The awkward part
LOUDNESS IS ONLY
KNOWN AFTERWARDS
A peak can be measured the instant it happens. Loudness cannot: it is an average of what has already been heard, so a figure for it only exists once the audio has played. That is the whole reason a level set at the start of a stream is a guess about the rest of the evening, and why a good guess still drifts.
A meter on its own does not solve it either. Watching a number and nudging a fader is exactly the right response, and nobody wants to do that for four hours while also playing, reading chat and talking.
The scale this is measured on is LUFS, and it is worth exactly one paragraph of your attention: it describes how loud something sounds over a stretch of time, rather than how tall its peaks are. Which number to aim at is its own question, answered on what LUFS to stream at.
Fix it before it leaves the PC. Whatever a platform does on its own side happens to the finished mix, after the two sources have already become one signal. Turning a whole stream down does not make a quiet microphone louder against a loud game, so the balance has to be right on the way out.