pdf-lab

/pdf-lab is a convergence loop that diagnoses PDF extraction failures, reproduces them on synthetic PDFs, discovers optimal parameters, and writes those fixes back to the extractor pipeline code permanently.

Why This Exists

There is no absolute ground truth for real PDFs. The system works with deltas between S00's estimate and actual extraction results. When S00 predicts 40 sections but extraction finds 15, something is wrong. /pdf-lab figures out what, fixes it, and makes the fix permanent.

Quick Start

cd ${HOME}/workspace/experiments/pi-mono/.pi/skills/pdf-lab