The open, short and load standards in the calibration kit must all have a mathematical model of them. For the load this is very simple - a 50 Ω resistor is assumed. The load must be a high quality, having a high return loss throughout the frequency range of the kit.
The opens and shorts are quite different. Each will be considered to have a transmission line will have some length L, typically of the order of a few mm, which causes a delay dT typically of the order of a few 10s of ps. A short will be terminated in an inductor, although the inductance of that is usually negligible. The open will be terminated in a capacitance due to the fringe capacitance, of the order of 100 fF. This fringe capacitance can not be ignored, and neither can it be considered independant of frequency. The capacitance is normally represented by a 3rd order polynomial as a function of frequency. HP, Agilent, Keysight and Copper Mountain all use the following definition
Cfringing=C0x10-15 + C1x10-27f + C2x10-36f2+C3x10-45f3
Rohde and Schwarz, Anritsu don't specifiy a delay of the standard, but a length. Care needs to be taken to be sure the units are correct. One of Kirkby Microwave's customers assumed the VNA wanted an offset length in mm, whereas actually the VNA was expecting the value in metres. Needless to say, having it 1000x wrong causes problems.