[Idea] Calibration graphs

[omarkohl]

e.g. display how often outcomes that were predicted as 80% actually resolved as correct. With perfect calibration it would be 80%.

Ideally it should be possible to dynamically choose which probability intervals to use and limit to specific categories.