Notes on prediction-market data
Microstructure, continuous capture, and how to get the most out of the archives. Written for quants, researchers, and data-driven traders.
The SupaGamma Python SDK is on PyPI
pip install supagamma. A dependency-light client for the SupaGamma API — sync and async from one package, pagination you do not have to write, and a paid-download path that tells you the cost before you spend it.
Read more →Backtesting prediction markets: resolution lookahead, survivorship, and the terminal-value trap
Prediction markets are unusually easy to backtest badly, because every contract has a known terminal value and it is sitting in the same table you are joining. Four failure modes and how to design them out.
Read more →Measuring execution quality when the order book is anonymous
You cannot observe queue position on an anonymized L2 book. You can bound it, and you can separate the fills from the cancels — which is usually the number that decides whether a strategy is real.
Read more →The delta tape vs the snapshot grid: what order-book sampling throws away
A book snapshot every five seconds and the event tape that produced it are not the same dataset. Here's exactly what the grid destroys, and why the tape is the thing worth capturing.
Read more →CSV, Parquet, or JSON: picking the right export for market data
The format you download in quietly decides how fast your research loop is. A quick, practical guide to when each of the three makes sense.
Read more →Why gapless, continuous capture is the whole game for market data
A dataset with holes is worse than no dataset — you can't trust a backtest run over data that silently skips the interesting moments. Continuity is the moat, and it's harder than it looks.
Read more →What L2 order-book data is, and why it matters for prediction markets
Level-2 order-book data captures the full depth of resting bids and asks over time — the microstructure most prediction-market datasets throw away. Here's what it is and what you can do with it.
Read more →