I would vote for #1. Unfortunately, your boss has a bigger vote!
Maybe he is thinking this:
-30 dBm in a 200 KHz bandwidth is like saying -33 dBm in a 100 KHz bandwidth is like saying (-33 dBm -50 dB) = -83 dBm in a 1 Hz bandwidth.
If your spectrum analyzer can measure in a 1 Hz resolution bandwidth AND you are measuring pure white noise, you can set it for that and do the measurement that way.
However, you are making life very hard on yourself if the signal contains any discrete spurious tones in addition to white noise. You might fail the test using the "#2 method" then, while you might legally pass it using the "#1" method".
Example: you are looking 20 MHz away from the carrier, and you have the spectrum analyzer set at 200 KHz RBW. Lets say that at this offset you have white noise at -50 dBm/200KHz, and you also have a discrete tone at -40 dBm. The spectrum analyzer will simply integrate all the noise it can see in the 200 KHz RBW, so it will show a roundy bump with a top of -39.5861 dBm.
This is because -50 dBm is 0.00001 mw (from 10^-50/10), and -40 dBm is 0.0001 mw, so the total integrated power is 0.00011 mw, or 10 Log(0.00011) = -39.5861 dBm. Since you are 9 db within the "-30 dBm in 200 KHz" spec, you ship the unit!
If instead you set the spectrum anlayzer to 1 Hz bandwidth, as it sweep exactly to 20 MHz, it sees a -40 dBm discrete tone, and shows it at -40 dBm. This is 43 dB worse than "-83 dBm per Hz", so now with that interpretation you can not ship the unit.
I would read thru your specification, line by line, and try to prove to your boss that the spec is designed to measure both white noise and discrete spurs, and that your's is the proper interpretation. A valid way to do this is to make the measurement, show him that there is a discrete spur 20 dB out-of-spec, and the only way to pass the test is your way. Bosses like this sort of argument, as they want to ship the product!
A lot of systems, communications systems for instance, only care about the total integrated noise power, and do not care what percentage comes from white noise or discrete noise sources--just that it is below a total integrated level.
Rich