Firstly - thanks all very much, especially HMS1021 for the replies.
When still putting together the list of equipment, clearly the "guaranteed" specifications are very much conservative.
A number like
+/- 1dB can be applied to kit that might well perform to
+/- 0.2 dB when actually delivered.
When doing the system analysis you RMS all of the component ripples together, this actually works in practice.
This, for me, is counter-intuitive. If I connect (say) a filter in line with a broader band amplifier, the sweep response is the straightforward sum of the individual responses. Two droops together make a bigger droop.
For design purposes, where one cannot know if the ripples coincide, but are maybe (randomly?) distributed across the band, then I guess the RMS sum of the specifications is a valid estimate provided one does not get unlucky! Again, the data sheet values need to be discounted (carefully!), because they reflect excessive manufacturer caution.
Since group delay is such a driver our RF and IF bandpass filters are fairly wide and are used to mitigate out of band interference and mixer spurs and images.
I take the point that group delay from using filters with too-steep attenuations too close to the band edges, and group delay response from various kit in general likely has more impact on messing up modulated signals than relatively small amplititude variations (ripple) in the passband response. Often the customer will be asking for filters that may include all sorts of tricky sections to make the response steeper on one side, or include a reject notch very close, as if one can make a passband begin directly where a transmit-reject band ended, with no room in between.
One thing I do say is to pay careful attention to the "+/-" in the specification. It somehow easily gets glossed over, or even left out of the client requirement. I can see why, but I am always surprised at how often I have to be very clear on this.