How accurate are the projections?

Each past week is replayed using only what was known before kickoff — closing Vegas lines, prior game logs, the players who dressed — then scored against what actually happened. Three models are compared: the Monte Carlo sim, a simpler baseline model, and a naive season-to-date average.

Loading…

Accuracy by position

MAE = average miss in DraftKings points (lower is better). r = correlation with actual scores (higher is better). Bias = projection minus actual (0 = no lean). Fantasy scores are right-skewed, so a well-calibrated mean will sit a little above the typical outcome.

Weekly error

Average miss each week for the three models.

Are the ranges honest?

Where each actual score landed inside that player's simulated range. A perfectly calibrated sim is flat along the dashed line; a tall right bar would mean ceilings are underrated, tall edges that ranges are too narrow.