Run the DC test docs through the live API and visualize the results: classify → extract, manifest vs expected, and column→field mapping with values side-by-side against ground truth. The scoreboard tracks the latest run's accuracy on the critical fields.