Then why not turn this around and say which (hopefully small set of) specific classes of curves are known to be avoided, but otherwise have a free for all?

> If that many curves are bad, then ECC was a mistake.

I mean that is the logical conclusion from the behaviour around this approach where a few good things are effectively endorsed without justification.

The fact is any application of an algorithm that involves so much "trust me bro" on the selection of what are a large number of opaque quasi random but not quite or provably so numbers only has as much trust in it as you have in the people that created it.