I see no reason to believe that, given his process for ripping an "equal" number of commit messages per language was broken, that anything else even approaches validity. It's simple arithmetic; a grade schooler who notices that the last number is 7 would realize something's off.

What about the process is broken? Did you read the code and find bugs? With a total commit count of 929857 missing a single commit to round out to a perfectly even number of commits in each language is insignificant.

Or he had 929857 commit's and then he randomly sampled an equal number for each language. Thus, no division etc.

Then, of course, you have problems of sample size. Nearly a million commits is a pretty good sample size; a hundred, not so much.