That's not what I'm saying. What I'm saying is it's easier to execute in an actual match, or against an actual person controlling the other character, that it's not as tight of a link as training mode makes it look. I'm saying it's easier to perform than training mode makes it appear. At least that's what seemed to be the case last time I was practicing my timing.
EDIT: Obviously I could be completely wrong, and since I can't test this until later tonight it's sort of just a theory atm, but I seemed to be 2 to 3 times as consistent performing this against a human opponent then against the CPU set to low block. The way I was practicing this in training mode: I set the opponent to duck, did b1u3 xx mb freeze, j2 to set up the reset, and the timing was very strict, clearly 1 frame. I hit it just a little over half the time I tried. But whenever I recorded the CPU doing this reset, I had like 90% consistency. The only difference in these two ways of testing is that in the first the computer was set to duck, and in the second I'm trying to duck. But for some reason my consistency doubled. Maybe it was just chance, but it really seemed like the window to execute this was MUCH bigger than 1-frame when tested with the record function.
