In that case we could just match by the CSS rules' content - i.e. match all elements that have style X, Y, Z applied to them. That kind of arms race would be over real quick.

The arms race is that the next time you come back to the page, the offending element has a slightly different style (randomized CSS class name, RGB values changed by 1, font size changed by 0.1...) and your browser doesn't recognise it as being the right element to block.

Match by ranges so changing values by 1 doesn't work. At the end of the day a human would have to come up with an entirely alternative style for the page to defeat your filters.

Pretty much any machine generated variations can also be defeated automatically by a machine. If defeating your filters takes a lot of work, and only helps until a few days later you update your filters in a few minutes, nobody is gonna bother.

Whitelist, not blacklist.

Which whould break CSS minifiers, but that's the point.

That's already the case with sites like Twitter. In order to remove the trending pane in uBO, I had to rely on a11y attributes:

##div[aria-label="Timeline: Trending now"]

I hate to tell you this, but we're already there.

You can have filters that look at parent or child elements of the randomized element or other patterns.

And random structure to break nth-child selectors.

Even better: draw everything in a canvas.