Hmm any way a similar method could be used to do efficient extraction from other things besides agent traces?

I’m thinking logs, documents, etc.