When 100% Accuracy Is a Red Flag

Building the regime detector for Kairox Vector, I got perfect out-of-sample accuracy on the first real run. It took two iterations to find the bug: the model was not learning market regimes, it was reading its own label definition back as a feature.

xG Overperformers: Skill or Luck?

When strikers consistently beat their expected goals, is it clinical finishing or statistical noise? The regression-to-mean data tells a different story than the pundits.

Bookmaker Calibration: The 0.52% Edge

Bookmakers achieve 99.48% calibration accuracy. But systematic biases of 1-7% exist in specific outcomes across leagues — and that microscopic edge is where profitable betting lives.