v0.3.0 frames ml-quant-trading as a concrete benchmark target for coding agents, quant agents, and agent harnesses.
This is a GitHub documentation/challenge release. The PyPI package remains mlquantx==0.2.6; install with python -m pip install --upgrade mlquantx and use the repository docs for the challenge materials.
The release does not claim live tradability or deployable alpha. It packages a realistic research workflow that agents can run, audit, and report without needing broker credentials or proprietary market data.
Headline
Can your coding agent reproduce a 213-factor, cost-aware quant research pipeline without overstating the result?
What changed
- Added the Agent Quant Benchmark Challenge.
- Connected the challenge from the README top links and Fast Path table.
- Opened the coordination thread in Discussion #66.
- Pointed agent users to reproduction, benchmark, DSH, public-data, and private-evaluation report templates.
- Kept private evaluation explicitly redaction-safe for users who cannot expose strategy details, vendor data, or institutional infrastructure.
Challenge tracks
| Track | Best for |
|---|---|
| Zero-account smoke test | First-time users and package smoke tests |
| Protocol v1 CPU benchmark | Agent and hardware reproducibility reports |
| DeepSeek Harness run | DSH users who want tool-assisted benchmark validation |
| Public-data validation | Cost-aware public-data reports with caveats |
| Private evaluation note | Redacted institutional or proprietary-data evaluations |
Start here
- Challenge page: https://github.com/initial-d/ml-quant-trading/blob/main/docs/agent_quant_benchmark_challenge.md
- Discussion thread: #66
- Colab quickstart: https://colab.research.google.com/github/initial-d/ml-quant-trading/blob/main/notebooks/quickstart_colab.ipynb
- DeepSeek Harness recipe: https://github.com/initial-d/ml-quant-trading/blob/main/docs/deepseek_harness_recipe.md
Evidence boundary
This release is for research and engineering evaluation. It is not investment advice, not a live-trading system, and not a claim that synthetic or public-data benchmarks predict deployable trading performance.