Maybe I wasn't clear enough. What I asked was how you would calculate the "end" of the saccade. If you say fixation begins from the end of a saccade, you are not really explaining anything. I don't understand why you say you would use EyeLink data to get a firm assessment, since ML has exactly the same data, but perhaps you can just explain how you would use the EyeLink data to determine the beginning and end of a saccade.
In ML2, you can calculate the time of any event to the precision of any offline analysis you can do. So there is no such thing like "the best read out". You just need to say what is the number you want to calculate.
If the number that you want to calculate requires an offline analysis anyway and there is no need to calculate it online, you may not want your computer to waste time in the middle of the experiment. For example, one of my experiments uses a visual search task in which the subject is required to find a match among 8 stimuli presented simultaneously. I need to know how long the subject watched each stimulus so I calculate the time of fixation acquisition and break with a velocity criterion: if the eye velocity is below 30 degrees per second, it is considered a fixation. For the velocity calculation, the eye signals are typically smoothed with a Gaussian window (otherwise, you will overestimate the instant velocity due to the sampling noise). If I use a window of sigma=3 ms, then the entire length of the window will be 18 ms (3 ms *6). This means that there will be a constant 9-ms delay (a half of the window length) in velocity calculation, if I do it online. However, I don't really need to know the fixation duration while running experiments. I just need the numbers for analysis afterwards. Therefore, I don't do an online calculation for that particular task.
So you need to think about what is the number you want to know and whether you need to know during the experiment or it is okay to calculate it offline.