@mllg
if you this

TuningInstance

      best = if (measure$minimize) which_min else which_max
      tab$resample_result[[best(y, na_rm = TRUE)]]

please add an inline code comment why NA removal is necessary (and not only: "for good measure / defensive / to make sure")

and also please point to the mlr3 docs which explain why an NA can be returned by BMR$aggregate