Today when I used the recommender, I tried putting the age slider at maximum (10 years).
In this case the first game recommended was "Maiden and Spell", released February 25th 2020 (4 months 5 days ago).

After that, I slided it all the way to the opposite (6 months). Now, the first recommended game was "One step from Eden" (released March 26th 2020), and "Maiden and Spell" was now ranked 16th.

This seems illogical: I would think that with a bigger selection of games across the 10 years time range, the fact that "Maiden and Spell" was first would mean that it won against all the other games. Then why, in a smaller time range of 6 months, did it lose against 15 other games that it won against in the broader range?

I suspect that this age slider is part of the machine learning module of the recommender. I guess so because sliding to the limit between different displayed years (let's say for example the limit between "10 years" and "5 years" at 0.09-0.1) will sometimes bring about errors, like CS:GO still appearing at 5 years despite it being 8 years old. Which proves that the borderline is "blurry", because it is outputted by the AI.

In my opinion, this is confusing, and the age slider should be a hard/accurate delimiter for the age of games (and so probably decorrelated from the machine learning). If you wanted to make it a weighted slider that allows some inaccuracy, then just stop writing 6 months/1 year/etc. under the slider. At the very least, it should be "~1 year", with a tilde character in front of the age.