2 alternative interpretations to consider:

1. 30 was a typo, it should have been 20.

2. 30 was the full suite, but they only ran 20 of those 30 on Nov 1st and so ran the same corpus again for a fair comparison.