Often, one finds that some nights of a run were of marginal photometric quality. There may have been clouds or cirrus visible; or the residuals may simply be anomalously large. How should these marginal nights be treated?
The safest thing is to reduce all the nights of a run together at first. If one or two nights then have a large number of rejected observations, try to decide whether the problem is instrumental or transparency problems. Nights with instrumental problems should probably be rejected completely. Nights with variable transparency may sometimes be salvaged, either by throwing out the worst part of the night, and reducing the rest together with the good data; or by removing a dubious night from the solution, and treating it separately.
In treating a bad night separately, one can either force the extinction coefficients to have particular values, or force the standard and extinction stars to have particular values (by creating a special standard-star file). Much of the time, the best thing to do with bad data is throw them away.