Quant-skeptic questions, answered honestly.
The questions a careful subscriber would ask before signing up: methodology, sample size, scope, and the things we will not claim. If you have one we have not answered, send it our way.
MLB, NFL, and NHL today. NCAA, NBA, EPL, and Golf are on the roadmap. We will not ship a sport until projection accuracy is at parity with the others. See /sports for the rationale on each.
DraftKings and FanDuel. Coverage is the same on both for shipped sports. The optimizer respects each platform's specific roster rules (DraftKings NHL three-teams minimum, FanDuel's stacking rules), and the consensus tier reads from both publishers in parallel.
Bayesian hierarchical models with a tier structure. Each player's component statistics (singles, doubles, walks, strikeouts, shots on goal, and so on) are modeled as posterior distributions, not point estimates. The /about page (#methodology) covers this at paragraph level; /insights/lineup-optimizer goes deeper on how the optimizer consumes the posterior.
A single label per player on every slate: green pill (esa_verified) when at least one trusted source has independently confirmed, blue pill (platform_projected) when no source has confirmed but at least one platform has projected. Sources include DraftKings, FanDuel, league feeds, and team feeds. See /insights/consensus-tier for the design decisions.
Full inference runs at slate-load, typically 5 AM CT for main slates. A delta inference job then runs hourly through the afternoon to detect lineup changes (scratches, late starters, goalie announcements) and patch the affected projections. Late changes do not silently leave the slate stale.
Yes. The delta inference job runs through the afternoon for late-swappable contests (NFL Sunday, MLB main slate). The optimizer recompiles with the updated player pool when you re-generate a lineup.
Yes, and we want to be clear about that. We publish MAE, rank correlation, and top-K overlap because those are the metrics the model is trained to optimize. ROI depends on contest selection, exposure management, and field composition. Those are decisions we do not make for you. Accuracy is the input; ROI is the outcome.
Per-slate, per-sport, per-position. The model-accuracy dashboard inside the product shows the last N slates for each cohort; sample sizes vary by how recently a sport ran. NHL has fewer historical slates than MLB simply because it is a younger product on our side.
Because we have not run the out-of-sample work that would defend the claim, and we will not assert beating a named competitor on a benchmark we have not committed to. We publish our own numbers and let you compare.
No. Daily fantasy contains irreducible variance. A model that improves your decisions is not a model that prints money. Anyone who tells you otherwise is selling you something other than a model.
We do not have a date. NBA introduces rotation variance and pace effects that make naive transplant of the MLB/NFL/NHL model architecture unreliable. We will not ship NBA until the projections are at parity with the other sports.
No formal trial. The weekly subscription is the lowest-commitment way to evaluate the product on a slate or two. Cancel anytime.