Our Failure Cases in the
Closed-World Benchmark Datasets