[In the system design review, Richard Hills suggests] substitute an 
estimate for the atmospheric residuals containing three terms:
1) the contribution from noise and short term instability in the 
radiometers;
2) a term proportional to the uncorrected path fluctuation, to allow for 
the uncertainties in the atmospheric conditions and modelling and the 
errors in the calibration of the radiometers;
3) a term to allow for atmospheric path fluctuations which are not 
measured by the radiometer at all (as discussed, the principle item here 
is the "dry" component due to temperature - or more precisely density - 
fluctuations).

For the first term I suggest you take the specification, which is
10(1+wv/1mm)microns and equals 19.6microns (65.3fs) for the 5th 
percentile conditions you have adopted.  This is a long established 
figure and I think it is best to stick with it, even though, as I shall 
describe below, it is quite conservative.  (Remember that this is a 
limitation on sensitivity, i.e. it corresponds to an error that would be 
there even if there were no fluctuations at all in the water along the 
paths through the atmosphere.  There is no reason to increase it because 
of uncertainties in the modelling, etc.;  those go in the next term.)

For the second term I again think that you should take the value of 2% 
that we have accepted as a specification.  This is a good deal less 
conservative because, as indicated above, it is intended to include all 
the uncertainties in the conversion from changes in measured brightness 
temperatures to the changes in path due to water molecules.  We think 
however that this is a reasonable figure for the "good" conditions that 
we are considering here and assuming that some effort gets put into 
optimizing the phase correction when we have some real experience on 
Chajnantor.  This term is in any case not significant here: 2% of the 
143fs of uncorrected phase fluctuation is only 2.9fs.  Adding this in 
quadrature to the 65.3fs above gives just 65.4fs.

It is difficult to give an estimate for the third term - the atmospheric 
path fluctuations due to things other than water.  As we have discussed 
previously, these may get quite large when there is strong convection 
driven by the Sun heating the ground in the afternoons.  Alison Stirling 
has started some work on this problem from the theoretical end, but it 
will be a while before we can draw any conclusions.  What we are trying 
to establish here, however, is a value for the "best" conditions, which 
will certainly be much lower and for which the physical causes are much 
less clear.  I suspect that the only way to get real numbers will be to 
fly balloons or kites on the ALMA site and measure the temperature 
fluctuations directly.  We can get some guidance from optical seeing 
measurements, but the problem there is the extrapolation of
these from the scale sizes of ~1m that they involved up to those 
relevant for ALMA, together with the vexed question of whether or not 
there is an "outer scale" on the density fluctuations.  Better 
information should soon be available from the optical interferometers 
(especially the VLTI on Paranal, which has ~100m baselines and is at 
least in the same general area even though it is 2500m lower down).  I 
will try to find out if they have any values yet.

Meanwhile I suggest you take a value of 10microns (33fs), which I freely 
admit is no more than an educated guess, but again, I do not think we 
can take speculation that it may be worse than this even under good 
conditions as a reason for setting a softer spec on the electronic 
stability.

This should again be added in quadrature (it will not be correlated with 
the radiometer noise) giving a total of 73.3fs.  Applying your 
methodology then gives 37fs for the structure and 64fs for the electronics.