What if you add the ability to forecasters to report one characteristic (interests/hobbies) from a group of characteristics? (for example, 8, this is important, but it is not yet clear how =)), then break all the questions on these characteristics, it is obvious that interests affect the amount of knowledge in a certain area. The accuracy of the forecasts I think is directly correlated with the amount of information on the issue. Well, respectively, the value / weight of some forecasts (with the desired attribute) of some forecasters (with the desired attribute) may increase, which in turn will lead to better performance of the model.
If you found me for a madman --->
https://cindicator.com/
P.S. The video is boring to for my taste.