Checked against 16 real slates — good for comparing strategies against each other, not a calibrated prediction of real payout. See the note at the bottom of this tab for what's verified vs. not.
Per-lineup results
Click a column header to sort. Click a row to see who's in that lineup.
ROI by lineup
Where the edge (or lack of it) is concentrated across your portfolio.
Portfolio (all lineups as combined entries in this contest)
What this simulation does and doesn't capture: player outcomes are sampled from a correlated model (game/team-level shared shocks, not independent per player) anchored to each player's own median/floor/ceiling. The opposing field mixes a naive ownership-weighted "public" portion with a small precomputed pool of near-optimal "sharp" lineups (default 15% of the field) — a deliberate improvement over pure noise, but the sharp fraction and correlation weights are reasoned placeholders, not fitted to real historical data. Each trial samples a few hundred field entries (not the full field) to estimate percentile, which adds noise to any single trial's extreme-tail estimate — trust the aggregate rates across thousands of trials, not any one number in isolation.
Checked against 16 real slates (2023–2024 contest history, matched to old ETR exports): the simulated field's score shape held up well — when a slate's projections were accurate, the simulator's full percentile curve tracked the real one closely (within ~1 point at every percentile, in the best case). The remaining gap is mostly explained by ordinary week-to-week projection error, not a structural flaw in the correlation/variance model. Across all 16 slates there's a small, borderline-significant upward bias — the simulator ran about 5 points optimistic on average (95% CI roughly [0, +9.5]), with wide slate-to-slate swings (−8 to +25) that a sample this size can't fully explain away as either real or noise.
What that means for the numbers above: treat ROI, cash rate, and top-X% figures as comparative — reliable for judging one strategy or lineup against another on the same slate — not as calibrated predictions of real-money return. The absolute level may run slightly optimistic. Confidence is highest for cash-line/median-range outcomes (best-sampled bucket); the extreme tail (top 0.1%, top 0.01% — what actually separates GPP strategies) has the least real-world validation, since real ceiling outcomes are rare by nature and this history didn't produce enough of them to check against.