`r.kappa` lacks: * User's and Producer's accuracies * Output in CSV * sort output differently: * arrange user and producer accuracies and omission and commission around the error matrix instead of the summary columns * add notes to "map1" and "map2" to explain what that means (more user friendly) For a working solution, see https://github.com/OSGeo/grass-addons/blob/master/grass7/raster/r.confusionmatrix/r.confusionmatrix.py