I'm growing increasingly confident that this is how people often work, as well.

Yes. But although they can't know "why" a specific "wrong" answer was selected, the response is often still informative, and it can highlight real weaknesses in process or code structure that should be addressed anyway.

To test this, change the history in the context to indicate that the model did or recommended something completely different than it actually did, and then ask it to explain why. You’ll still get a plausible explanation.

True, but even a hallucinated explanation of where things went wrong added to the context can force the model down a better path over the next few inputs.

The point of the parent post is that the explanation shows they made the error themselves, so it's immediately validated.

You can also literally tell them: "Here is your session ID: $ID, lookup the .jsonl session, trace exactly why this decision was being made, present evidence and concrete proof, no guessing or assumptions" and you'll get an evidence-based report without guesses.