Assuming the data from this story is the data from the NYT election scrap data (which I am 99% sure it is) it is a decimal because of how it was calculated. In other words, this was an ANALYST error, not an error of the voting machines or reporting software. The scrap data was NOT the actual data. The actual data was from Edison. The Edison data was similar, but more accurate (and all whole numbers). The scrap data was only displayed to three decimal places. So same data, but less accurate. The scrap data was given in the format of:

Date, Total votes, .xxx, .yyy, .zzz

where .xxx is the percent of of the total vote for Biden, .yyy is the percent of the total vote from Trump, and .zzz is the percent of total vote for other candidates. Again, assuming this is the scrap data, the "# OF VOTES" column is (in this spreadsheet) calculated from the Total votes (current line minus previous lines of the Total votes column). It is because the scrap data was only given to three decimal places (and because of how multiplication works) that all of the other columns are decimals. This has been an egregious oversight by many "analysts."

This error has given a lot of bad analyses. Even real mathematicians have made this mistake. Unfortunately mathematicians and basement analysts often don't work with actual data and make such rookie mistakes, drawing false conclusions from their own calculation errors.

Don't get me wrong. There are PLENTY of REAL problems with the data that show it is not data from an election. I spent almost a month analyzing tons of election data and proved it was impossible, but this particular problem is analyst error, nothing more (if I am correct on the data source).