天) / medium(7天) / long(30天) +# 实际窗口 = min(策略窗口, NEWS_MAX_AGE_DAYS) +# NEWS_STRATEGY_PROFILE=short # 新闻最大时效（天），搜索时限制结果在近期内，避免使用过时信息 # NEWS_MAX_AGE_DAYS=3 # 乖离率阈值（%），偏离 MA5 超过此值提示不追高；强势趋势股自动放宽到 1.5 倍 diff --git a/README.md b/README.md index 7e4defb07..c1ff202c7 100644 --- a/README.md +++ b/README.md @@ -164,6 +164,7 @@ | `TUSHARE_TOKEN` | [Tushare Pro](https://tushare.pro/weborder/#/login?reg=834638 ) Token | 可选 | | `PREFETCH_REALTIME_QUOTES` | 实时行情预取开关：设为 `false` 可禁用全市场预取（默认 `true`） | 可选 | | `WECHAT_MSG_TYPE` | 企微消息类型，默认 markdown，支持配置 text 类型，发送纯 markdown 文本 | 可选 | +| `NEWS_STRATEGY_PROFILE` | 新闻策略窗口档位：`ultra_short`(1天) / `short`(3天) / `medium`(7天) / `long`(30天)，默认 `short` | 可选 | | `NEWS_MAX_AGE_DAYS` | 新闻最大时效（天），默认 3，避免使用过时信息 | 可选 | | `BIAS_THRESHOLD` | 乖离率阈值（%），默认 5.0，超过提示不追高；强势趋势股自动放宽 | 可选 | | `AGENT_MODE` | 开启 Agent 策略问股模式（`true`/`false`，默认 false） | 可选 | diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md index f15bfd29f..79061f408 100644 --- a/docs/CHANGELOG.md +++ b/docs/CHANGELOG.md @@ -19,6 +19,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/). ### 修复 - 🐛 **港股 Agent 实时行情去重与快速路由** — 统一 `HK01810` / `1810.HK` / `01810` 等港股代码归一规则；港股实时行情改为直接走单次 `akshare_hk` 路径，避免按 A 股 source priority 重复触发同一失败接口；Agent 运行期对显式 `retriable=false` 的工具失败增加短路缓存，减少同轮分析中的重复失败调用。 +- 📰 **新闻时效硬过滤与策略分窗**（#697）— 新增 `NEWS_STRATEGY_PROFILE`（`ultra_short/short/medium/long`）并与 `NEWS_MAX_AGE_DAYS` 统一计算有效窗口；搜索结果在返回后执行发布时间硬过滤（时间未知剔除、超窗剔除、未来仅容忍 1 天），并在历史 fallback 链路追加相同约束，避免旧闻再次进入“最新动态/风险警报”。 ## [3.7.0] - 2026-03-15 diff --git a/docs/full-guide.md b/docs/full-guide.md index 900b60250..a6e7d36cd 100644 --- a/docs/full-guide.md +++ b/docs/full-guide.md @@ -228,6 +228,7 @@ daily_stock_analysis/ | `BRAVE_API_KEYS` | Brave Search API Key（美股优化） | 可选 | | `SERPAPI_API_KEYS` | SerpAPI 备用搜索 | 可选 | | `SEARXNG_BASE_URLS` | SearXNG 自建实例（无配额兜底，需在 settings.yml 启用 format: json） | 可选 | +| `NEWS_STRATEGY_PROFILE` | 新闻策略窗口档位：`ultra_short`(1天)/`short`(3天)/`medium`(7天)/`long`(30天)；实际窗口取与 `NEWS_MAX_AGE_DAYS` 的最小值 | 默认 `short` | | `NEWS_MAX_AGE_DAYS` | 新闻最大时效（天），搜索时限制结果在近期内 | 默认 `3` | | `BIAS_THRESHOLD` | 乖离率阈值（%），超过提示不追高；强势趋势股自动放宽到 1.5 倍 | 默认 `5.0` | diff --git a/main.py b/main.py index 40d6ab32c..48689e635 100644 --- a/main.py +++ b/main.py @@ -642,6 +642,7 @@ def main() -> int: minimax_keys=config.minimax_api_keys, searxng_base_urls=config.searxng_base_urls, news_max_age_days=config.news_max_age_days, + news_strategy_profile=getattr(config, "news_strategy_profile", "short"), ) if config.gemini_api_key or config.openai_api_key: diff --git a/src/analyzer.py b/src/analyzer.py index 435ca9e93..c8b221588 100644 --- a/src/analyzer.py +++ b/src/analyzer.py @@ -22,7 +22,14 @@ from json_repair import repair_json from litellm import Router from src.agent.llm_adapter import get_thinking_extra_body -from src.config import Config, get_config, get_api_keys_for_model, extra_litellm_params, get_configured_llm_models +from src.config import ( + Config, + extra_litellm_params, + get_api_keys_for_model, + get_config, + get_configured_llm_models, + resolve_news_window_days, +) from src.storage import persist_llm_usage from src.data.stock_mapping import STOCK_NAME_MAP from src.schemas.report_schema import AnalysisReportSchema @@ -1161,6 +1168,22 @@ class GeminiAnalyzer: """ # 添加新闻搜索结果（重点区域） + news_window_days: Optional[int] = None + context_window = context.get("news_window_days") + try: + if context_window is not None: + parsed_window = int(context_window) + if parsed_window > 0: + news_window_days = parsed_window + except (TypeError, ValueError): + news_window_days = None + + if news_window_days is None: + prompt_config = get_config() + news_window_days = resolve_news_window_days( + news_max_age_days=getattr(prompt_config, "news_max_age_days", 3), + news_strategy_profile=getattr(prompt_config, "news_strategy_profile", "short"), + ) prompt += """ --- @@ -1168,10 +1191,14 @@ class GeminiAnalyzer: """ if news_context: prompt += f""" -以下是 **{stock_name}({code})** 近7日的新闻搜索结果，请重点提取： +以下是 **{stock_name}({code})** 近{news_window_days}日的新闻搜索结果，请重点提取： 1. 🚨 **风险警报**：减持、处罚、利空 2. 🎯 **利好催化**：业绩、合同、政策 3. 📊 **业绩预期**：年报预告、业绩快报 +4. 🕒 **时间规则（强制）**： + - 输出到 `risk_alerts` / `positive_catalysts` / `latest_news` 的每一条都必须带具体日期（YYYY-MM-DD） + - 超出近{news_window_days}日窗口的新闻一律忽略 + - 时间未知、无法确定发布日期的新闻一律忽略 ``` {news_context} @@ -1226,6 +1253,7 @@ class GeminiAnalyzer: - **持仓分类建议**：空仓者怎么做 vs 持仓者怎么做 - **具体狙击点位**：买入价、止损价、目标价（精确到分） - **检查清单**：每项用 ✅/⚠️/❌ 标记 +- **消息面时间合规**：`latest_news`、`risk_alerts`、`positive_catalysts` 不得包含超出近{news_window_days}日或时间未知的信息 请输出完整的 JSON 格式决策仪表盘。""" diff --git a/src/config.py b/src/config.py index 5667a67a5..73ea8952b 100644 --- a/src/config.py +++ b/src/config.py @@ -43,6 +43,12 @@ class ConfigIssue: _MANAGED_LITELLM_KEY_PROVIDERS = {"gemini", "vertex_ai", "anthropic", "openai", "deepseek"} SUPPORTED_LLM_CHANNEL_PROTOCOLS = ("openai", "anthropic", "gemini", "vertex_ai", "deepseek", "ollama") _FALSEY_ENV_VALUES = {"0", "false", "no", "off"} +NEWS_STRATEGY_WINDOWS: Dict[str, int] = { + "ultra_short": 1, + "short": 3, + "medium": 7, + "long": 30, +} def parse_env_bool(value: Optional[str], default: bool = False) -> bool: @@ -55,6 +61,19 @@ def parse_env_bool(value: Optional[str], default: bool = False) -> bool: return normalized not in _FALSEY_ENV_VALUES +def normalize_news_strategy_profile(value: Optional[str]) -> str: + """Normalize news strategy profile to known values.""" + candidate = (value or "short").strip().lower() + return candidate if candidate in NEWS_STRATEGY_WINDOWS else "short" + + +def resolve_news_window_days(news_max_age_days: int, news_strategy_profile: Optional[str]) -> int: + """Resolve effective news window days from profile and global max-age.""" + profile = normalize_news_strategy_profile(news_strategy_profile) + profile_days = NEWS_STRATEGY_WINDOWS.get(profile, NEWS_STRATEGY_WINDOWS["short"]) + return max(1, min(max(1, int(news_max_age_days)), profile_days)) + + def canonicalize_llm_channel_protocol(value: Optional[str]) -> str: """Normalize a protocol label into a LiteLLM provider identifier.""" candidate = (value or "").strip().lower().replace("-", "_") @@ -335,6 +354,7 @@ class Config: # === 新闻与分析筛选配置 === news_max_age_days: int = 3 # 新闻最大时效（天） + news_strategy_profile: str = "short" # 新闻窗口策略档位：ultra_short/short/medium/long bias_threshold: float = 5.0 # 乖离率阈值（%），超过此值提示不追高 # === Agent 模式配置 === @@ -896,6 +916,9 @@ class Config: social_sentiment_api_key=os.getenv('SOCIAL_SENTIMENT_API_KEY') or None, social_sentiment_api_url=os.getenv('SOCIAL_SENTIMENT_API_URL', 'https://api.adanos.org').rstrip('/'), news_max_age_days=max(1, int(os.getenv('NEWS_MAX_AGE_DAYS', '3'))), + news_strategy_profile=cls._parse_news_strategy_profile( + os.getenv('NEWS_STRATEGY_PROFILE', 'short') + ), bias_threshold=max(1.0, float(os.getenv('BIAS_THRESHOLD', '5.0'))), agent_mode=os.getenv('AGENT_MODE', 'false').lower() == 'true', _agent_mode_explicit=os.getenv('AGENT_MODE') is not None, @@ -1310,6 +1333,26 @@ class Config: ) return 'simple' + @classmethod + def _parse_news_strategy_profile(cls, value: Optional[str]) -> str: + """Parse NEWS_STRATEGY_PROFILE, fallback to short for invalid values.""" + normalized = normalize_news_strategy_profile(value) + raw = (value or "short").strip().lower() + if raw != normalized: + logging.getLogger(__name__).warning( + "NEWS_STRATEGY_PROFILE '%s' invalid, fallback to 'short' " + "(valid: ultra_short/short/medium/long)", + value, + ) + return normalized + + def get_effective_news_window_days(self) -> int: + """Return effective news window days after profile + max-age merge.""" + return resolve_news_window_days( + news_max_age_days=self.news_max_age_days, + news_strategy_profile=self.news_strategy_profile, + ) + @classmethod def _parse_market_review_region(cls, value: str) -> str: """解析大盘复盘市场区域，非法值记录警告后回退为 cn""" diff --git a/src/core/config_registry.py b/src/core/config_registry.py index 8010b7761..7b3db21cd 100644 --- a/src/core/config_registry.py +++ b/src/core/config_registry.py @@ -372,6 +372,20 @@ _FIELD_DEFINITIONS: Dict[str, Dict[str, Any]] = { "validation": {"min": 1, "max": 30}, "display_order": 60, }, + "NEWS_STRATEGY_PROFILE": { + "title": "News Strategy Profile", + "description": "News window profile: ultra_short(1d), short(3d), medium(7d), long(30d). Effective window = min(profile, NEWS_MAX_AGE_DAYS).", + "category": "data_source", + "data_type": "string", + "ui_control": "select", + "is_sensitive": False, + "is_required": False, + "is_editable": True, + "default_value": "short", + "options": ["ultra_short", "short", "medium", "long"], + "validation": {"enum": ["ultra_short", "short", "medium", "long"]}, + "display_order": 61, + }, "BIAS_THRESHOLD": { "title": "Bias Threshold (%)", "description": "Deviation threshold from MA5 (%). Exceeding this triggers 'do not chase' warning. Strong trend stocks auto-widen to 1.5x.", @@ -384,7 +398,7 @@ _FIELD_DEFINITIONS: Dict[str, Dict[str, Any]] = { "default_value": "5.0", "options": [], "validation": {"min": 0.0, "max": 50.0}, - "display_order": 61, + "display_order": 62, }, "PYTDX_HOST": { "title": "Pytdx Host", diff --git a/src/core/pipeline.py b/src/core/pipeline.py index e4bba48d7..0fe82344c 100644 --- a/src/core/pipeline.py +++ b/src/core/pipeline.py @@ -91,6 +91,7 @@ class StockAnalysisPipeline: serpapi_keys=self.config.serpapi_keys, minimax_keys=self.config.minimax_api_keys, news_max_age_days=self.config.news_max_age_days, + news_strategy_profile=getattr(self.config, "news_strategy_profile", "short"), ) logger.info(f"调度器初始化完成，最大并发数: {self.max_workers}") @@ -450,6 +451,9 @@ class StockAnalysisPipeline: enhanced['stock_name'] = stock_name elif realtime_quote and getattr(realtime_quote, 'name', None): enhanced['stock_name'] = realtime_quote.name + + # 将运行时搜索窗口透传给 analyzer，避免与全局配置重新读取产生窗口不一致 + enhanced['news_window_days'] = getattr(self.search_service, "news_window_days", 3) # 添加实时行情（兼容不同数据源的字段差异） if realtime_quote: diff --git a/src/search_service.py b/src/search_service.py index c5302e6fb..6cf43361c 100644 --- a/src/search_service.py +++ b/src/search_service.py @@ -13,10 +13,12 @@ A股自选股智能分析系统 - 搜索服务模块 import logging import random +import re import time from abc import ABC, abstractmethod -from dataclasses import dataclass, field -from datetime import datetime +from dataclasses import dataclass +from datetime import date, datetime, timedelta, timezone +from email.utils import parsedate_to_datetime from typing import List, Dict, Any, Optional, Tuple from itertools import cycle import requests @@ -30,6 +32,11 @@ from tenacity import ( ) from data_provider.us_index_mapping import is_us_index_code +from src.config import ( + NEWS_STRATEGY_WINDOWS, + normalize_news_strategy_profile, + resolve_news_window_days, +) logger = logging.getLogger(__name__) @@ -1338,6 +1345,9 @@ class SearchService: "{name} technical analysis", "{name} {code} performance volume", ] + NEWS_OVERSAMPLE_FACTOR = 2 + NEWS_OVERSAMPLE_MAX = 10 + FUTURE_TOLERANCE_DAYS = 1 def __init__( self, @@ -1348,6 +1358,7 @@ class SearchService: minimax_keys: Optional[List[str]] = None, searxng_base_urls: Optional[List[str]] = None, news_max_age_days: int = 3, + news_strategy_profile: str = "short", ): """ 初始化搜索服务 @@ -1360,9 +1371,25 @@ class SearchService: minimax_keys: MiniMax API Key 列表 searxng_base_urls: SearXNG 实例地址列表（自建无配额兜底） news_max_age_days: 新闻最大时效（天） + news_strategy_profile: 新闻窗口策略档位（ultra_short/short/medium/long） """ self._providers: List[BaseSearchProvider] = [] self.news_max_age_days = max(1, news_max_age_days) + raw_profile = (news_strategy_profile or "short").strip().lower() + self.news_strategy_profile = normalize_news_strategy_profile(news_strategy_profile) + if raw_profile != self.news_strategy_profile: + logger.warning( + "NEWS_STRATEGY_PROFILE '%s' 无效，已回退为 'short'", + news_strategy_profile, + ) + self.news_window_days = resolve_news_window_days( + news_max_age_days=self.news_max_age_days, + news_strategy_profile=self.news_strategy_profile, + ) + self.news_profile_days = NEWS_STRATEGY_WINDOWS.get( + self.news_strategy_profile, + NEWS_STRATEGY_WINDOWS["short"], + ) # 初始化搜索引擎（按优先级排序） # 1. Bocha 优先（中文搜索优化，AI摘要） @@ -1402,6 +1429,13 @@ class SearchService: self._cache: Dict[str, Tuple[float, 'SearchResponse']] = {} # Default cache TTL in seconds (10 minutes) self._cache_ttl: int = 600 + logger.info( + "新闻时效策略已启用: profile=%s, profile_days=%s, NEWS_MAX_AGE_DAYS=%s, effective_window=%s", + self.news_strategy_profile, + self.news_profile_days, + self.news_max_age_days, + self.news_window_days, + ) @staticmethod def _is_foreign_stock(stock_code: str) -> bool: @@ -1482,6 +1516,228 @@ class SearchService: for k in oldest: del self._cache[k] self._cache[key] = (time.time(), response) + + def _effective_news_window_days(self) -> int: + """Resolve effective news window from strategy profile and global max-age.""" + return resolve_news_window_days( + news_max_age_days=self.news_max_age_days, + news_strategy_profile=self.news_strategy_profile, + ) + + @classmethod + def _provider_request_size(cls, max_results: int) -> int: + """Apply light overfetch before time filtering to avoid sparse outputs.""" + target = max(1, int(max_results)) + return max(target, min(target * cls.NEWS_OVERSAMPLE_FACTOR, cls.NEWS_OVERSAMPLE_MAX)) + + @staticmethod + def _parse_relative_news_date(text: str, now: datetime) -> Optional[date]: + """Parse common Chinese/English relative-time strings.""" + raw = (text or "").strip() + if not raw: + return None + + lower = raw.lower() + if raw in {"今天", "今日", "刚刚"} or lower in {"today", "just now", "now"}: + return now.date() + if raw == "昨天" or lower == "yesterday": + return (now - timedelta(days=1)).date() + if raw == "前天": + return (now - timedelta(days=2)).date() + + zh = re.match(r"^\s*(\d+)\s*(分钟|小时|天|周|个月|月|年)\s*前\s*$", raw) + if zh: + amount = int(zh.group(1)) + unit = zh.group(2) + if unit == "分钟": + return (now - timedelta(minutes=amount)).date() + if unit == "小时": + return (now - timedelta(hours=amount)).date() + if unit == "天": + return (now - timedelta(days=amount)).date() + if unit == "周": + return (now - timedelta(weeks=amount)).date() + if unit in {"个月", "月"}: + return (now - timedelta(days=amount * 30)).date() + if unit == "年": + return (now - timedelta(days=amount * 365)).date() + + en = re.match( + r"^\s*(\d+)\s*(minute|minutes|min|mins|hour|hours|day|days|week|weeks|month|months|year|years)\s*ago\s*$", + lower, + ) + if en: + amount = int(en.group(1)) + unit = en.group(2) + if unit in {"minute", "minutes", "min", "mins"}: + return (now - timedelta(minutes=amount)).date() + if unit in {"hour", "hours"}: + return (now - timedelta(hours=amount)).date() + if unit in {"day", "days"}: + return (now - timedelta(days=amount)).date() + if unit in {"week", "weeks"}: + return (now - timedelta(weeks=amount)).date() + if unit in {"month", "months"}: + return (now - timedelta(days=amount * 30)).date() + if unit in {"year", "years"}: + return (now - timedelta(days=amount * 365)).date() + + return None + + @classmethod + def _normalize_news_publish_date(cls, value: Any) -> Optional[date]: + """Normalize provider date value into a date object.""" + if value is None: + return None + if isinstance(value, datetime): + if value.tzinfo is not None: + local_tz = datetime.now().astimezone().tzinfo or timezone.utc + return value.astimezone(local_tz).date() + return value.date() + if isinstance(value, date): + return value + + text = str(value).strip() + if not text: + return None + now = datetime.now() + local_tz = now.astimezone().tzinfo or timezone.utc + + relative_date = cls._parse_relative_news_date(text, now) + if relative_date: + return relative_date + + # Unix timestamp fallback + if text.isdigit() and len(text) in (10, 13): + try: + ts = int(text[:10]) if len(text) == 13 else int(text) + # Provider timestamps are typically UTC epoch seconds. + # Normalize to local date to keep window checks aligned with local "today". + return datetime.fromtimestamp(ts, tz=timezone.utc).astimezone(local_tz).date() + except (OSError, OverflowError, ValueError): + pass + + iso_candidate = text.replace("Z", "+00:00") + try: + parsed_iso = datetime.fromisoformat(iso_candidate) + if parsed_iso.tzinfo is not None: + return parsed_iso.astimezone(local_tz).date() + return parsed_iso.date() + except ValueError: + pass + + normalized = re.sub(r"(\d+)(st|nd|rd|th)", r"\1", text, flags=re.IGNORECASE) + + try: + parsed_rfc = parsedate_to_datetime(normalized) + if parsed_rfc: + if parsed_rfc.tzinfo is not None: + return parsed_rfc.astimezone(local_tz).date() + return parsed_rfc.date() + except (TypeError, ValueError): + pass + + zh_match = re.search(r"(\d{4})\s*[年/\-.]\s*(\d{1,2})\s*[月/\-.]\s*(\d{1,2})\s*日?", text) + if zh_match: + try: + return date(int(zh_match.group(1)), int(zh_match.group(2)), int(zh_match.group(3))) + except ValueError: + pass + + for fmt in ( + "%Y-%m-%d %H:%M:%S", + "%Y-%m-%d %H:%M", + "%Y-%m-%d", + "%Y/%m/%d %H:%M:%S", + "%Y/%m/%d %H:%M", + "%Y/%m/%d", + "%Y.%m.%d %H:%M:%S", + "%Y.%m.%d %H:%M", + "%Y.%m.%d", + "%Y%m%d", + "%b %d, %Y", + "%B %d, %Y", + "%d %b %Y", + "%d %B %Y", + "%a, %d %b %Y %H:%M:%S %z", + ): + try: + parsed_dt = datetime.strptime(normalized, fmt) + if parsed_dt.tzinfo is not None: + return parsed_dt.astimezone(local_tz).date() + return parsed_dt.date() + except ValueError: + continue + + return None + + def _filter_news_response( + self, + response: SearchResponse, + *, + search_days: int, + max_results: int, + log_scope: str, + ) -> SearchResponse: + """Hard-filter results by published_date recency and normalize date strings.""" + if not response.success or not response.results: + return response + + today = datetime.now().date() + earliest = today - timedelta(days=max(0, int(search_days) - 1)) + latest = today + timedelta(days=self.FUTURE_TOLERANCE_DAYS) + + filtered: List[SearchResult] = [] + dropped_unknown = 0 + dropped_old = 0 + dropped_future = 0 + + for item in response.results: + published = self._normalize_news_publish_date(item.published_date) + if published is None: + dropped_unknown += 1 + continue + if published < earliest: + dropped_old += 1 + continue + if published > latest: + dropped_future += 1 + continue + + filtered.append( + SearchResult( + title=item.title, + snippet=item.snippet, + url=item.url, + source=item.source, + published_date=published.isoformat(), + ) + ) + if len(filtered) >= max_results: + break + + if dropped_unknown or dropped_old or dropped_future: + logger.info( + "[新闻过滤] %s: provider=%s, total=%s, kept=%s, drop_unknown=%s, drop_old=%s, drop_future=%s, window=[%s,%s]", + log_scope, + response.provider, + len(response.results), + len(filtered), + dropped_unknown, + dropped_old, + dropped_future, + earliest.isoformat(), + latest.isoformat(), + ) + + return SearchResponse( + query=response.query, + results=filtered, + provider=response.provider, + success=response.success, + error_message=response.error_message, + search_time=response.search_time, + ) def search_stock_news( self, @@ -1502,20 +1758,10 @@ class SearchService: Returns: SearchResponse 对象 """ - # 智能确定搜索时间范围 - # 策略： - # 1. 周二至周五：搜索近1天（24小时） - # 2. 周六、周日：搜索近2-3天（覆盖周末） - # 3. 周一：搜索近3天（覆盖周末） - # 4. 用 NEWS_MAX_AGE_DAYS 限制上限 - today_weekday = datetime.now().weekday() - if today_weekday == 0: # 周一 - weekday_days = 3 - elif today_weekday >= 5: # 周六(5)、周日(6) - weekday_days = 2 - else: # 周二(1) - 周五(4) - weekday_days = 1 - search_days = min(weekday_days, self.news_max_age_days) + # 策略窗口优先：ultra_short/short/medium/long = 1/3/7/30 天， + # 并统一受 NEWS_MAX_AGE_DAYS 上限约束。 + search_days = self._effective_news_window_days() + provider_max_results = self._provider_request_size(max_results) # 构建搜索查询（优化搜索效果） is_foreign = self._is_foreign_stock(stock_code) @@ -1529,7 +1775,15 @@ class SearchService: # 默认主查询：股票名称 + 核心关键词 query = f"{stock_name} {stock_code} 股票 最新消息" - logger.info(f"搜索股票新闻: {stock_name}({stock_code}), query='{query}', 时间范围: 近{search_days}天") + logger.info( + "搜索股票新闻: %s(%s), query='%s', 时间范围: 近%s天, 目标条数=%s, provider请求条数=%s", + stock_name, + stock_code, + query, + search_days, + max_results, + provider_max_results, + ) # Check cache first cache_key = self._cache_key(query, max_results, search_days) @@ -1538,19 +1792,46 @@ class SearchService: logger.info(f"使用缓存搜索结果: {stock_name}({stock_code})") return cached - # 依次尝试各个搜索引擎 + # 依次尝试各个搜索引擎（若过滤后为空，继续尝试下一引擎） + had_provider_success = False for provider in self._providers: if not provider.is_available: continue - response = provider.search(query, max_results, days=search_days) - - if response.success and response.results: + response = provider.search(query, provider_max_results, days=search_days) + filtered_response = self._filter_news_response( + response, + search_days=search_days, + max_results=max_results, + log_scope=f"{stock_code}:{provider.name}:stock_news", + ) + had_provider_success = had_provider_success or bool(response.success) + + if filtered_response.success and filtered_response.results: logger.info(f"使用 {provider.name} 搜索成功") - self._put_cache(cache_key, response) - return response + self._put_cache(cache_key, filtered_response) + return filtered_response else: - logger.warning(f"{provider.name} 搜索失败: {response.error_message}，尝试下一个引擎") + if response.success and not filtered_response.results: + logger.info( + "%s 搜索成功但过滤后无有效新闻，继续尝试下一引擎", + provider.name, + ) + else: + logger.warning( + "%s 搜索失败: %s，尝试下一个引擎", + provider.name, + response.error_message, + ) + + if had_provider_success: + return SearchResponse( + query=query, + results=[], + provider="Filtered", + success=True, + error_message=None, + ) # 所有引擎都失败 return SearchResponse( @@ -1673,7 +1954,18 @@ class SearchService: ), 'desc': '行业分析'}, ] - logger.info(f"开始多维度情报搜索: {stock_name}({stock_code})") + search_days = self._effective_news_window_days() + target_per_dimension = 3 + provider_max_results = self._provider_request_size(target_per_dimension) + + logger.info( + "开始多维度情报搜索: %s(%s), 时间范围: 近%s天, 目标条数=%s, provider请求条数=%s", + stock_name, + stock_code, + search_days, + target_per_dimension, + provider_max_results, + ) # 轮流使用不同的搜索引擎 provider_index = 0 @@ -1692,12 +1984,27 @@ class SearchService: logger.info(f"[情报搜索] {dim['desc']}: 使用 {provider.name}") - response = provider.search(dim['query'], max_results=3, days=self.news_max_age_days) - results[dim['name']] = response + response = provider.search( + dim['query'], + max_results=provider_max_results, + days=search_days, + ) + filtered_response = self._filter_news_response( + response, + search_days=search_days, + max_results=target_per_dimension, + log_scope=f"{stock_code}:{provider.name}:{dim['name']}", + ) + results[dim['name']] = filtered_response search_count += 1 if response.success: - logger.info(f"[情报搜索] {dim['desc']}: 获取 {len(response.results)} 条结果") + logger.info( + "[情报搜索] %s: 原始=%s条, 过滤后=%s条", + dim['desc'], + len(response.results), + len(filtered_response.results), + ) else: logger.warning(f"[情报搜索] {dim['desc']}: 搜索失败 - {response.error_message}") @@ -1978,6 +2285,7 @@ def get_search_service() -> SearchService: minimax_keys=config.minimax_api_keys, searxng_base_urls=config.searxng_base_urls, news_max_age_days=config.news_max_age_days, + news_strategy_profile=getattr(config, "news_strategy_profile", "short"), ) return _search_service diff --git a/src/services/history_service.py b/src/services/history_service.py index f31f59cc5..cd4f93fe4 100644 --- a/src/services/history_service.py +++ b/src/services/history_service.py @@ -12,9 +12,10 @@ Responsibilities: from __future__ import annotations import json import logging -from datetime import datetime, timedelta +from datetime import date, datetime, timedelta from typing import Optional, Dict, Any, List, Tuple, TYPE_CHECKING +from src.config import get_config, resolve_news_window_days from src.storage import DatabaseManager from src.utils.data_processing import normalize_model_used, parse_json_field @@ -381,7 +382,31 @@ class HistoryService: if item.fetched_at and start_time <= item.fetched_at <= end_time ] - return matched[:limit] + # 历史兜底链路也做发布时间硬过滤，避免旧库脏数据重新冒出。 + cfg = get_config() + window_days = resolve_news_window_days( + news_max_age_days=getattr(cfg, "news_max_age_days", 3), + news_strategy_profile=getattr(cfg, "news_strategy_profile", "short"), + ) + # Anchor to analysis date instead of "today" to preserve historical context. + anchor_date = analysis.created_at.date() + latest_allowed = anchor_date + timedelta(days=1) + earliest_allowed = anchor_date - timedelta(days=max(0, window_days - 1)) + + filtered = [] + for item in matched: + if not item.published_date: + continue + if isinstance(item.published_date, datetime): + published = item.published_date.date() + elif isinstance(item.published_date, date): + published = item.published_date + else: + continue + if earliest_allowed <= published <= latest_allowed: + filtered.append(item) + + return filtered[:limit] def _get_sentiment_label(self, score: int) -> str: """ diff --git a/src/services/name_to_code_resolver.py b/src/services/name_to_code_resolver.py index 6c4214666..de8d409cd 100644 --- a/src/services/name_to_code_resolver.py +++ b/src/services/name_to_code_resolver.py @@ -86,6 +86,19 @@ def _get_akshare_name_to_code() -> Optional[Dict[str, str]]: return None +def _is_single_char_typo(input_name: str, candidate_name: str) -> bool: + """Return True when two names only differ by one character position.""" + if not input_name or not candidate_name: + return False + if len(input_name) != len(candidate_name): + return False + # Keep typo fallback conservative: only for names with enough signal. + if len(input_name) < 3: + return False + diff = sum(1 for a, b in zip(input_name, candidate_name) if a != b) + return diff == 1 + + def resolve_name_to_code(name: str) -> Optional[str]: """ Resolve stock name to code. @@ -147,10 +160,19 @@ def resolve_name_to_code(name: str) -> Optional[str]: # e.g. '中国' matching arbitrary company names in a pool of 5000+ stocks. # Use a higher cutoff (0.8) to reduce mis-hits on longer inputs as well. if len(s) > 2: - matches = difflib.get_close_matches(s, list(all_name_to_code.keys()), n=1, cutoff=0.8) + names = list(all_name_to_code.keys()) + matches = difflib.get_close_matches(s, names, n=1, cutoff=0.8) if matches: logger.debug(f"[NameResolver] 命中模糊匹配: input={s}, matched={matches[0]}") return all_name_to_code[matches[0]] + # Conservative fallback for one-character typo in medium/long names. + # This keeps the strict default threshold while fixing obvious misspellings + # such as "贵州茅苔" -> "贵州茅台". + typo_matches = difflib.get_close_matches(s, names, n=1, cutoff=0.7) + if typo_matches and _is_single_char_typo(s, typo_matches[0]): + logger.debug(f"[NameResolver] 命中单字误写兜底: input={s}, matched={typo_matches[0]}") + return all_name_to_code[typo_matches[0]] + logger.debug(f"[NameResolver] 解析失败: {s}") return None diff --git a/tests/test_analyzer_news_prompt.py b/tests/test_analyzer_news_prompt.py new file mode 100644 index 000000000..36938b794 --- /dev/null +++ b/tests/test_analyzer_news_prompt.py @@ -0,0 +1,63 @@ +# -*- coding: utf-8 -*- +"""Tests for analyzer news prompt hard constraints (Issue #697).""" + +import sys +import unittest +from types import SimpleNamespace +from unittest.mock import MagicMock, patch + +try: + import litellm # noqa: F401 +except ModuleNotFoundError: + sys.modules["litellm"] = MagicMock() + +from src.analyzer import GeminiAnalyzer + + +class AnalyzerNewsPromptTestCase(unittest.TestCase): + def test_prompt_contains_time_constraints(self) -> None: + with patch.object(GeminiAnalyzer, "_init_litellm", return_value=None): + analyzer = GeminiAnalyzer() + + context = { + "code": "600519", + "stock_name": "贵州茅台", + "date": "2026-03-16", + "today": {}, + } + fake_cfg = SimpleNamespace( + news_max_age_days=30, + news_strategy_profile="medium", # 7 days + ) + with patch("src.analyzer.get_config", return_value=fake_cfg): + prompt = analyzer._format_prompt(context, "贵州茅台", news_context="news") + + self.assertIn("近7日的新闻搜索结果", prompt) + self.assertIn("每一条都必须带具体日期（YYYY-MM-DD）", prompt) + self.assertIn("超出近7日窗口的新闻一律忽略", prompt) + self.assertIn("时间未知、无法确定发布日期的新闻一律忽略", prompt) + + def test_prompt_prefers_context_news_window_days(self) -> None: + with patch.object(GeminiAnalyzer, "_init_litellm", return_value=None): + analyzer = GeminiAnalyzer() + + context = { + "code": "600519", + "stock_name": "贵州茅台", + "date": "2026-03-16", + "today": {}, + "news_window_days": 1, + } + fake_cfg = SimpleNamespace( + news_max_age_days=30, + news_strategy_profile="long", # 30 days if fallback is used + ) + with patch("src.analyzer.get_config", return_value=fake_cfg): + prompt = analyzer._format_prompt(context, "贵州茅台", news_context="news") + + self.assertIn("近1日的新闻搜索结果", prompt) + self.assertIn("超出近1日窗口的新闻一律忽略", prompt) + + +if __name__ == "__main__": + unittest.main() diff --git a/tests/test_history_news_fallback.py b/tests/test_history_news_fallback.py new file mode 100644 index 000000000..1c4c48711 --- /dev/null +++ b/tests/test_history_news_fallback.py @@ -0,0 +1,77 @@ +# -*- coding: utf-8 -*- +"""Tests for history fallback published_date hard filtering (Issue #697).""" + +import unittest +from datetime import datetime, timedelta +from types import SimpleNamespace +from unittest.mock import MagicMock, patch + +from src.services.history_service import HistoryService + + +class HistoryNewsFallbackTestCase(unittest.TestCase): + def test_fallback_filters_by_published_date_window(self) -> None: + now = datetime.now() + analysis = SimpleNamespace(code="600519", created_at=now) + + # All entries are within fetched_at window; only one should pass published_date window. + candidates = [ + SimpleNamespace( + fetched_at=now, + published_date=now - timedelta(days=20), # too old + title="old", + ), + SimpleNamespace( + fetched_at=now, + published_date=None, # unknown -> drop + title="unknown", + ), + SimpleNamespace( + fetched_at=now, + published_date=now - timedelta(days=1), # valid + title="fresh", + ), + ] + + mock_db = MagicMock() + mock_db.get_analysis_history.return_value = [analysis] + mock_db.get_recent_news.return_value = candidates + + svc = HistoryService(db_manager=mock_db) + fake_cfg = SimpleNamespace(news_max_age_days=30, news_strategy_profile="short") + with patch("src.services.history_service.get_config", return_value=fake_cfg): + result = svc._fallback_news_by_analysis_context("q-1", limit=20) + + self.assertEqual([item.title for item in result], ["fresh"]) + + def test_fallback_uses_analysis_date_as_window_anchor(self) -> None: + analysis_time = datetime.now() - timedelta(days=40) + analysis = SimpleNamespace(code="600519", created_at=analysis_time) + + candidates = [ + SimpleNamespace( + fetched_at=analysis_time, + published_date=analysis_time - timedelta(days=10), # too old for short profile + title="too_old_for_analysis_window", + ), + SimpleNamespace( + fetched_at=analysis_time, + published_date=analysis_time - timedelta(days=1), # valid around analysis date + title="valid_near_analysis_date", + ), + ] + + mock_db = MagicMock() + mock_db.get_analysis_history.return_value = [analysis] + mock_db.get_recent_news.return_value = candidates + + svc = HistoryService(db_manager=mock_db) + fake_cfg = SimpleNamespace(news_max_age_days=30, news_strategy_profile="short") + with patch("src.services.history_service.get_config", return_value=fake_cfg): + result = svc._fallback_news_by_analysis_context("q-1", limit=20) + + self.assertEqual([item.title for item in result], ["valid_near_analysis_date"]) + + +if __name__ == "__main__": + unittest.main() diff --git a/tests/test_news_strategy_config.py b/tests/test_news_strategy_config.py new file mode 100644 index 000000000..c1f536572 --- /dev/null +++ b/tests/test_news_strategy_config.py @@ -0,0 +1,23 @@ +# -*- coding: utf-8 -*- +"""Tests for NEWS_STRATEGY_PROFILE parsing and effective window calculation.""" + +import unittest + +from src.config import Config, resolve_news_window_days + + +class NewsStrategyConfigTestCase(unittest.TestCase): + def test_invalid_profile_fallback_to_short(self) -> None: + self.assertEqual(Config._parse_news_strategy_profile("bad_value"), "short") + + def test_window_respects_news_max_age_days(self) -> None: + # medium=7 but max-age=3 -> effective=3 + self.assertEqual(resolve_news_window_days(3, "medium"), 3) + # long=30 with max-age=30 -> effective=30 + self.assertEqual(resolve_news_window_days(30, "long"), 30) + # ultra_short=1 with max-age=30 -> effective=1 + self.assertEqual(resolve_news_window_days(30, "ultra_short"), 1) + + +if __name__ == "__main__": + unittest.main() diff --git a/tests/test_pipeline_realtime_indicators.py b/tests/test_pipeline_realtime_indicators.py index 269228344..7e59c819f 100644 --- a/tests/test_pipeline_realtime_indicators.py +++ b/tests/test_pipeline_realtime_indicators.py @@ -197,6 +197,16 @@ class TestEnhanceContextRealtimeOverride(unittest.TestCase): self.assertIn("price_change_ratio", enhanced) self.assertIn("volume_change_ratio", enhanced) + def test_enhance_context_injects_runtime_news_window_days(self) -> None: + context = {"code": "600519", "today": {"close": 15.0}} + enhanced = self.pipeline._enhance_context( + context, None, None, None, "贵州茅台" + ) + self.assertEqual( + enhanced["news_window_days"], + self.pipeline.search_service.news_window_days, + ) + def test_today_not_overridden_when_trend_missing(self) -> None: context = {"code": "600519", "today": {"close": 15.0}} quote = _make_realtime_quote(price=15.72) diff --git a/tests/test_search_news_freshness.py b/tests/test_search_news_freshness.py index 732cba1d0..5e5dc5834 100644 --- a/tests/test_search_news_freshness.py +++ b/tests/test_search_news_freshness.py @@ -1,10 +1,12 @@ # -*- coding: utf-8 -*- """ -Unit tests for search_stock_news and search_comprehensive_intel news_max_age_days logic (Issue #296). +Unit tests for strict news freshness filtering and strategy window logic (Issue #697). """ import sys import unittest +from datetime import datetime, timedelta, timezone +from types import SimpleNamespace from unittest.mock import MagicMock, patch # Mock newspaper before search_service import (optional dependency) @@ -17,107 +19,199 @@ if "newspaper" not in sys.modules: from src.search_service import SearchResponse, SearchResult, SearchService -def _fake_search_response() -> SearchResponse: - """Return a successful SearchResponse for mocking.""" +def _result(title: str, published_date: str | None) -> SearchResult: + return SearchResult( + title=title, + snippet="snippet", + url=f"https://example.com/{title}", + source="example.com", + published_date=published_date, + ) + + +def _response(results) -> SearchResponse: return SearchResponse( query="test", - results=[ - SearchResult( - title="Test", - snippet="snippet", - url="https://example.com/1", - source="example.com", - published_date=None, - ) - ], + results=results, provider="Mock", success=True, ) class SearchNewsFreshnessTestCase(unittest.TestCase): - """Tests for news_max_age_days in search_stock_news and search_comprehensive_intel.""" + """Tests for strategy window and strict published_date filtering.""" - def _create_service_with_mock_provider(self, news_max_age_days: int = 3): - """Create SearchService with a mock provider that records search() calls.""" + def _create_service_with_mock_provider( + self, + *, + news_max_age_days: int = 3, + news_strategy_profile: str = "short", + response: SearchResponse | None = None, + ): service = SearchService( bocha_keys=["dummy_key"], news_max_age_days=news_max_age_days, + news_strategy_profile=news_strategy_profile, + ) + mock_search = MagicMock( + return_value=response + or _response([_result("default", datetime.now().date().isoformat())]) ) - mock_search = MagicMock(return_value=_fake_search_response()) service._providers[0].search = mock_search return service, mock_search - @patch("src.search_service.datetime") - def test_search_stock_news_days_monday_limit_by_news_max_age( - self, mock_dt: MagicMock - ) -> None: - """Monday + news_max_age_days=1 -> search_days=1 (min(3,1)=1).""" - mock_dt.now.return_value.weekday.return_value = 0 # Monday -> weekday_days=3 + def test_effective_window_uses_profile_and_news_max_age(self) -> None: + """window = min(profile_days, NEWS_MAX_AGE_DAYS).""" service, mock_search = self._create_service_with_mock_provider( - news_max_age_days=1 + news_max_age_days=3, + news_strategy_profile="medium", # 7 ) - service.search_stock_news("600519", "贵州茅台") - mock_search.assert_called_once() - call_kwargs = mock_search.call_args[1] - self.assertEqual(call_kwargs["days"], 1) + service.search_stock_news("600519", "贵州茅台", max_results=5) + kwargs = mock_search.call_args[1] + self.assertEqual(kwargs["days"], 3) - @patch("src.search_service.datetime") - def test_search_stock_news_days_tuesday_weekday_dominates( - self, mock_dt: MagicMock - ) -> None: - """Tuesday + news_max_age_days=3 -> search_days=1 (min(1,3)=1).""" - mock_dt.now.return_value.weekday.return_value = 1 # Tuesday -> weekday_days=1 + def test_invalid_profile_falls_back_to_short(self) -> None: + """Invalid profile should fallback to short (3 days).""" service, mock_search = self._create_service_with_mock_provider( - news_max_age_days=3 + news_max_age_days=30, + news_strategy_profile="invalid_profile", ) - service.search_stock_news("600519", "贵州茅台") - mock_search.assert_called_once() - call_kwargs = mock_search.call_args[1] - self.assertEqual(call_kwargs["days"], 1) + service.search_stock_news("600519", "贵州茅台", max_results=5) + kwargs = mock_search.call_args[1] + self.assertEqual(kwargs["days"], 3) - @patch("src.search_service.datetime") - def test_search_stock_news_days_monday_news_max_age_dominates( - self, mock_dt: MagicMock - ) -> None: - """Monday + news_max_age_days=5 -> search_days=3 (min(3,5)=3).""" - mock_dt.now.return_value.weekday.return_value = 0 # Monday -> weekday_days=3 - service, mock_search = self._create_service_with_mock_provider( - news_max_age_days=5 - ) - service.search_stock_news("600519", "贵州茅台") - mock_search.assert_called_once() - call_kwargs = mock_search.call_args[1] - self.assertEqual(call_kwargs["days"], 3) + def test_search_stock_news_strict_filters(self) -> None: + """Drop old/unknown/future+2, keep future+1 and within-window dates.""" + today = datetime.now().date() + fresh = today.isoformat() + old = (today - timedelta(days=30)).isoformat() + future_1 = (today + timedelta(days=1)).isoformat() + future_2 = (today + timedelta(days=2)).isoformat() - @patch("src.search_service.datetime") - def test_search_stock_news_days_weekend(self, mock_dt: MagicMock) -> None: - """Saturday + news_max_age_days=5 -> search_days=2 (min(2,5)=2).""" - mock_dt.now.return_value.weekday.return_value = 5 # Saturday -> weekday_days=2 - service, mock_search = self._create_service_with_mock_provider( - news_max_age_days=5 + service, _ = self._create_service_with_mock_provider( + news_max_age_days=7, + news_strategy_profile="medium", + response=_response( + [ + _result("old", old), + _result("unknown", None), + _result("future_2", future_2), + _result("future_1", future_1), + _result("fresh", fresh), + ] + ), ) - service.search_stock_news("600519", "贵州茅台") - mock_search.assert_called_once() - call_kwargs = mock_search.call_args[1] - self.assertEqual(call_kwargs["days"], 2) - def test_search_comprehensive_intel_uses_news_max_age_days(self) -> None: - """search_comprehensive_intel passes news_max_age_days directly to provider.search.""" + resp = service.search_stock_news("600519", "贵州茅台", max_results=5) + titles = [r.title for r in resp.results] + self.assertEqual(titles, ["future_1", "fresh"]) + for item in resp.results: + self.assertRegex(item.published_date or "", r"^\d{4}-\d{2}-\d{2}$") + + def test_search_stock_news_overfetch_before_filter(self) -> None: + """Provider request size should be increased before filtering.""" service, mock_search = self._create_service_with_mock_provider( - news_max_age_days=2 + news_max_age_days=3, + news_strategy_profile="short", ) - with patch("src.search_service.time.sleep"): # avoid delay in tests - service.search_comprehensive_intel( + service.search_stock_news("600519", "贵州茅台", max_results=4) + args, kwargs = mock_search.call_args + requested = kwargs.get("max_results") + if requested is None: + requested = args[1] + self.assertEqual(requested, 8) + + def test_search_stock_news_try_next_provider_when_filtered_empty(self) -> None: + """If provider-A passes API call but all results are filtered, continue to provider-B.""" + today = datetime.now().date() + old = (today - timedelta(days=90)).isoformat() + fresh = today.isoformat() + + service = SearchService( + bocha_keys=["dummy_key"], + news_max_age_days=3, + news_strategy_profile="short", + ) + + p1 = SimpleNamespace( + is_available=True, + name="P1", + search=MagicMock(return_value=_response([_result("too_old", old)])), + ) + p2 = SimpleNamespace( + is_available=True, + name="P2", + search=MagicMock(return_value=_response([_result("fresh", fresh)])), + ) + service._providers = [p1, p2] + + resp = service.search_stock_news("600519", "贵州茅台", max_results=3) + self.assertEqual([r.title for r in resp.results], ["fresh"]) + p1.search.assert_called_once() + p2.search.assert_called_once() + + def test_search_comprehensive_intel_uses_same_filter(self) -> None: + """Comprehensive intel should use same strict date filtering logic.""" + today = datetime.now().date() + old = (today - timedelta(days=20)).isoformat() + fresh = (today - timedelta(days=1)).isoformat() + + service, mock_search = self._create_service_with_mock_provider( + news_max_age_days=3, + news_strategy_profile="medium", # min(7,3)=3 + response=_response([_result("old", old), _result("fresh", fresh)]), + ) + with patch("src.search_service.time.sleep"): + intel = service.search_comprehensive_intel( stock_code="600519", stock_name="贵州茅台", max_searches=2, ) + self.assertGreaterEqual(mock_search.call_count, 1) for call in mock_search.call_args_list: - call_kwargs = call[1] - self.assertEqual( - call_kwargs["days"], - 2, - msg=f"Expected days=2, got {call_kwargs.get('days')}", - ) + kwargs = call[1] + self.assertEqual(kwargs["days"], 3) + self.assertEqual(kwargs["max_results"], 6) # target 3 -> overfetch 6 + + for resp in intel.values(): + self.assertTrue(all(item.title != "old" for item in resp.results)) + + def test_effective_window_helper_has_no_side_effect(self) -> None: + """_effective_news_window_days should not mutate stored news_window_days.""" + service, _ = self._create_service_with_mock_provider( + news_max_age_days=3, + news_strategy_profile="short", + ) + service.news_window_days = 99 + resolved = service._effective_news_window_days() + self.assertEqual(resolved, 3) + self.assertEqual(service.news_window_days, 99) + + def test_unix_timestamp_normalizes_to_local_date(self) -> None: + """Unix timestamp should be converted to local date before window filtering.""" + dt_utc = datetime(2026, 3, 15, 23, 30, tzinfo=timezone.utc) + timestamp = str(int(dt_utc.timestamp())) + expected_local_date = dt_utc.astimezone().date() + parsed = SearchService._normalize_news_publish_date(timestamp) + self.assertEqual(parsed, expected_local_date) + + def test_iso_utc_string_normalizes_to_local_date(self) -> None: + """ISO datetime with timezone should be converted to local date.""" + dt_utc = datetime(2026, 3, 15, 23, 30, tzinfo=timezone.utc) + iso_text = "2026-03-15T23:30:00Z" + expected_local_date = dt_utc.astimezone().date() + parsed = SearchService._normalize_news_publish_date(iso_text) + self.assertEqual(parsed, expected_local_date) + + def test_rfc_utc_string_normalizes_to_local_date(self) -> None: + """RFC datetime with timezone should be converted to local date.""" + dt_utc = datetime(2026, 3, 15, 23, 30, tzinfo=timezone.utc) + rfc_text = "Sun, 15 Mar 2026 23:30:00 +0000" + expected_local_date = dt_utc.astimezone().date() + parsed = SearchService._normalize_news_publish_date(rfc_text) + self.assertEqual(parsed, expected_local_date) + + +if __name__ == "__main__": + unittest.main()