replicalpha v0.1.0

Research-reproduction Agent: PDF → factor code → backtest → Red Team → reproducibility score.

replicalpha takes a quant research PDF and produces a complete reproduction bundle in one command — structured paper metadata, executable Python factor code, backtest results, automated Red Team findings, and a plain-English reproducibility score.

📖 English README · 简体中文 README

Pipeline

PDF ─▶ paper2alpha (PyMuPDF + LLM JSON mode)
       └─▶ ResearchCard
           └─▶ DSL interpreter (+ LLM fallback) → Python factor code
               └─▶ qtype static check (look-ahead / future-function lint)
                   └─▶ pandas quintile IC backtest
                       └─▶ Red Team Validator (5 fixed checks)
                           └─▶ reproducibility scorer
                               └─▶ markdown report

What's in v0.1.0

• End-to-end pipeline glued by core.orchestrator.run_pipeline:
  • PDF extraction via vendored paper2alpha (PyMuPDF + OpenAI JSON-mode)
  • Factor code generation — DSL interpreter for a 7-operator whitelist, LLM fallback for unsupported formulas, qtype-gated
  • Minimal pandas quintile long-short backtest (IC time series, cumulative return, max drawdown, annualized Sharpe)
• Red Team Validator v1 — 5 fixed computational checks (overfitting hint, small-cap exposure, data leakage, sample concentration, factor redundancy)
• Reproducibility score (sign match + magnitude) with plain-English interpretation
• Markdown research-report aggregator
• CLI replicalpha run <pdf> with offline-friendly --extractor-mock / --codegen-llm
• FastAPI JSON server (POST /runs, GET /runs/{id}, GET /runs/{id}/report)
• Bundled 20-ticker × 500-day synthetic market data for offline tests
• Unit + end-to-end subprocess tests — 52 tests, 84% coverage (3.11 + 3.12 CI matrix)
• Bilingual README (English + 简体中文) + issue templates + SUPPORT.md

Quick start

```bash
uv sync
export OPENAI_API_KEY=sk-...
uv run replicalpha run path/to/paper.pdf \
--out ./out \
--data path/to/market_data.csv \
--start 2022-01-03 --end 2024-12-31
```

See examples/reproduce_demo.md for the offline demo walkthrough (no API key needed).

Part of the alpha-kit stack

replicalpha vendors and composes:

• paper2alpha — PDF → ResearchCard extraction
• qtype — factor-code look-ahead lint

License

MIT.