They claim their neuro-symbolic approach closes a 40%+ accuracy gap in reasoning tasks by forcing the LLM to construct a formal model rather than just predicting the next token. Curious if anyone has tried their CodeLogician agent yet, or if this is just more symbolic AI hype?