Is there a tool for the following match based on pgn?

[Uri Blass]

Take a pgn of many games of humans.
call it pgn A
Take every position inside pgn A and let engine C to play against itself from the relevant position at level of average of 1 second per move(or different time control that you choose) in order to save the games in a new pgn that you call pgn B.

After you finish one game you look for the first position in pgn A that is not in pgn B and make a new game from it and continue in this way until you have pgn B that include all the positions of pgn A with evaluations at 1 second per move.

Now you basically have evaluation for every position in pgn B at 1 second per move and result at 1 second per move.

you use the data to build a table to translate evaluation to expected result.

You can decide that if the number of games in pgn B with a specific evaluation is smaller than 10000 you do not trust the results to be accurate enough so you may have a table like this
evalation (below -300) games 10020 expected result 0.0004(wins 0 draws 8 losses 10012)
evaluation -300<=x<-290 games 10300 expected result 0.0007(wins 1 draws 13 losses 10286)
evaluation -290<=x<-283 games 10430 expected result 0.0008
...