Extended Data Fig. 2: Accuracy results.

Accuracy during different stimulation experiments, for A) open-loop and B) closed-loop capsular stimulation. Boxes show the mean and confidence intervals for accuracy with stimulation at each site. Colours indicate stimulation sites as in the main text. The p-value above each bar represents a binomial exact test of accuracy compared to the non-stimulated baseline condition, with Benjamini-Hochberg false discovery rate correction. All accuracies are above 95%, with accuracy during stimulated blocks being very slightly higher in most cases. No results exceed chance significance. We did not have open- and closed-loop data from the same participants. To compare the CL and OL conditions, we therefore compared their accuracies across participants with a Fisher exact test for each of the three stimulation sites (L Dorsal, R Ventral, R Dorsal) that were used in both conditions. L Dorsal: p = 0.645. R Ventral: p = 0.440. R Dorsal: p = 0.655. These provide no evidence for a difference between OL and CL conditions. These results do not support a change in accuracy with any stimulation type. That is, the observed decrease in reaction times is a true performance improvement, not a shift along a speed-accuracy tradeoff. We were unable to analyse accuracy in the GLME framework because the differences between stimulation sites are so small as to make the models non-identifiable in all cases.