Insights

Notes on how the model actually works — backtesting, calibration, market efficiency, and the mistakes we made along the way. Written from a fully public betting record, losing runs included.

Methodology

Four backtest "edges" that died live

Look-ahead bias hiding in the data pipeline, why oversized edges signal your own error, winner's curse against sharp markets, and why a probability floor beats an edge filter.

29 July 2026
More methodology notes coming — calibration, staking, and league selection.