what is openmarket fr?

OpenMarket is an open source Rust research project that collects, matches, and studies high frequency data from two financial venues at once: Binance's Bitcoin market and Polymarket's short term Bitcoin prediction markets. Rather than being a trading bot, it is a data and methods release meant to let researchers reproduce and check the work themselves. The project has now reached what its authors call archival shutdown. It is frozen as a public research record, and the large datasets it produced, together with a trained model, are published separately on Hugging Face rather than inside the code repository itself. Over about 109 days of collection, the project gathered 202 database snapshots covering millions of order book updates and trades, which were later combined into one unified dataset of more than 727 million rows and nearly 9 gigabytes of Parquet files, plus about 2.9 million matched pairs of events across the two venues. The Rust codebase itself contains the pieces needed to reproduce this pipeline: collectors that connect to Binance and Polymarket over WebSocket connections, a recorder that saves snapshots, a timestamp synchronizer that lines up events from both venues, feature generation code, a machine learning and signal engine, and a backtesting and evaluation system. All of this is organized as a single Rust workspace of about 17,800 lines of code. The README is candid that the released prediction model, while showing reasonable calibration and ranking skill, does not outperform a simple baseline strategy out of sample, and that a simulated trading strategy using it loses money once realistic fees and slippage are included. The project frames itself as documenting stylized facts about how these two markets relate rather than proving a profitable trading strategy. Findings described include a typical delay of about 16 milliseconds between related events on the two exchanges, a slower roughly 347 millisecond delay in how Polymarket prices react to large Binance price moves, and evidence that most order book spreads sit at the tightest possible single tick, which limits how much real trading profit could be captured from any price gap. The project is released under the Apache 2.0 license, and its published datasets are provided in multiple splits on Hugging Face depending on whether a user wants the full raw archive, the deduplicated research timeline, or a small quickstart sample.