Under the hood: the models & the results board
Reading the results & ROI board
5 min
FinalSkore publishes how its tracked picks actually did. Reading that board well is a skill — done wrong, it either flatters or scares you for no reason.
Win rate and ROI are different questions
Win rate (WR) is how often picks win. ROI is how much money they'd have made per unit staked. They can disagree: a run of low-odds favourites can post a high WR but a thin ROI, while selective picks at bigger prices can win less often yet earn more. ROI is measured in units (see Bankroll & discipline) so it doesn't depend on how big your bankroll is.
Sample size is everything
A 60% win rate over 20 picks is almost meaningless — that's within the range of pure luck. The same 60% over 500 picks is a real signal. Before you conclude anything from a hot or cold stretch, check how many picks it covers. Short windows swing wildly; that swing is variance, not the model breaking or fixing itself.
Calibration by band
Because picks are grouped into 65% / 75% / 85% bands, you can check something powerful: are the bands calibrated? Over a large sample, the 85% band should win about 85% of the time, the 65% band about 65%. If it does, the probabilities mean what they say. The board also lets you toggle markets — for basketball, over/under versus moneyline — so you compare like with like.
Reading it honestly
- Judge the model over hundreds of picks, never one night.
- Expect the high bands to lose sometimes — an 85% band that never lost would be mis-calibrated.
- Past ROI is a track record, not a promised future return.
The board is a rear-view mirror, not a windshield. It tells you whether the process has been sound — it cannot tell you tonight's result.