Grubbs outlier test can remove large outliers as well, so if there are few large undercuts and many small ones, it will remove the large ones unexpectedly.
For now the code has been reverted to a strict smallness test.
We need better statistics for removing outliers, i.e. spurious data.