I noticed that too. I don’t really see this as a bug. It just tells me that Shotcut is doing stuff (maybe gathering infos about the project or something?) before the actual export job can start.
But maybe it would make things clearer if it said something like preparing… instead of paused, and if an animated spinning arrow icon was used instead of the pause icon.
Interesting that you went to that from “paused”. For me the first reaction was “oh I must have clicked big pause button by mistake” (or the pause this job one).
Between another word and the dashes I’d defintely choose the dashes, there’s no reason to add complexity in this small area. The “Preparing” keyword would have to actually represent the stuff you mention so it would be “preparing/starting/etc” for a while and then start from 0% (–;–;–) to show that the exporting has now started so there would be a fix for this but also an enhancement to the code to clearly separate those extra steps. This is just too much detail to be worth the hassle.
The “paused” at the beginning of the job was a regression that was already fixed. The dashed time value means there is not yet enough data to compute an estimate. On previous versions, the Windows on Arm build was flickering between dashes and time values while running beyond 2% that was fixed but caused the regression.
I gave a visual example that the filter is buggy and needs to be fixed. In your example, there are also color artifacts in the sky. I.e. the result is unsatisfactory.
I have attached an example with another editor, and I have even amplified the blue saturation to the maximum. Artifacts and banding are not visible. I haven’t touched any of the other colors.
Pour assensible
Essayez avec le curseur Overlap
For assensible
Try with the Overlap slider
Just because the filter does not work the same as some other tool that you like does not mean it is a bug.
When I developed the filter, I compared it to the GIMP Hue-Saturation filter and I saw very similar results:
https://docs.gimp.org/2.10/en/gimp-tool-hue-saturation.html
Please try the “Overlap” parameter as @meiber has suggested (twice). This banding is exactly what “overlap” is for.
This morning I downloaded the latest Github version (25.01.13) and that problem is gone now. So whatever you did, thanks 
