Output-referred kT/C noise combined with T&H and with sw.cap integrator

Best is to read the attached .pdf, but here a very brief summary:
*) For kT/C noise combined with a Track And Hold, equations provided by Ken Kundert (thanks a lot!) for the T&H fit perfectly to Cadence sim. results. (Adding his "Ideal S&H” makes no difference here.)
*) However, if I add a switched-capacitor integrator, NOW SUDDENLY I HAVE TO TAKE THE EQUATION FOR A SAMPLE & HOLD (S&H) to fit the Cadence sim. result (at least up to fs/2). Adding his "Ideal S&H” in this case does make a difference.
=> What is from the concept point of view the correct procedure / flow diagram for obtaining the correct discrete-time total integrator output-referred kT/Cs noise and why?
Example: Generate kT/C noise ==> Process it with the T&H and/or S&H (??) ==> Multiply by integrator transfer function ==> Do sampling (again)?!

Up to my knowledge, it seems, no one has conceptually described this in an understandable way up to now.
As said, I would be VERY happy if one could bring some light into this topic!
Many 1000 thanks in advance and best regards,
Bernd


(Reference = Paper used: "Simulating Switched-Capacitor Filters with SpectreRF - Ken Kundert")