Track Record

Pre-season

We grade ourselves in public. Every number below comes from held-out evaluation data -- the model never saw the weeks it is being graded on. We include weak results alongside strong ones.

How we evaluate: Backtest numbers use held-out evaluation: walk-forward for weekly rankings (the model only uses data from before each game) and leave-one-season-out for streaming picks (each season is withheld entirely). Neither method cherry-picks favorable windows.
Live in-season data will appear here once games are played (September). The numbers below are from rigorous held-out backtests, not cherry-picked samples.

Weekly Start/Sit Rankings

Walk-forward (Gate v2): the model never sees data from the week it grades. Evaluated across 15,093 player-weeks, seasons 2021-2025.15,093 player-weeks evaluated across 2021, 2022, 2023, 2024, 2025 seasons (PPR). Zero leakage violations: each prediction was made before the game it grades.

Quarterback
62.3%
top-12 hit rate
Running Back
51.8%
top-12 hit rate
Wide Receiver
37.9%
top-12 hit rate
Tight End
56.2%
top-12 hit rate

Calibration

Our 80% prediction band covers 82.4% of actual outcomes — right on target (PICP vs. 80% nominal). Brier skill score 0.18 means we outperform the no-skill baseline. The model is honest about uncertainty: when it says 80%, it means 80%.

WR hit-rate (~0.38) is modest -- WRs are the hardest position to rank weekly due to target volatility. We include this number because hiding it would be dishonest. All other positions beat or match naive baselines.

Streaming Picks (K & DST)

Evaluated with leave-one-season-out: each season is completely withheld from training, then graded against actual results. The baseline is Vegas implied scoring totals alone.

Kicker Streaming Picks

Top-3 kicker streaming recommendations per week, evaluated against actual scores. Uses Vegas implied totals, weather, and prior season performance. Baseline = Vegas implied totals only (naive).

Model hit rate
38.0%
top-3 picks
Naive baseline
34.3%
Vegas totals only
Rank correlation
0.142
Pearson r
Season held outModelNaiveResult
202229.6%29.6%Tied / behind
202335.2%29.6%Beat naive
202442.6%40.7%Beat naive
202544.4%37.0%Beat naive

The 2022 kicker season (held out) was our weakest: the model exactly tied the Vegas-only baseline (29.6% for both) -- we include it rather than hide it. The other three held-out seasons (2023, 2024, 2025) each beat the baseline, so 3 of 4 seasons beat naive and none fell behind it.

Defense Streaming Picks

Top-3 DST streaming recommendations per week. Uses Vegas implied totals, opponent offensive profile, and weather. Baseline = Vegas implied totals only (naive).

Model hit rate
55.1%
top-3 picks
Naive baseline
48.1%
Vegas totals only
Rank correlation
0.297
Pearson r
Season held outModelNaiveResult
202251.8%46.3%Beat naive
202348.1%37.0%Beat naive
202459.3%55.6%Beat naive
202561.1%53.7%Beat naive

All four held-out DST seasons (2022-2025) beat the Vegas-only baseline, but the margins varied: 2024 was closest (+3.7 points of hit rate) while 2023 was widest (+11.1). We report every season rather than cherry-picking the strongest -- a narrow win is still shown as a win.

No third-party proprietary ranking or value products are published on this page. Every accuracy metric is our own model's output, graded against publicly available actual fantasy scores.