This repository is documented around two active operating assumptions:

• vendor-backed historical data is local-first: mirror PMXT raw archive hours and Telonex full-book parquet parts onto local disk, then replay through L2-native adapters
• public Python backtest runners expose run() and build their explicit experiment inputs inline: MarketDataConfig, BookReplay, strategy configs, ExecutionModelConfig, MarketReportConfig, and either build_replay_experiment(...) or ParameterSearchExperiment(...)

PMXT uses local raw files plus remote archives instead of a separate service surface. Telonex uses materialized OrderBookDeltas replay cache, api: as the public-runner first source, and a local Hive-partitioned full-book mirror as a fallback. The primary operating docs are grouped under Core Framework; optimizer research and ledger replay are grouped under Advanced / Experiments.

Start Here

• Setup

Core Framework

• Backtests And Runners
• Sandbox And Live Runners
• Data Loading
• Data Vendors And Local Mirrors
• Vendor Fetch Sources And Timing
• Execution Modeling
• Plotting
• Testing

Advanced / Experiments

• Research And Optimizers
• Polymarket Account Ledger Replay

Project

• Project Status
• License Notes

Acknowledgements

I'd like to thank everybody who I talked to along the way, as well as everybody who has starred, forked, filed issues, and asked questions about this project. Being only 19, I started with very little knowledge about the inner workings of markets on a microstructure level, and now have a lot more experience in strategy research and optimization. This repository started when I wanted to test my friend's hypotheses, and serves as my attempt at an all-in-one backtesting solution for prediction markets, with easy access to data and abstractions around a well-known backtesting framework.