·
Two records answer two questions
Forecast accuracy counts the correct and incorrect original selections within a checkpoint. Forecast units settle the same forecasts as 1 unit each at the best sportsbook price when they were published. A high hit rate on short prices can still lose units, so neither is a substitute for the other.
Losses stay visible
Eligible historical losses remain in the reporting period. A model becoming unavailable does not remove its old results. Search and status filters on a daily board do not change the reporting cohort.
Some results need their own category
Ties, voids and dead heats are not ordinary binary wins. Suspensions and unresolved evidence may remain pending or require review. Accuracy uses correct divided by correct plus incorrect; ROI is units over priced wins and losses, and a tie or void returns the stake.
Read the dates and the sample
Forecast periods follow game or tournament start dates. The report shows its as-of time and sample counts, including how many settled forecasts were priced. An empty denominator is no graded results, not a 0% track record.
SharpDeck field notes. Product explanations and worked examples, not betting recommendations.