Including the comparison we lose
Fantasy tools are full of confident numbers and empty of evidence. So here is ours, measured the only way that means anything: on seasons the model never saw while it was being built.
Against consensus market rankings and against FantasyPros, our season projections win in every gradeable season we have measured. In 2025 that was +85.7 points against the market and +101.7 against FantasyPros.
On held-out weekly projections we lose to ESPN, by −18.1, −20.2, −40.7 and −32.4 points per week across 2022–2025. We tested whether blending our model into theirs helped. It did not — the confidence interval sat entirely below zero.
So the app serves ESPN’s weekly projection where ESPN is better, and ours where ours is. A tool that insists on its own number when a free one beats it is selling pride, not accuracy.
Every model is graded on data it could not have seen when the call was made. A backtest that peeks is a press release.
Fit on earlier seasons, graded on later ones. If a result only exists in the seasons it was fitted on, it does not ship.
When the data cannot support an answer the app says so instead of guessing. A confident wrong number costs you a week; an honest “we don’t know” costs you nothing.
Because the alternative is asking you to trust a number with nothing behind it. Knowing where a tool is weak is what lets you use it where it is strong — and our strength was never weekly point projections. It is the decisions around them: what a keeper actually costs once his price escalates, which players return to your draft pool when every rival keeps theirs, and what your league’s own roster shape does to a player’s value.
Those are the questions the big platforms do not answer, and they are the ones we grade ourselves on hardest.