When I was doing T168455: [Epic] Implement code generation for model makefile maintenance I took the liberty of small changes in the make file to make it more consistent, one of them was sorting the arguments of merge_labels alphabetically which I assumed should not matter but while working on T182611: Train/test reverted edittype model for Catalan Wikipedia it seems human_labeled dataset should come first. This also brings another problem I caused in merge_label by making rules that give only one argument to merge_labels (like English Wikipedia) and my expectation was that it would only output the file but it seems it doesn't even work :/
Fixing the first part is rather easy, just swipe place of the arguments. Fixing the second problem is rather complex