[amelius]

Just wait until LLMs are fast and cheap enough to be run in a breadth first search kind of way, with "fuzzy" pruning.