Usually, the reason we do SI simulation is to verify compliance with some specification. If the quantity (overshoot, timing, etc.) is specified at the device pin by the vendor, then you need to measure the signal at the pin. This is traditionally the location that IC vendor use in data sheets. That is why HyperLynx defaults to measurement at the pin. In newer specification that realize that designers will simulate, the specified quantities are defined at the die. That's where it really matters, but before simulation was popular, it simply was not practical to define requirements this way.
So, measure at the pin or die to match the method used to define the constraints you are verifying.