Question about the result

Hi, could you please confirm if the "predict-add.json" in the Adveta dataset is complete? There are only 100 data samples, but the reported accuracy in the paper is 68.1%. If there are only one hundred data samples, shouldn't the accuracy be an integer?