Judged is how many rows carry a disposition. It is zero until a label plane writes one, and reporting it plainly is what lets a model plane refuse to rank rather than name a winner it cannot justify.
Rows is how many rows the version holds across every split. It is the size of the version, not of the source window — the horizon, the cuts and the row cap all bind before this number.
Subjects is how many distinct subjects the rows belong to. Every row of one subject is in ONE split, so this is the real sample size — the row count flatters it whenever a subject is active.
Unproductive is how many carry the other. With Productive it accounts for Judged, so the class imbalance is visible before anyone trains on it; both stay 0 while Judged is 0.