What's a system you stopped trusting once you understood how it worked?

Standardized testing. Specifically: how much of the score variance is explained by test-taking strategy training rather than the underlying construct.

Insurance underwriting. I used to think the actuarial models were neutral. They're not — they're optimizing for portfolio variance, which is a perfectly legitimate goal that has nothing to do with what most policyholders think they're being measured on.

Hospital quality metrics. Patient satisfaction scores correlate weakly with clinical outcomes and strongly with how comfortable the waiting room is.