I've been trying to play with this, but am frustrated. I'd imported a csv version with 8215 non-zero time stamped state (no DC) entries into database software and was able to sort them by time within each state series. Broke the cumulative vote totals by the included 3 digit dem. and rep. shares and assigned the reminder to a generic 3rd party column. Computed the change in dem. rep., 3rd, and total vote for each incremental state entry compared to previous timestamp. Figured switch and losses would show up as negative increments, but there were lots of negative increments, some Trumps, more Bidens, many 3rds and 38 in which the total vote count dropped. I see the data pointed to as Antrim, MI but am missing something in terms of how to separate "switch" and "lost" from merely negative increments. I WANT this to pan out, but we can't afford to make claims that don't hold up when other candidate fraud claims are available.
I see sources for the data in JSON format and my db claims to be able to import JSON, but it refuses to import these JSON files and I don't know enough about JSON to understand why.