basically I think the only criteria that matters for describing ability is how well you perform
I think this is a short-sighted view of ability. It gives you an instantaneous view, but it doesn't easily make provision for learning. For example, if a player tries a new strategy and they lose, you might say "ah, not performing well - this means low ability". But it's likely that player learned something from the new strategy that will improve their overall game.
Let's look at it from another angle. Take a player who pours all their attention into the game, but isn't very creative. So they have a couple of very strong BM strategies and they're really good at the obvious engines. They win a lot - at the things they're good at. But, by focusing so much on a narrow portion of the game, they've limited their ability to perform in the subtler scenarios. I think this shows that choosing your personally-strongest strategy for each game does not necessarily make you a better overall player. Sometimes the lessons that come from experimenting and losing are more valuable than another notch on the belt.
You seem to think that experimentation and trying to give yourself the best chance to win are mutually exclusive. This isn't true at all, and I don't know why anybody would think that. And of course I think people can learn from their mistakes, you don't have to play absurd strategies to make mistakes though. Every single game has instances where your play can be improved.
Other thing: basically I think the only criteria that matters for describing ability is how well you perform.
Yes, but how you perform when is the issue. If I'm assessing a sprinter, I want to know how fast he runs on the track in big competitions, not how fast he runs when he's chasing his kids around the park. In the same way if I want to know how good a player is at Dominion I want to know how good he is when he is actually trying. I love Isotropic to bits, but that doesn't mean it's the be all and end all of Dominion. Some people blatantly care about (and go to great lengths to increase/maintain) their isotropic rating a lot more than others.
I don't know of a tool for measuring a person's effort. So, until I find one, I will continue to use performance as my standard of ability. This doesn't mean I think it's perfect, I think fluctuations will be common, and that's why we have upsets.
Nobody looks at the dad in the park chasing his kids and says "that guy must be a world class sprinter, he caught that 5 year old without breaking a sweat". So when he shows up and beats Usain Bolt, we are surprised. That's the nature of upsets. But we don't think his ability is high until he demonstrates it.
Also, I'm not sure who you're thinking of with your comment about isotropic level and the "great lengths" people go to increase/maintain theirs, but with the exception of a few cheaters, as far as I can tell the only methods anybody uses is "play some games of Dominion". I have not (yet) developed a complex training schedule/diet. I suspect -Stef- is genetically modified though, so yeah, that is taking it a bit far.