The JG stutter is people just noticing that the rule on JG means that tapping G into pressing it again exactly 6 frames later maximizes the use of the JG window while giving you guard if you miss the JG by too early.
The JG flash is long enough you can still release your G-button in time to drop guard for whatever punish, right, but if you didn't JG you have guard up "seamlessly" after the JG window.
Diagrammatically*, with C being before the G input,
-C -| -------------- --- A --- ---------------------- | D -| -- B ------
---- | --------- --------| ------- +6 frames+ - ----| - - |------------

input , release

input , ................ (x) ,

input
You're covered for guarding the whole time; in region -A- because of JG window, in region -B- because you're in the G state. You're reducing the size of the -D- span of frames to zero.
You still disable the JG option for the 29 frames or whatever. If you cut into the A region, you're just reducing the window in which a JG will activate and obtain only a G.
EDIT: And 6 frames, at 60 frames/s, is 100 milliseconds, or "10 centiseconds", the quantity you'd actually see on a stopwatch that shows you two digits after the seconds.
*
this diagram is incomplete; I've assumed you press G for exactly 1 frame. Rather, there is a span, call it -F-, which precedes -A-, in which you may hold G for up to 4 frames (the JG "tap" rule). The span -A- is the 6 frames of JG state.
---
k, on #3, it uses some concepts from my background in computation, and dynamical systems.
If you strategize to one 'objective', then you'll maximize that objective. But some actually good "goals" or "states" are impossible to state as a single objective - rather, only as the mediation of two (or more) inherently opposed extremes.
When speaking, for instance, you want to say what's true (putting aside malicious language). But also, you want to say what is economical - understood quickly, and used in relevant context However, each thing you leave out of a statement is a factor that, technically, reduces its absolute truthfulness. And yet, to be completely true, you have to say infinitely many things.
Neither of those two 'goals' (truth, economy) is the uttermost valued 'good' in the scenario. The good is "balancing both" even though they are logically opposed to each other. Somehow, by using other measures, people intuit how much to weight either one.
Theorists came up with dynamical systems theory to get more precise ways of talking about how processes can
automate these balancings and middle-ground-findings - and to explain processes in nature that use it (for instance, our own use of language as just mentioned; sorry I don't know of any more purely physical example offhand).
I described in my post my emphasis on defense as how I bedrock my entire mindset... but noticed that it plays off of my belief that I have an advantage to interrupt somewhere. The two are opposites, but I look for when to listen to one or the other. My adherence to either one prevents me from hitting the extreme of the other - where I would be turtling, or I would think too highly of my chances to be an aggressor.
edit:
I don't want to get hit, but sometimes, I think my best chance to live, is dash up and throw you. Or BB confirm B, or 2A or whatever. I don't know yet how to put it, the difference between this, and going "I think my thing will be faster NYAAAHHG! *tries poke*" But I feel one.
I just have a bunch of abstract ideas about what can work. I still suck. Not technically proficient enough, and nowhere near fast-thinking enough to beat someone who concentrates on doing good sequences.