A lot of things actually 1. Polyclaw setup to do system checking, before every trade it came back to audit the risk, which it wasnt supposed to 2. Any analysis of logs and action, it always came back again for human in the loop for every single one I could have solved it with Claude or GPT maybe with more hardening adding more examples when not to ask me With Minimax I didnt need to and still get the same quality per token as these models Well now use Minimax for everything