Just ran across this thread. I can't speak for all companies, but for Keysight (and this should not be construed as an official response), all kits of a particular model number have the same calibration coefficients, so all 85033E kits have the same C0, C1, C2.... There is another kind of kit called a verification kit, and those have verification artifacts that do have individualized measurements of each part. On sending a new disk: not sure, I'll have to check but from time to time we do change the values of the constants if we have a refinement in the calculation of the model, so we might send a new disc if the definition changed. They don't change often. I will note, for example that a kit specified to 6 GHz might have different constants than a kit specified for 18 GHZ even if the physical part is identical, because the open circuit capacitance cannot be adequately described by 5 values (the C values and delay). The polynomial does not exactly fit the Bessel function curve that mostly describes the open circuit capacitance (which is the key thing in the kit, along with the shorts' offset delay. The set of coefficients for the 6 GHz kit fit the 6 GHz performance best but are not good at 18 GHz. The 18 GHz fit well over the whole band but are slightly worse at 6 GHz. Above 50 GHz, the polynomial fit falls apart rather badly so we use databased fits. But all the data-based numbers are the same. We do have very specialized kits called CD (characterized device kits) we use internally and occasionally sell to NIST or NPL or PTB. They are individually characterized. For the most part, the specifications are derived from the mechanical dimensions of the kit which are controlled so well that the constants for each kit are essentially the same (precision Swiss screw machines then precision plating). The biggest thing that changes the performance is damage to the slotless connectors, and physical distortion due to over-torqueing which can change the pin-depth and cause the model to not be correct.