MLB, NFL, and NHL fantasy projections, scored against what actually happened.
Every model we run is graded on the slates it projected: mean absolute error, rank correlation, and top-10 overlap, under DraftKings and FanDuel scoring. This page reads from the same performance feed our subscribers see inside the product. Nothing on it is hand-entered.
Updated after each sport's monitor run. Pick a sport.
Loading the live feed.
Numbers come straight from the model performance endpoint. One moment.
Three metrics, each answering a different question.
On average, how many fantasy points a projection missed by, in either direction. A DK MAE near 5 means a player projected at 12 typically lands between 7 and 17. MAE is scale-dependent: pitchers and QBs score more points than bench bats, so compare MAE within a position group, not across sports.
Whether the model ordered the slate correctly: did the players projected higher actually score higher. This is the metric that builds lineups. You can have a mediocre MAE and still win the ordering, and vice versa, which is why we publish both.
Of the 10 players the model ranked highest over the window, how many finished in the actual top 10. The bluntest of the three, and the hardest to fake: it measures whether the model finds the players that decide tournaments.
One number this page will never show: guaranteed ROI. Accuracy is measurable and ours to own. Contest results depend on entry selection, field behavior, and variance. We publish our accuracy. We don't claim guaranteed ROI.
An accuracy number without a baseline is a decoration.
A fair question about every number above: is that good? The honest answer requires a baseline computed on the identical slates, and that is what lands on this page next: the same MAE, correlation, and top-10 overlap for two projections anyone could build themselves. A trailing-average projection (each player projected at his recent per-game mean) and a salary-implied projection (points inferred from DK and FD pricing). If our edge over the naive baselines is small, this page will say so.
On competitors: our commitment on the About page is that we will not name a competitor and assert we beat them without a public backtest. That cuts both ways, and the offer is standing. Any projection service that publishes slate-level accuracy under a stated methodology is welcome on this page, scored side by side on the same slates. As of this writing, we know of no major DFS projection vendor that publishes theirs. If that changes, this paragraph changes.
The methodology behind the models is public too: the Insights notes cover the Bayesian posterior, the correlated Monte Carlo layer, the ILP optimizer, and the consensus tier. The projections are graded by the same pipeline that produces them, and the grading code runs on a schedule, not when the numbers look good.
Subscribers see this scoreboard per slate and per player, next to the projections themselves.
Get Started