Back
Judge prediction error fell from 11.87 to 9.59–9.97 percentage points (16–19% less error than BASE). B1000 gameplay showed no clear strength gain; all paired 95% confidence intervals include 50%.