As an avid stock DSLR APer, I find calibration to be exceptionally important, for very specific reasons. My goal is usable data at 100% scale, so that is my point of view.
1. Flats. Or better yet, flat—as RawTherapee uses just one flat. I don't have a lot of dust motes in any of my setups, so the lens vignette model is the most critical part of the calibration. It's not that I can't create a synthetic flat. It just saves me hours of processing, especially in tricky sky gradient conditions. Better to fight one largely linear light pollution gradient than a convolution of it and a radial vignette gradient. And even a bad flat—such as a dark in the wrong orientation from my C5/750—is more helpful than no flat. And I've successfully used the same flat for each configuration for over a year now.
2. Darks. I've joined the dither or die bandwagon, and indeed, I could get away with just a handful of bias frames and no darks if I dither, let the mount drift, and use sigma-reject stacking. This eliminates hot pixel paths and averages out the residual fixed pattern noise quite nicely. But in going back through some old data from back when I didn't know to dither, I paradoxically found that cherry picking the lights and darks to where their average temperatures matched to within ±0.1˚C produced better results than just adding more data—especially in the red channel. Because of the mount drift, the full stack was walking noise limited from the residual dark noise error getting walked across the image. But with the reduced-but-matched stack, I had less residual fixed pattern noise and a shorter distance any noise got walked. And I found that at f/2.8, my images were rarely shot noise limited—so less subs was actually quite viable.
3. Bias. I have a library for ISO800 and above, but I only use them for short exposure subs (such as for the moon, assuming I'm imaging at high ISOs). For longer exposures, I use my darks library.
4. Flat darks. I find them to be unnecessary, and I've purged the library. And RawTherapee has no way to use them. But I've been told that it's wrong, so not to tell anyone. So I'm not tell you here.
There are more opinions than DSOs in the sky. So find what works for your workflow for the output you're aiming for.
Edited by BQ Octantis, Yesterday, 12:18 AM.