It's plausible. To see 100% accuracy esp on well studied datasets or small set of new data. On a 20 example test set in the wild i witnessed exactly this. 20: is super small. The rules forbid me from using more .