seamaster:
i was concerned (or should I?) about drift.
Do you mean the vessel drifting on its anchor, or drift of the readings from the '6050 sensor over time? If the latter, then drift is not a problem. The '6050 has an accelerometer, so it always knows exactly which way is down (in calm water, at least).
seamaster:
the accuracy of 0.5 deg is plenty. sampling as often as needed, but the output 2 to 5Hz.
Well, that's trickier. You seem to be asking for the swell-compensated tilt reading to be able to react very quickly, in a much shorter time than the wavelength of the swell, contradicting the subject you put. This will mean the Arduino must analyse the swell in 3 dimensions, determining wavelength and amplitude, so it can remove the effect of the swell from each reading at any instant in time. An interesting challenge to be sure! You'll need someone with a Maths degree to help you!
If the output could be slower, i.e. much slower than the wavelength of the swell, like your subject suggests then it would be a simple matter of averaging the readings, but that would have to be over a period of at least a minute I would have thought.
Paul