I am trying to accurately measure the parameters of some adapters. While checking the accuracy, the S21 error is the largest when measured in transmission.

When I would perform the same measurement in reflection, by first calibrating the port, and then calibrating the port including the pin-saver, I can make T-parameter networks of the error terms. Then by applying T_nwaport^{-1} * T_pinsaver, I should also be able to get the full set of S-parameters of the adapter (except for the sqrt(S21S12) issue). I suspect this is the same as is used during adapter removal.

My simulations of the errors seem to indicate this would give me better accuracy. Furthermore, when the adapter is small compared to the wavelength, its T-paramers approach the identity matrix. In this case the effects of directivity, match and tracking terms seem to be suppressed significantly (assuming the terms remain constant).

Could somebody comment on this procedure? Or iow. what is my failure in thinking?

I was also wondering: are the error terms between the male and female adapters of a 3.5mm kit correlated, or should these be treated as independent sets?

When I would perform the same measurement in reflection, by first calibrating the port, and then calibrating the port including the pin-saver, I can make T-parameter networks of the error terms. Then by applying T_nwaport^{-1} * T_pinsaver, I should also be able to get the full set of S-parameters of the adapter (except for the sqrt(S21S12) issue). I suspect this is the same as is used during adapter removal.

My simulations of the errors seem to indicate this would give me better accuracy. Furthermore, when the adapter is small compared to the wavelength, its T-paramers approach the identity matrix. In this case the effects of directivity, match and tracking terms seem to be suppressed significantly (assuming the terms remain constant).

Could somebody comment on this procedure? Or iow. what is my failure in thinking?

I was also wondering: are the error terms between the male and female adapters of a 3.5mm kit correlated, or should these be treated as independent sets?

And you are also correct that it is -exactly- mathematically equivalent to Adapter Removal of the 8510 and 8753. That is, all the terms but the two one-port terms drop out (in fact, I won a little bet on that fact!).