β For the views below to match the numbers, each model's GT + preds SLPs must come from the same train/val split as its metrics NPZ. Use the per-model GT slot on each card if your models have different splits.
In Both (compare) mode this applies to every loaded model β otherwise single-model views are unaffected, and per-model errors stay correct for whichever model you select.
Frame Coverage (instance count mismatches Β· GT + preds SLP required)