One thing I tripped over initially was fuji’s DR modes. In DR400, the raw file is underexposed one stop, and in DR800 it’s two stops. My Lua script detects this using exiftool, and adjusts the tone equalizer accordingly. This usually gets me pretty close to the JPEG’s exposure. One could probably do a similar thing to implement in-camera white balance shifts.
Film simulation colors however are a different matter. Actually, three different matters:
The first is white balance, which is implemented differently by Fuji than in darktable. That’s what you discovered with your Color Correction change. The problem is that this change varies with color temperature, so it can’t be consistently replicated by a LUT alone.
Second is the tone curve. My LUTs are intended to be applied after filmic. This means, however, that they rely on filmic’s default tone curve. In 4.2.0, filmic’s defaults changed, so my LUTs are probably no longer correct. Stuart Sowerby’s LUTs have a strong tone curve built-in, and are intended to replace filmic. But of course they don’t implement the scene-to-display mapping, so you should probably put them after a zero-contrast filmic.
The third issue is colors. It seems to me that my LUTs try to partly “undo” filmic’s color preservation, which leads to issues with saturated highlights. I should try to redo the LUTs for 50% sigmoid, which should fit better. I suspect Stuart Sowerby’s LUTs to have similar mismatch with filmic’s color preservation modes. So maybe they’ll work better after a zero-contrast sigmoid, too.
Personally, I’ve found LUTs ultimately didn’t quite work out for me. I’m currently experimenting with a separate rgb tone curve after sigmoid, and then a lookup table instance to correct colors. Splitting the film simulations thus into a tonal part and a color part seems to work more robustly.
Other raw developers seem to solve these issues by providing camera-specific tone curves and LUTs (ICC profiles) to get them all to a common baseline, and then implement film simulations as a second-level LUT from there. Since darktable lacks this common baseline, a LUT-based approach is bound to be difficult. Especially since darktable’s default rendering changed subtly version-to-version.
However, I must say that most of my color rendering issues have been resolved by switching from filmic to sigmoid. This is highly personal, and perhaps mathematically misguided, but I just prefer the look of per-channel, hue-twisted sigmoid highlights.