Did the model fail, or did the environment?

A runnable evaluation example separating wrong answers, memory limits, timeouts and setup failures, with all 27 attempt records retained.