getting an estimate for the percentage of games with no losing mistake

[Uri Blass]

1)Take some big pgn of games that have final result of a draw and let a strong engine to play against itself at some fixed time control from every position inside the games.

If all the games end in a draw then the game probably has no losing mistake.
If at least one game end in a win then probably the game has a losing mistake.

Now the question is what is the percentage of draws in human-human games with probably no losing mistake based on this test and if this percentage is getting bigger or lower if you increase the time control of testing(for example from 1 second per move to 10 seconds per move).

I expect the percentage of probably games with no mistake to become bigger with long time control because at fast time control the engine may win some drawn positions against itself because of mistakes of the engine(with weak engine even the opening position may not be a draw based on engine-engine games so weak engines are useless for this test).