[gonc]

Curious about the OCR + Vision pipeline: do you run frame sampling, or analyze every frame? How do you deal with false positives when extracting gem names or code snippets?