Todo's sorted by priority:

(1) Should we spend time to improve BNC's convergence? Should we follow a
procedure like the one Oscar implemented? If not: Can we do something which
helps a bit and does not involve too much work? Or should we better do nothing?

(a) Calculate, for each satellite, the STD of the code multipath in the
iono-free combination Pc, as a combination the carrier phase and the code.
Do this continuously, using a circular register that contains the observations
of the previous 10 minutes.
(b) Introduce a default STD that is used when, either at the beginning or
after an interruption in the data, there is less than 10 minutes worth of
data for finding the multipath STD as in (1), and the STD with insufficient
data is less than this default.
(c) Introduce use a "fudge factor" to multiply the data-based STD and get the
actual value used to get the data. This is necessary, because the correct weight
of the data depends on numerous factors, particularly the weights assigned to
the phase and to the a priori values of the unknowns. This problem also is true
for (b), (c), and (d). So the fudge factor really depends on the overall strategy
and cannot be used in different software implementations.
Everyone has to find it the hard way, by trial and error, making lots of solutions.
(The variance of unit weight is of very limited help for this purpose.)
(d) Downweight the code by a large factor when a satellite is below 20 degrees
in elevation, or the 3-D precision of the coordinates being estimated falls
below something like 25 cm (again, this will be a different threshold with
different software).