For anyone else that has this issue you need to redownload the test data, then reindex your predictions to the sample you get with the test labels in _common, if you have one more sample than you should delete field id 0. After this you'll still get bad results, you need to reorder your columns (including field_id as column 0) like: df = df[[0, 8, 3, 1, 2, 9, 6, 5, 7, 4]]. Pretty annoying tbh, so I hope there's eventually a fix for this!