Perhaps another approach would be to retain logs on clean-all if the previous pants run was a failure? Another approach would be to more aggressively record information on failures in a durable way.