It would make sense if all the required data was supplied by your broker, but as your modelling quality of 40% indicates, the tester is having to guess what happens within a bar, so the result isn't going to be a true reflection of live trading.
Besides, if your broker's datafeed differs...