I found that it takes some effort to make manual calculations match pnoise analysis. It is easy to forget about the assumptions you make for a manual calculation and it is also easy for a new SpectreRF user to set the analysis up incorrectly. Without seeing your pnoise set up, I only have the following comments: Are you sure the assumptions you make for hand calculations and simulations are consistent, especially regarding terminal impedances and interstage loading? Also, pss will account for noise mixing from other harmonics besides the desired ones, whereas the typical manual calculation may not. It is very very easy to end up with an errors of 3dB, which seems to be in the neighborhood of your error. You might try playing around with the number of sidebands you are using in your pnoise analysis. That might give you some idea of where any inconsistent assumptions lie.

Assuming the objective is to resolve the descrepancy between hand calculations and simulation, I would start by using a relative sweep only over the band of interest. I would also limit the the maximum number of sidebands to 1, and then maybe even use an array of sidebands and select only the + or - 1 side band. These changes may bring your simulation in line with your manual calculation. Then you can decide which is more appropriate. Also, as I said before, make sure your manual calculation accounts for changes in loading. For example, if everything in the chain is driven from 50 ohms and loaded with 50 ohms, the usual Friis formula applies. But if you start out with a 50 ohm system then down stream switch to high input impedances, you will have to modify the Friis formula.

i have another more interesting issue to share with. it is base on measurement, the measured nf is much less than the simulated one. don't know why after many people and a lot of time spent on figuring out how simulation should go, so the company is thinking of switching to other software. we are group of experience rf/rfic/mmic designer, just by forced to used this software.
i am looking for the solution of this discrepancy now, so don't need to switch software.

measure agree with manual. any suggestion in debugging the simulation? such as
-looking at the gain first,
-single stage nf,
-noise summary,
-transient,
-op, dc,
-etc.
but we are not spectre common user, that's why need some advice.