@DemonSynth hm yeah worth trying something like this - it always seemed intractable to score many options on many criteria, but if the LLM can fill in reasonable scores then maybe...