I think it would be interesting to re-run experiments / data analyses (ones sensitive to FP correctness) to see what holds up. I imagine that would be a pain though; reproducibility isn’t usually easy.