Player explorer

Projections next to reality

The main chart puts a player's actual production in every game next to what the projection model expected for that game. The predictions shown for past seasons are an out-of-sample backtest: each one was produced by a model trained only on earlier seasons, so it never saw the game it was predicting. That makes the comparison a fair test rather than a curve fitted to the answers.

Clearing a line

The line tool answers a common fantasy question: how often does this player get to a given number? The curve shows, for every possible line, the share of games in your chosen window that beat it. With only five or ten games the confidence interval is wide by design. Small samples really are that uncertain, and the tile says so rather than hiding it.

For low-count stats like assists, steals, blocks, turnovers and threes, the model treats each game as a Poisson count around the player's projected rate. Comparing that model-implied rate with the actual hit rate shows whether a player is streakier, or steadier, than the model assumes.

Prediction error and the next matchup

The error chart shows each miss as predicted minus actual, so a consistent positive bias means the model has been too optimistic about this player. The team comparison at the bottom lines up the player's team against its next opponent on pace, efficiency and rebounding: the context the next projection will lean on most.