Flawed benchmarks have one or more substantive flaws we believe users need to be aware of to accurately interpret results, most commonly that >20% of the tasks have accuracy-impacting errors. In this case, we publish a limited writeup of the flaws we found.