Holdout (or out-of-sample validation) accuracy; null when not reported.
Accuracy of always guessing the most common class on the same rows (sidecar baseline_accuracy).
True when the finding graded the model Poor: rule enforce, or an accuracy-only verdict under either rule.
One plain sentence with the numbers.
How many distinct classes the model predicted on those rows (sidecar predicted_classes).
The rule recorded in the model's Lineage.evaluation ('enforce' when the lineage has none).
The majority-baseline check's result on one model (decision D2-C5d).