Diego Garcia

Technical Briefs

Notes from the workshop. Lessons learned from building ML models, insights from failed experiments, and technical deep-dives into what actually works.

Filter by Project

Filter by Tags

Prometheus

Rolling Windows vs Cumulative Stats: Why 'Recent Form' Made My Predictions Worse

I thought tracking 'last 5 matches' would capture team momentum better than season averages. Validation accuracy dropped from 60.8% to 57.4%. The lesson: in noisy domains, recency bias amplifies noise instead of capturing signal.

Dec 30, 2025

16 min read

Machine LearningSports AnalyticsFeature Engineering

Read brief

Prometheus

Temperature Scaling vs Isotonic Regression: How I Fixed My Overconfident Predictions and Improved Betting Performance by 84%

My Prometheus football prediction models were accurate but useless for betting until I calibrated their probabilities. Expected Calibration Error dropped from 0.25 to 0.02, turning -€847 into +€3,247 on a €10k bankroll.

Dec 11, 2025

18 min read

Machine LearningModel CalibrationSports Analytics

Read brief

Prometheus

I Made My Football Prediction Model More Sophisticated. Accuracy Dropped From 57% to 53%.

How I tested 10 goal prediction approaches, picked the best one, fed it to an over-engineered classifier, and watched my accuracy drop. The lesson: in noisy domains like football, complexity is technical debt.

Dec 4, 2025

15 min read

Machine LearningSports AnalyticsData Science

Read brief