If a main peak and a child peak are separated by a valley, the child peak is a rider peak. If there is no valley, but there are three inflection points, the child peak is a shoulder. However, depending on your data and on the integration settings, a child peak may appear as a shoulder although it actually is a rider.
This interpretation as a shoulder can occur if your main peak is rather tall, and the child peak is small compared to the main peak. In this scenario, averaging data points and combining them into one can level out the valley between main peak and rider peak. The higher you choose the Peak width integration parameter, the more data points are averaged, and the higher the risk to interprete a small rider as a shoulder.
To proof that a peak is a shoulder, use the following integration parameters:
Tail peak skim height ratio = 0.0
Front peak skim height ratio = 0.0
Shoulders mode = Tangential
Height reject = 0.0
Area reject = 0.0
After processing the data with these settings, every nested peak with a straight line is then definitely a shoulder. You can also verify this by with the fourth character of the baseline code: f or b for shoulders with a tangent baseline, and F or B for shoulders with a drop line.