Gradient Removal in Multi-Session Stacking

Dear all,
I was wondering what the best approach is to gradient removal with data from multiple nights.

It seems to me, that as my data grows over multiple nights, the performance of the gradient removal feature deteriorates, resulting in local “hazes”, like so:

This haze in the middle is nearly not as bad as when I stack the data from each night separately.

Now, if I suppose the gradient is linear per night, then the polynomial degree necessary to eliminate this gradient grows per night, meaning if each individual session has a linear gradient, the gradient after, let’s say 6 nights, should be an order 6 polynomial?

Is there any consensus on how to approach this? Would it make sense to add an option in e.g. sirilic to first remove the gradient per session before merging everything and stacking the multi-night data?

It is unlikely that the gradient is linear on the stacking result of a night. It is likely that the gradient is linear per exposure. I see two possibilities: remove a linear gradient from each exposure or remove a high-degree polynomial gradient from the stacking of the session.
If sirilic doesn’t propose the second solution, I think it proposes the first.
But if you have a manageable gradient on the session stack, it would be faster to use the second solution.

You can remove the gradients per sub and still use Sirilic.

Instead of running the stack from start to final finish (with the Run Scripts command, under the Actions menu, select Run/fine tuning (Experimental) and then run all processes up to and including Group registration.

Once this is complete, you can then open Siril and the registered group sequence and then perform gradient removal per sub before the final stack stage. You also have more control with the various stacking algorithms within Siril.

I also get into the habit of “blinking” my subs through Siril (with a histogram stretched sequence) before I start loading everything into Sirilic. Then you can easily cull any bad subs.