After Two Corrections, Usually Start Over

It gets it wrong. You correct it. Still wrong. You correct again, and the third attempt is, more often than not, somehow worse.

That's not a fixed law, just a pattern I keep seeing: every failed attempt stays in the context and shapes the next one. You are not steering by then. You are adding noise on top of noise.

The heuristic I use, usually and not always: after two failed corrections, clear the session and write a better first prompt, one that carries what you just learned. "Don't use approach X; the constraint is Y."

A clean session with a smarter prompt beats a long session full of accumulated corrections, in my experience almost every time.