When a pipeline produces an unexpected result, the first questions are usually simple: Which data did it read? Which code ran? What parameters were used? If the answers are scattered across logs, shell history, and mutable object paths, reproducing the run becomes guesswork.