There’s generally three failure modes for a physical activity:

For a slower activity like climbing, these failure modes can be identified in the moments before a move. One can loop through these and spot the most likely failure.

It’s like a premortem but happens within the context of the activity, not in a carved out spacetime beforehand. It’s obviously not like post-event learning, where you fail then update. But it’s not quite as integrated as predictive processing, either, within which perception and action are much more tightly coupled. The starting vector itself is wrong but gets corrected mid-move.

I suspect this intra-action corrective might be a differentiator for experts versus non-experts. In contemplative practices, it’s said that the true practice is getting back on the path, not avoiding being knocked off it. This feels like that; it is the ability to save a moment even when it begins like one that typically would be lost.