It's not like human can set same level of difficulty problems in every contest. You found some 1500 rated problem from div1 + div2 feels harder then a 1500 rated question from div2. I also found vice verse.

In my experience, the length of the contest affects the calibration of the problem rating (at least in terms of implementation difficulty). I think it's because in a longer contest more people will have time to solve it so the user rating of the average solver will be lower

i think the rating of a problem is somehow related to the number of people who solve that problem in the contest.

I believe that a rating of x on a problem means that half of the people around that rating will be able to solve that problem. Since Div 1 + Div 2 usually have higher rated participants in general, a 1500 rated problem might be harder in general than a Div 2 1500 rated problem.

True. And it's much harder to get the same performance at Div.1 than at Div.2.