When you are manually deriving a feature and you digitise an edge, loop or face, then the depth of the feature is set between the highest and lowest points on all the faces that neighbour this, as in this example based on profiles from edges .
However you can change these; at the next step you are prompted to 'Digitise one or more vertices to specify depth range (or return)'.
EdgeCAM guesses what point you are altering:
If you digitise near the highest point, you alter the highest point.
If you digitise near the lowest point, you alter the lowest point.
If you make two digitises, the depth is set between them.