The BDF pipeline is designed to take raw bias, dark, and flat files and combine them into "master" files. Conceptually, this involves taking all the files of each type taken during a single night and processing them into a "first-level" master file. It would be possible, then to combine masters from different nights to obtain a "better" product-- but whether or not that yields better results is TBD. There can be (at least) two approaches to applying the masters in the RAW pipeline. One could just take the newest "acceptable" B, D, and F files, or one could attempt to combine files from multiple nights (e.g., by re-computing a larger set after new data are added or by combining "master" files from different nights in some kind of (possibly weighted) average.
Things which should be in the output fits header