End-to-end Python implementation of He's (2026) audit pipeline for LLM factor-mining claims. Implements: hash-verified custody, anti-look-ahead feature/label separation, freeze-before-score cohorts, average-tie daily rank-IC, a three-state ledger that keeps every record in the denominator, a versioned v1-to-v2 evaluator fix and claim governance.
python reproducible-research jupyter-notebook quantitative-finance multi-agent-systems backtesting nonparametric-statistics data-provenance selection-bias paper-implementation metascience baostock large-language-models empirical-asset-pricing llm-agents information-coefficient china-a-shares alpha-mining rank-ic alpha-factor-mining
-
Updated
Nov 14, 2026 - Jupyter Notebook