IAWYC, but I disagree with this:

"Carl Shulman would be heartbroken if I didn't mention Philip Tetlock, so here goes."

One interesting proposal of Tetlock's is for an 'adversarial collaboration' where he and the IAT team assign Bayesian probabilities and how they will update in response to specified evidence, then conduct the specified experiments.

Negotiations apparently broke down, which seems to be the usual outcome for attempted high-stakes adversarial collaborations, but I would like to some of them succeed.