The match's final BV differential was given to every decision in it, so a fit's effective sample size was the match count and not the decision count: 15156 candidate comparisons from 20 matches came out at R-squared 0.016 with three weights contradicting their own descriptions. The host now records BV and live units per round, and a decision is labelled by the change over the three rounds after it. A corpus without the round log still fits on the old label, and the report says which was used.