I bumped into this thread in SoulCalibur forum "There are two methods that i can vouge for for testing frame data: Frame Counting Method - Requires a Video Capture Card, a PC, and a video editing program. Frame Collision Method - Requires some existing frame data to test with Both methods are accurate and reliable. i reccomend the frame counting method for finding base speeds of moves and such, and the frame collision method for finding hit/grd/ch properties, as well as in some cases for verifying move speed. Allow me to explain both methods in detail: Frame Counting Method The things you will need for this method are: a video capture card or similar device that allows you to capture video from your console(ps3 or 360)(usually comes with some kind of video capture software) a pc, almost any will do, it doesnt have to be extraordinary some video editing software *make sure you have the input logs turned on in training mode(the log at the bottom of your screen that shows all your previous inputs) as it may make things easier for you in determining moment of execution/input. Now for the Semantics of this method: 1) set up your capture card, and capture settings: this will typically vary, depending on what kind of capture card you have. simply follow the instructions, and then hook the capture card up to your TV/console, using cable splitters as neccessary. Start your game, and open your video capture software, make sure if you can, that you choose to record at 60fps. if you cannot, 30fps will do, but your data will not be as accurate, and will require verification later on. 2) start recording now for this one, i reccomend you give it at minute to buffer after you start recording to make sure that you dont get lag or dropped frames. after this, you'll want to test the speed of each move. just hitting the opponent will be fine. try to test each move at least 5 times on video, so that just incase one time drops a frame or such, you have 4 other test results that you can validate with. just simply go through all of your characters moves, testing in a manner such as this: A,, A, A, A, A, B, B, B, B, B... until you have a recorded result of all the moves your character has *note: if a move has multiple hits in a string, you need only to test the first hit of the string for speed. 3) breaking down the data into frames now this is a bit tedious, but by no means difficult, just time consuming. what you do now, is open your recorded video into your video editing software, and simply advance the video frame by frame, and count the number of frames from move initialization to move impact. that will be the speed of the move. in this situation, recording at 60fps is very very useful since it will give you exact results, but if you recorded at 30fps, you will have to take your results and multiply them by two(which will give you all even numbers unfortunately) however once this is all done, and you have recorded all of your results (check them over to make sure you havent missed anything) you will now have the impact/speed values for all the moves*. *note: if you recorded at 30fps, you will also need to verify impact frames with the collision method. Frame Collision Method The things you will need for this method are: some existing frame data to cross reference with(preferably multiple characters, or characters that have little or no RV's) Ok, so for this method... finding the speed of a move: the best way to do this is to use the record command function in practice mode. (note that for some moves you will have to tap G before executing the recording to allow it to buffer through hit/blockstun) to record the move you want to test. then you set up the practice dummy in freestyle and have the dummy set up in the following manner: action 1 - standing guard all/standing normal action 2 - command start #1, 2 or 3 next, you trigger the dummy to input your command by hitting it with a move, preferably a move that is neutral on grd or neutral on hit(both methods are good for different situations) and then input a move after to see if the two moves collide. a useful trick for finding frames that are inbetween that of your test character(for example i18, but say your test character only has an i17 and an i19) is to a move that is slightly +on hit or grd to compensate for the speed, then test until you collide and subtract the +frames of your trigger move from the end collision speed of your opponents move(or if your using disadvantage to test, simply add the frames) For testing stances, you'll want to use a neutral move to initiate. have the recorded command enter stance, and then attack. in order to find the speed of the transition, you'll have to keep hitting the opponent until you get CH, to find what the slowest move you can hit them with is before getting counter hit, and that is the speed of the transition. finding the grd frames of a move: first what you'll want to do 44 guard your dummy's move. then try to punish with one of the known impact frame values that you know(note that you may need to use more than one character for such)simply start with K(i13)(or in some cases i10 or i11 AA) in an attempt to punish. if K punishes, try to punish with a slower move, such as an i14, i15 and so on until you cannot punish anymore. if K doesnt punish, have your opponent use a move after(in most cases K) which will come out at disadvantage(so if the move was not K punishable, its at most -12 on block, which means the opponents K will come out at i25 and thusly colliding with i25 moves, telling you the move is -12 ie. i25 total - i13K = -12 disadvantage.) the same method can also be used to see how much advantage a move gives, although in these cases you'll want your opponent to use much slower moves to go against your i13K. this also comes in handy for finding GI, TC, TS, and TJ frames. finding the frames of a move on hit and CH: same as above. when testing frame data, generally, you have achieved a result when you get the following: collision, clash, or break. Range values: some moves have multiple impact, hit and grd values depending on what range/position they are executed from. generally a good rule of thumb is: the further away from the origin, the slower the impact frames are, but the better the hit/grd frames are, and vice versa. note however that there are exceptions to this rule (such as Siegfried's b6 in SoulCaliburIV) to get accurate results for range values, you'll want to test moves that give inconsistent values at short, mid, and long range from the point of origin, when applicable. that way you will be able to get more accurate move properties. *NOTE: when testing frame data, it is almost always best to test near a wall to eliminate any possible range values. plus it makes punishing easier. try to test using kick attacks when possible. also note that there may be some moves in the game that generally break all of the rules of testing, and may require special testing methods. this however goes on a case by case basis, and is extremely uncommon. Hopefully you found this tutorial useful! Please feel free to let me know if i missed anything." Tl;Dr With this guide in mind, is it possible to break down all scrolls to their actual frame data? If so, we can have an extremely accurate tier list for scrolls. Source: http://8wayrun.com/threads/tutorial-how-to-test-frame-data.12706/
I remember seeing this video about competitive smash bros melee players and they did the exact same thing. One of them even went as far as to skip eating to play smash, and would dream about frames and reaction times and controllers.
