cmsy10.tfm gives a positive depth to the minus sign:
$ tftopl /usr/local/texlive/2019/texmf-dist/fonts/tfm/public/cm/cmsy10.tfm | grep -A4 'CHARACTER O 0' (CHARACTER O 0 (CHARWD R 0.777781) (CHARHT R 0.583334) (CHARDP R 0.083334) # <- emphasis mine )
(the minus sign is at position zero, as confirmed by:
$ grep 'C 0' /usr/local/texlive/2019/texmf-dist/fonts/afm/public/amsfonts/cm/cmsy10.afm C 0 ; WX 777 ; N minus ; B 83 230 694 270 ;
Where is there this positive depth? The minus sign obviously has no descender.
This is problematic for the Python plotting library matplotlib, which needs to figure out the baseline of certain tex constructs to properly align tex-processed strings -- see https://github.com/matplotlib/matplotlib/issues/6323 for the issue, and https://github.com/matplotlib/matplotlib/pull/6333/files for a proposed fix which basically amounts to special casing this glyph to ignore the tfm data.