Create another reference set of 50-100 labels

[ntatonetti]

Manually curate a new set of FDA labels (~50-100 should be sufficient) according to the same process that the Denmer-Fishman paper performed. Use this set to independently validate the performance metrics we got from the gold standard.

This can be broken down into two parts. First, someone needs to review the process in detail and come up with a strategy of how we'll do it. Then as a lab we can dedicate a couple of days to the manual annotation.