[Shark69]

I've been checking what you're saying and maybe the approach would be to use data that's on the internet rather than doing the work all over again.
For example, with reference to Wedlex, I have been studying the Spanish language and I think it is quite good.
A problem could be that there are a lot of words. Many of them are derivations of other.
As you say, the issue is to take the thresholds, but my proposal is that they be dynamic thresholds.
I explain. In the Spanish language there are 208,078 entries in Wedlex. By default, equal thresholds could be set for the 5 levels of difficulty. 208,078/5 = 41615 entries per level. If I look at the entries since 41615, it may seem to me that the thresholds are not correct. Well, it could be that each person decided the threshold based on percentages.
For example:
Level 1: 0-10%
Level 2: 20-50%
Level 3: 50-60%
Level 4: 60-70%
Level 5: 70-100%
But, the user could change the thresholds and the level should change.
I don't know what you think.