I'm trying to find out whether back-and-forth between two different LLMs improves task success beyond strong alternatives under a controlled resource budget — and I'd rather hear that it's already settled than spend money finding out. AI-assisted throughout: the protocol was developed and critiqued
