Please ... The evidence here is pretty weak. The main evidence is that the total N votes drops off a cliff a number of times, despite being sorted by timestamp. Simple answer: The data quality from the third-party commercial API, Edison Research, is of low quality. Or the NYT feeder parser has a bug (e.g. they collect the data in a set which loses ordering when it's dumped to json on disk on server). Fraud is by far the least likely explanation.

Please ... The evidence here is pretty weak. The main evidence is that the total N votes drops off a cliff a number of times, despite being sorted by timestamp. Simple answer: The data quality from the third-party commercial API, Edison Research, is of low quality. Or the NYT feeder parser has a bug (e.g. they collect the data in a set which loses ordering when it's dumped to json on disk on server). Fraud is by far the least likely explanation.