[martimchaves]

I'm guessing that human accuracy may be lower or around that value, given that handwritten notes are generally difficult to read. A better metric for document parsing might be accuracy relative to human performance (how much better the LLM performs compared to a human).