Writing
Notes on evaluating models.
Evidence-led technical notes on validation design, leakage, calibration and the operational decisions that machine-learning models actually support.
2026-10-12
Why random cross-validation overstated fraud performance
An IEEE-CIS fraud case study on random splits, temporal holdouts, review-queue precision and the difference between a risk rank and a calibrated probability.
Model evaluationFraud riskTemporal validation2026-10-12
What a lucky holdout taught me about model validation
A capital-market forecasting investigation into why strong walk-forward and holdout scores did not transfer to the Kaggle leaderboard.
Model evaluationFinancial MLWalk-forward validation