"To properly test that script, we are of course just going to feed it back to another LLM that is supposed to score in the same general area as the DeepSeek Special model. That of course being Gemini 3 Pro. Google Gemini 3 Pro said the script is mathematically sound."