He claims that any regular linear relationship at all would be problematic, though.

He describes the expected case as being a flat line.

So according to that logic, a positively sloping line should also be an indicator of fraud.

It wouldn't be. It'd just be an indicator of there being a >=1 slope in the correlation between split and straight ticket votes, as you've described.

If a negative slope were proof of Biden stealing--why would a positive slope not be proof of Trump stealing?

The point I'm making is: there would be some linear slope no matter what--no fraud necessary.