We publish two types of prediction. The algorithm predictions come from a probabilistic model that, for each match, estimates the probability of every outcome (1X2, double chance, under/over, both teams to score) and picks the selection with the best value. The studied predictions are our editorial columns: for the leagues we follow, an analyst builds up to four predictions per match in order of reliability.
The outcome check is automatic and objective: every prediction is compared with the real final result of the match, imported from official data. A prediction is "won" if the outcome occurred, "lost" otherwise. We neither rewrite nor hide anything after the fact: the rates above are the ratio between correct predictions and assessed predictions, over the entire available history.