Paul is right that there will be lots of opinions. Here is mine based on many years of imaging work.
My calibration steps involve having a library of dark frames of different exposures so that I never have to scale darks. Since I don’t scale my darks I don’t need bias frames. I generate the darks when it is cloudy and generally average 30 – 100 raw dark frames at each exposure to get my library dark (I do 5, 10, 15, 30, 60, 120, 300s darks). My camera has good temperature regulation and so I can use dark frames from 6 months ago and they are fine. If you have a CMOS camera you probably will find that scaling darks is not always successful so, again, it is a good reason to keep a library of darks at different exposures rather than trying to scale them.
I have always used sky flats, again taking 30 – 100 raw flats, subtracting the flat dark and then normalizing and averaging them to get the master flat. Since the right conditions for making twilight flats don’t come along very often I tend to re-use flats for a month or so. I’m lucky, I have a permanent observatory and the camera never comes off the telescope but new dust spots appear with monotonous regularity. You can see an example of that in the lower right of this image.
When you do the averaging is best if you can use floating point format output FITS files rather than 16-bit integers. This applies both for the calibration frames (flats, darks) and for your final stacked images. Certainly don’t using summing with integer files since you will end up saturating brighter stars as their summed pixels hit 2^16.