Looking at the actual Chicago data at https://www.chicagoelections.gov/en/election-results-specifics.asp by precinct as of late November 7, the charts for Chicago look credible but the assumption that Benford’s law should apply do not, at least for Biden/Harris or the minor candidates.
Of the 2069 precincts (most of which are of broadly similar size), Biden/Harris won fewer than 100 votes in 12 precincts, and more than 999 votes in 4 precincts. All the rest (more than 99%) had three digits for their votes, violating the requirement that natural data satisfying Benford’s law should span several orders of magnitude. More than half the precincts (1100) gave Biden/Harris from 300 through to 499 votes, making 3 and 4 the most common first digits (the chart reflects this and is close to showing the actual frequencies by hudreds of votes, so 300-399 was the most common).
For Trump/Pence, votes were more widely dispersed: 99 precincts with 1-9 votes, 1339 precincts with 10-99, and 633 precincts with 100 or more votes. This dispersion over orders of magnitude allowed a greater chance of coming closer to matching Benford’s law.
For the minor candidates, they only reached double digits in a very small number of precincts (and got 0 votes in hundreds of precincts – not shown on the charts) so the charts are close to showing their actual vote distribution with censoring of 0 and 10+; again you would not expect Benford’s law to apply.
Chicago was an odd choice to investigate for suspected cheating in 2020 where the gap in Illinois was 12 percentage points (1960 when it was 0.2 percentage points might have been more interesting). I suspect it was chosen simply because the data is publicly available and the distortions caused by similar precinct size led to this non-Benford law result. You will see this elsewhere for similar reasons: in 2019 very few British MPs won a number of votes starting with 5-9, as their constituencies are of broadly similar sizes and the winners usually got in the range from 10,000 to 49,999 votes, again failing the spanning several orders of magnitude requirement.
