It usually refers to a model that is tested by applying it to the past. It is run with past data to see if it could correctly predict past events.
I think there is value in this concept of retrodiction. It is important to look at the past and see if the results make sense based on what we know. It typically is easy but that is handicapped by a cognitive bias where we think past events and conclusions seem obvious in retrospect. Take these last years in the NBA playoffs. The Lakers were the best of the West. Their series with the Rockets was unusual but was never very close. Just like the Celtics seven game series against the Hawks in 2008, the result was never really in question.
In the East though, it gets curious. The Cavs were the #1 seed but didn't match up physically with the Magics athletic shooters and center. The Cavs weren't made to beat the Celtics; they were made to beat the Pistons! Hell yeah! Awesome insight!! They were based on plodding bigmen, a slow wingman (Rasheed Wallace), etc. And the Magic came out with a super athletic Center and a wingman who could shoot and drive. And Varejao and Ilgauskas were abused!