In the round 2 datasets, the numbers in the truth files, which are supposed to be TPM (transcripts per million), don't quite add up to 1 million. Here is what I see: sim31: 999,999.7 sim32: 999,999.8 sim33: 1,000,000 sim34: 1,000,000 sim36: 1,000,003 I know these are small discrepancies, but it suggests a bug somewhere. Also, one of the aims of the contest is to evaluate how well programs can measure low levels of expression, so these discrepancies may have a greater effect on lower-expressing transcripts.