Evaluate a screening signal

Compare a complete-run machine signal with the paper, then agree, disagree, or override each axis. Responses are retained as evaluation evidence and reviewed before they become calibration targets; one response is never treated as ground truth automatically.