[dan-robertson]

I think the bug was that your first sheet came from a small set and the people entering bad data would refresh instead of doing the actually random next sheet, so entries for most of the sheets came only from people who had long sessions who were apparently more likely to enter good data.