Keysight ADS EM co-simulation: port reference plane

I’m designing a matching network at TLs, and I’m observing a dependence of network input impedance on open stub length using ADS co-simulation. I tried it by two ways – first, I placed input port directly at point of the open stub connection. The second one is by using a TL section 10mm length inserted between input port and open stub connection point with port reference plane shift 10mm, i.e. to the point of open stub connection. The port feed type is selected to “Auto “. As I understand, there will be an automatic port calibration to reduce the impact of high-order modes. Hence, the results of both simulations should be identical. However, after the EM simulation I obtained completely different dependencies. The “TML” calibration also gives some different result for first case and evokes the error “Singular matrix found at row 1.” for the second case (with reference plane shift).

Please explain me, why is this happening and which result is correct?

Try making a very short input segment (but not zero length) at the Tee-junction, instead of placing the pin exactly at the Tee. I think your issue is that Momentum doesn't understand what port width you want, and uses the entire width - including the TL2 stub length - as the port width.

You could also use a controlled "drawn" port width, by using an edge pin with manually drawn size, as I described in this appnote:
https://muehlhaus.com/support/ads-application-notes/edge-area-pins