Methodology, racing analysis, and honest simulated-performance analysis.
One living article per track: trap bias, favourite strike rate at SP, early pace, running lines, trainers and pars, counted from the last year's results and recomputed as new results arrive. 35 tracks.
Track of the week · week 38
167 meetings in the last year, last raced 2026-09-13.
Read the numbers →11 June 2026
SHAP values decompose a machine-learning prediction into per-feature contributions. How they work, how to read the 'Why this pick' panel, and the honest limits of model explanations.
Read more →11 June 2026
Circumference, trap bias, going and grade depth vary far more between greyhound tracks than most analysts assume. A data-driven tour of what separates Newcastle from Romford — and why it matters for racing analysis.
Read more →30 May 2026
Each-way selections look like insurance but the maths is unforgiving on tight place markets. Why TrapStats only flags win selections — and what the numbers showed.
Read more →30 May 2026
A practical guide to every column on a UK or Irish greyhound racecard. What trap numbers tell you, how grades stack, what the form figures mean, and where the market focuses first.
Read more →30 May 2026
A transparent look at the ML pipeline behind TrapStats — what features the model uses, how it's calibrated, and what its honest accuracy looks like on live UK and Irish greyhound racing.
Read more →30 May 2026
Inside-trap dogs do win more often at most UK tracks — but the effect is smaller than analysts think and varies dramatically by venue. What the 180-day data says and how to use it.
Read more →30 May 2026
Two different probability calibrators on a narrow-input ML problem produce wildly different live behaviour. Why isotonic collapsed to a phantom-EV machine — and what Beta does differently.
Read more →29 May 2026
How the Kelly Criterion calculates a mathematically optimal allocation for each selection, why we use a fractional Kelly, and a worked example on a real greyhound race.
Read more →