Two independent ways to see whether scan values behave like real genetic differences. Both are diagnostic only: they report on scan values that already exist and never change any animal's value.
Every model is trained on exactly one target and is named after it. A model trained on marble scores gives a marbling grade, never an IMF %.
Progeny scan values (adjusted for contemporary group) are regressed on their sires' EBVs. A sire passes on half its breeding value, so the slope should be about 0.5 when the scan value and the EBV are on the same scale. EBV accuracy affects how certain the estimate is, not where it should land. At least 20 sires and 100 progeny are needed before a result is shown; even then the 95 % interval is about ±0.35, enough to tell a real genetic signal from none, but not to fine-tune.
Genemap's own EBVs are accepted only from an evaluation that left out every ultrasound record (no US_* trait); otherwise the scan values would be checked against breeding values partly built from those same scans. Runs that included ultrasound traits are listed but cannot be chosen. Breed association EBVs are used only for sires whose EBVs no Genemap run has overwritten.
A scanned mob's predicted values compared with the farm's own MSA kill history (same scale only) and with mobs scanned within 120 days. A mob is flagged when its average is more than 1 standard deviation from the reference, or its spread is less than half or more than double. Mobs are chosen groups, so ordinary differences are expected — a flag points at something to check (scanner settings, the wrong animals), it changes nothing.