Comment by sesm

I wonder, did they consider an approach of vibe-coding a deterministic converter and then running it? This should be much more token efficient.