Thanks. Interesting for me is that the LLM and the parser have inverse failure mode, the LLM nails intent and misses structure, and the parser is the opposite.