>>>>>>I'd imagine if all of your images have more noise than useful data, than there's no point in stacking.
That sounds sensible. And it would be true if noise was a thing in itself. However, your analysis kinda misses what "noise" is. Noise is part of the signal, just as the recorded energy of the collected light photons are signal. If you could record precisely the energy of the photons hitting the sensor, you would have no noise. But you cannot. Too many things get in the way. Heat photons, cosmic rays, variations in the way sensors react, "rounding" (quantization) errors in converting photons to ADU electron units, and such...... So noise is the uncertainty in your signal.
Your analysis sounds like noise is something above and beyond signal, as if it were added to the signal. It is not. It can be. But it can also be stuff taken away from the signal. It is uncertainty in the signal.
In fact, if you stack enough noisy data the noise starts to disappear. You see, noise is not "noise" in the sense we usually think of it. For astroimaging purposes it is best to think of it in terms of "uncertainty." And the purpose of stacking is not to "add" the signal together, but to "average" out the signal.
Lets say you get the following reading from a given pixel.
9, 10, 9, 11, 10, 11, 9, 10, 11, 10, 10, 11, 9, etc......
What is the "true" value that the pixel should be reading? I'd eyeball it at 10. But relatively few subframes are actually reading at 10. The noise puts the signal near 10, but rarely right at it.
What stacking does is allow one to average out the readings to get you to that "10."
But it is more powerful than that.
Let us say that your readings were instead:
9, 10, 1, 17, 10, 11, 9, 10, 11, 10, 10, 11, 27, etc...... (the same as before except this time, you had three weirdos for some reason).
C'mon, that 1, 17, and 27 are goofy. Before averaging you should throw them out. And average out the rest. That would get you back to your "10." Getting rid of the 1, 17, and 27 reading is called "rejection" and your software has a "rejection algorithm" which helps decide which readings are spurious and should be ignored.
So, remember, when properly handled, stacking noisy pictures together is not ADDING the noise, but instead AVERAGING OUT the noise.