OpenMarket: A Synchronized Polymarket–Binance Dataset for High-Frequency Prediction-Market Research

Battle Count: 62

Highly relevant as infrastructure and dataset for prediction market microstructure research. The paper provides the first public millisecond-level Polymarket-Binance paired corpus with explicit pairing metadata, enabling reproducible cross-venue timing studies. However, the central empirical result is explicitly negative: no tradable edge was found, and the multivariate model underperforms the market prior out-of-sample. The value is primarily as a research tool and benchmark rather than a trading alpha source. The synchronization methodology and clock-offset validation are directly applicable to any cross-venue HFT research.

Moderate complexity. The Rust workspace spans exchange collectors, multi-market recorder, lag-pairing export, signal/execution engines, backtesting, and Parquet-native ML crates. However, the pipeline is well-documented with pinned commands, Docker scaffolding, and a Jupyter quickstart. The core ML pipeline (step3 export + binary-outcome-trainer) runs in ~130s on a laptop. The main complexity lies in understanding the synchronization layer, pairing quality flags, and walk-forward calibration protocol. No GPU required; runs on CPU.

Exceptional reproducibility: full Rust pipeline published on GitHub (tag v0.5.2, Apache 2.0), Hugging Face dataset (v0.4.3-unified, v0.2-full, v0.1-sample) and model artifacts (v0.2.1/, v0.1/), pinned commands, Docker scaffolding, Jupyter quickstart, mdBook documentation, and a paper/scripts/compile.sh to regenerate all statistics and figures. Version map provided for all artifacts. Clock-offset validation scripts included. The frozen archive is explicitly designed as an immutable public research record.