ImageNet has five orders of magnitude more answers, which I would assume makes QA a completely different category of problem.
The authors could probably have carefully review all ~300 of their questions. If they couldn't they could have just reduced their sample size to say 50.
I admit that Imagenet isn't the best analogy here. But I'm pretty confident that this data cleaning issue would be caught in peer review. The biggest issue which I still don't understand was the removal of the test set. That was bad practice on the authors' part.
In general with evaluations LLVMs I keep seeing issues that would be caught by a human carefully looking at the results. To a nonexpert who occasionally peeks at things it seems like having a small dataset that's just slightly too big for you to manually review is a bad practice.
It also seems like 100% accuracy should have raised red flags, especially if you know your dataset isn't perfectly cleaned.
Most of the damning claims in the conclusion section (Obligatory: I haven't read the paper entirely, just skimmed it.) usually get ironed out in the final deadline run by the advisors anyway. I'm assuming this is a draft paper for the EMNLP deadline this coming Friday published on arxiv. So this paper hasn't even gone through the peer review process yet.