The Metric You Optimize Has a Dark Side You Didn't Name

Confidence: hypothesis. The failures below are drawn from published research ethics guidance and widely reported industry patterns, not firsthand deployment. The pattern at the end is an argument, not a shipped system. Argue with it.

Give a system one thing to maximize, and it will find every path to that thing, including the paths you did not want it to take. This is not a flaw in the system. It is the system working. The flaw is in the objective, which named a goal and forgot to name the harm.

Country selection for a clinical trial makes this concrete and serious. If you reward the choice on enrollment speed alone, the optimizer has a clear and ugly path to more speed, and it will take it.

The clinical why

Patients tend to enroll fastest in places where they have the fewest alternatives. Where care access is poor, a trial can be the best treatment available, so people join quickly and in large numbers. That is a real and uncomfortable fact. It means a country-selection process rewarded purely on speed will gradient, quietly and reliably, toward populations that are enrolling fast because they are underserved.