I would have to agree that the numbers are illogical, even if you propose mail-in votes would be heavily favorable to Biden.

Two things I would expect to see are:

• The graph would never be quite that clean.
• A subtracted percentage is not natural, even if you would expect mail-in to be more favorable to biden. I'm a mathematician but not a statician, so I don't want to say exactly what kind of formula you should expect, but I could probably figure it out if I had more free time.

GP does seem to get ahead of itself quite often, so I might want to see some independent verification before getting too excited though.