Research
Methodology deep-dives, backtests, and modeling education — the receipts behind what we publish, not just the claims.
Daily picks process: schedule, models, prices and research
Our current operating flow, exact scheduled times, selection rules, review budget and missing-data behavior.
-
NBA Forecasting and Market Pricing: What v1 Can and Cannot Do
Our first NBA model, tested once on 2025–26. Projecting minutes first beat recent averages for every player stat. DraftKings' game lines beat the model, and matchup, home/away and back-to-back splits added nothing.
-
What Makes Games Go Under? Weather, Fatigue, Travel and Officials Against the Closing Total
60 pre-registered situations in 90,866 NFL, NHL, NBA and MLB games. Most lower scoring and are already in the closing total. Strong NFL wind is the clear exception; officials and back-to-backs gave no edge.
-
Player Matchups and Home Field: Do Head-to-Head History and Venue Splits Predict the Next Game?
A backtest across NFL, NHL and MLB of whether a player's record against one opponent, or his own home/away split, predicts his next game. It doesn't. League venue effects, platoon and player bias did help, and the models changed accordingly.
-
NHL Forecasting and Market Pricing: Evidence, Coherence, and Limits
A four-season evaluation of game-score and player-count models, with equations, timestamped sportsbook comparisons, and a plain-language explanation of what improved—and why a betting edge remains unproven.
-
Brier Scores and Calibration: Grading Our Own Model Honestly
What a Brier score is, why it's the right way to grade a probabilistic model, and the real, unfiltered results of backtesting our own NFL player prop model against 2025 outcomes — including a validation bug we caught and fixed along the way.