docs(naming): stop overstating Alpha158 / causal chains / IC (P1.5)

README.md

@@ -42,7 +42,7 @@ An LLM-powered intelligent analysis platform for the A-share (Chinese stock) mar
 | 🌊 事件总线 | Redis Streams 事件驱动微 OODA（行情 / 新闻 / 情绪 / 信号） | Event Bus | Redis-Streams event-driven micro-OODA (market / news / sentiment / signal) |
 | 🛡️ 风险引擎 | 熔断器 + Kelly 仓位 + VaR + A股约束（T+1 / 涨跌停 / 100 股） | Risk Engine | circuit breaker + Kelly sizing + VaR + A-share constraints (T+1 / price limits / 100-share lots) |
 | 🌐 全球情报 + 新闻 | 全球指数 / 大宗 / 汇率关联 + AI 新闻聚合 | Global Intel + News | global indices / commodities / FX correlation + AI news aggregation |
-| 📈 量化回测 | backtrader 策略回测，可选 Qlib Alpha158 | Backtesting | backtrader strategy backtesting, optional Qlib Alpha158 |
+| 📈 量化回测 | backtrader 策略回测，可选 Qlib 自定义 alpha 因子 | Backtesting | backtrader strategy backtesting, optional Qlib custom alpha factors |
 | 📱 Discord | 交易信号 / 情报自动推送到 Discord | Discord | auto-push trade signals / intel to Discord |
 | 🖥️ Web UI | FastAPI + React 19：ControlTower / Portfolio / Recommendations / Review | Web UI | FastAPI + React 19: ControlTower / Portfolio / Recommendations / Review |
 | ⚙️ 自动化调度 | Celery 45+ 定时任务 + 常驻 agent 守护进程 | Automation | Celery beat (45+ tasks) + always-on agent daemon |
@@ -88,7 +88,7 @@ src/web/ FastAPI  ·  frontend/ React SPA  ·  src/discord_bot/  ·  openclaw/ C
 | 数据 / Data | AKShare, adata, EastMoney push2 (curl_cffi), QMT/XtQuant (optional), pandas, numpy, pyarrow, yfinance |
 | 分析 / Analysis | ta (technical indicators), plotly, matplotlib |
 | AI 预测 / AI | Anthropic Claude, Google Gemini, OpenAI, DeepSeek, Claude Code bridge (fallback) |
-| 量化 & Agent / Quant | hmmlearn (HMM 市场状态), networkx (知识图谱), scikit-learn, Qlib Alpha158 (optional) |
+| 量化 & Agent / Quant | hmmlearn (HMM 市场状态), networkx (知识图谱), scikit-learn, Qlib custom alpha factors (optional) |
 | 策略回测 / Backtest | backtrader, Qlib (optional) |
 | 后端 / Backend | FastAPI, uvicorn, Redis (cache + Streams 事件总线), Celery + Beat |
 | 前端 / Frontend | React 19, TypeScript, Vite, shadcn/ui, Tailwind CSS 4, React Query |

docs/guides/development-guide.md

@@ -42,7 +42,7 @@ See @requirements.txt for Python dependencies and @frontend/package.json for fro
 |-------|-----------------|
 | Data | AKShare, adata, EastMoney push2 (curl_cffi), XtQuant (QMT), pandas, numpy, pyarrow, Qlib (optional) |
 | Intelligence | NetworkX (knowledge graph), feedparser, ddgs/searxng |
-| Quant / Agent | hmmlearn (HMM regime), scikit-learn, Qlib Alpha158 (optional) |
+| Quant / Agent | hmmlearn (HMM regime), scikit-learn, Qlib custom alpha factors (optional) |
 | Prediction / LLM | Anthropic Claude, Google Gemini, OpenAI, DeepSeek, Claude Code bridge |
 | Strategy | backtrader |
 | Web Backend | FastAPI, uvicorn, Redis (cache + Streams event bus) |

scripts/qlib_worker.py

@@ -160,7 +160,9 @@ def cmd_predict(symbols: list[str], horizon: int) -> dict[str, dict[str, Any]]:
                     # Use mean recent return as raw score, sigmoid to [0,1]
                     raw = float(returns.mean()) * 100  # scale up
                     score = round(1.0 / (1.0 + math.exp(-raw)), 4)
-                    # IC: autocorrelation of returns as proxy
+                    # NOTE: serial autocorrelation of returns, used as a rough
+                    # proxy only — this is NOT a true cross-sectional Information
+                    # Coefficient (rank-corr of factor vs forward returns).
                     if len(returns) >= 10:
                         ic = round(float(returns.autocorr()), 4)
 
@@ -278,6 +280,9 @@ def cmd_ic(symbol: str) -> dict[str, Any]:
         if data is None or data.empty or len(data) < 10:
             return {"symbol": symbol, "ic": None, "reason": "insufficient data"}
 
+        # Serial return autocorrelation used as a rough proxy — NOT a true
+        # cross-sectional Information Coefficient. The "ic" key is kept for
+        # config/back-compat; treat it as an autocorrelation signal.
         ic_val = float(data.iloc[:, 0].dropna().autocorr())
         threshold = _ACTUARY_CFG.get("ic_threshold", 0.03)
         return {

src/intelligence/impact_chain.py

@@ -1,7 +1,9 @@
 """Impact Chain Engine — builds event-to-asset transmission chains.
 
-Per PRD v34.0 FR-GI002: Automatic construction of causal chains from
-macro events to affected sectors and stocks.
+Per PRD v34.0 FR-GI002: maps macro events to affected sectors and stocks using
+hand-authored, keyword-matched transmission **templates** (a curated rule base).
+These are domain-knowledge templates, NOT statistically inferred causality — the
+edge confidences are fixed constants, not estimated from data.
 
 Example chain:
   中东战争 → 原油↑ → 航运成本↑ → 航运股↓

src/quant/qlib_alpha.py

@@ -1,7 +1,9 @@
-"""Qlib Alpha factor computation — enhanced factor library for stock screening.
+"""Qlib alpha factor computation — custom factor library for stock screening.
 
-FR-QL001: Wraps Qlib-style Alpha158 factor categories into a computation
-engine that works via the QlibAdapter (remote/subprocess/in-process).
+FR-QL001: a hand-written set of ~20 alpha factors expressed in the Qlib DSL,
+computed via the QlibAdapter (remote/subprocess/in-process). This is *inspired
+by* Qlib's Alpha158 factor categories but is NOT Qlib's Alpha158 dataset handler
+— it does not instantiate or reproduce the full 158-factor set.
 
 Factor categories:
 - Momentum: 5d, 10d, 20d, 60d price momentum