- Project status, limitations, and improvement roadmap
- Methodology equations
- Final research-reference version
- 2026-07-16 consolidation release notes
- GitHub push plan
- Live-forward monitoring and daily operations
- GitHub Pages dashboard deployment
The repository now has two documentation audiences:
mainshould keep the finalized reusable pipeline, concise project narrative, and current evidence-backed limitations.archive-experimentsshould preserve the full dated experiment trail, failed gates, rejected designs, and development-only research artifacts.
Do not treat dated experiment logs as active production modules unless the corresponding gate decision explicitly says they are active.
- Data ingestion, caching, and point-in-time inputs
- Sentiment and event NLP
- Feature engineering
- Prediction targets
- Forecasting and ranking models
- Portfolio construction and risk
- Backtesting, execution, costs, and performance metrics
- Baselines and benchmarks
- Pair trading and statistical arbitrage
- Experiment governance and confirmation
- Final evidence-backed system proposal
- Rebalance-buffer experiment draft
- Next research direction: transparent selection, topics, LLM events, and pairs
- Sentiment and event overlay
- 2026-07-29 - Live daily-selection versus frozen-holdings ablation
- 2026-07-28 - First refreshed 20-session live-forward run
- 2026-07-28 - Live horizon evidence and design decision
- 2026-07-28 - Matched 5/10/20-session forecast-horizon sweep
- 2026-07-28 - Legacy live-settings backtest
- 2026-07-27 - Ridge-XGBoost rank ensemble
- 2026-07-20 - Model-specific tuning and ensemble v10
- 2026-07-20 - Qlib Alpha158 and LightGBM U.S. comparison
- 2026-07-17 - Survival-risk filter diagnostic for v8
- 2026-07-17 - Repaired rank-40 ML stock-selection evidence
- 2026-07-16 - Transparent smart-universe experiment v7
- 2026-07-16 - Fama-French target and attribution study v6
- 2026-07-16 - Timestamped-news matched event study v5
- 2026-07-16 - Point-in-time topic features v5 baseline
- 2026-07-16 - Weekly slow-NLP incremental test v5
- 2026-07-16 - Point-in-time pair-trading development test v5
- 2026-07-15 - Turnover, sizing, and public confirmation v4
- 2026-07-14 - Sparse ranking, sentiment, and correlation audit
- 2026-07-14 - Residual-rank v2 implementation
- 2026-07-14 - Revealing forecast and universe controls
- 2026-07-13 - Canonical weekly pipeline rerun
- 2026-07-13 - Economic and market-regime stability
- 2026-07-12 - News feature A/B backtest
Experiment logs are dated, tracked Markdown records. Generated CSV files, equity curves,
charts, cached datasets, and model artifacts remain under ignored reports/ and data/
directories so the repository stays lightweight and reproducible.