1/sample rate [in Hz]
* samples [number of samples taken under the hood and hidden from user]
+ ~0,1 [processing time]
= time in seconds it takes for one line of log to appear at “no wait”.

1/125 * 256 + 0,1 = 2,148 s (slowest possible for reference)
1/125 * 32 + 0,1 = 0,356 s
1/4000 * 32 + 0,1 = 0,108 s (fastest with stable results)
1/8000 * 32 + 0,1 = 0,104 s (might generate random errors in data)

0,1 s is just a bit longer than it actually takes for processing and saving results, but you get to the ballpark with that nice round number. 8 kHz is possible, but it has been a bit unstable and in most cases you get more usable and less noisy results at 4 kHz. And as my calculations above state, you cant get that much faster with 8 kHz anyway, that you should tolerate it adding random spikes in your results.

But then, there is this thing called buffer mode “hidden” in settings for graphs. There you can see actual live samples taken as a graph. 50/60 Hz mains sine wave is easily recognized and that is probably what makes some people think that mooshimeter could be used as an oscilloscope. There is no way of aiming or triggering, you just need to wish for the best and let mooshimeter shoot in the dark. 8 kHz sample rate would give you 160 (for 50 Hz, 133 for 60 Hz) points per full mains wave, so it still looks quite smooth and you might see when the waveform is not what it should be etc. Now guess what you see when frequency starts to rise. 16 points from 500 Hz wave is close to useless, it might be possible to tell what frequency that graph is supposed to represent. Anything higher and you most likely need something other than a multimeter. I think i know a device that would help and it is called an oscilloscope.

And as always, what was the question? Some day I should try to answer to someone and not just ramble something :D