Perhaps. It’s possible fodder for a request.
In all honesty though, I think the best way to improve the algorithm is to simply look at the grading data and see how self-consistent it is. Some books & tv shows are bound to have more variance for all the reasons @DuoUnderscore and others have stated and it’d be nice to surface that information. I can’t find the request, but I thought there was a feature request to surface historical rating level changes and compute how large the variant / how consistent something is.
I also agree with a lot of @DuoUnderscore says around Video especially. I think since a lot of books have a narrator that speaks in one style, that can make books feel a bit more consistent? Not true for manga of course. I also agree with @DuoUnderscore that people are watching stuff that’s above their level in Video more consistently, especially with tools like language reactor built into Netflix.
I really do like @bibliothecary 's Detailed language evaluations suggestion too. I think that’d illuminate the ‘why’ questions you’re asking on how difficulty is being assesed.
Ultimately though, it comes down:
- is it helpful for tracking your progress?
- is it helpful for discovering new content?
I think you’re right that it’s not 10/10 for each of those, especially for video. And there is a lot of work that can be done to analyze possible systematic biases. And I like to hear opinions from people who don’t find it useful and prefer not to use them. I must admit though, that while I do get frustrated it’s not totally accurate, even if flawed, I find it very nice.
Granted, I did start it, so hopefully I think that 
