There's a lot of proposals to uncover in the original post. Some I agree with, some I disagree with, some I partially agree with and i'll try to explain in detail how I would tweak the design a little. I think first it's good to understand there's roughly two groups of players that use quirks. SP/Co-op players and test players. I think combining everything into one and slapping just 'test' or 'vanilla' onto them is going to add confusion.

I'm more in favour of introducing two settings, each for their original purpose: /vanilla, or /oldphysics or /standardjj2. (name can be maybe polled later) for what I assume to be the same as speedrunmode to reintroduce all the original behaviours and from vanilla jj2. I think /test is mainly to introduce a neutral team that can pass through others, maybe with a better win condition at the end than flags and maybe some kind of level/timer from the test manager. I'm in favour of seeing /test similar to how /plusonly and /latestversiononly interact, whereas /test gives a "you can only run /test if you enable /vanillamode first'" checkbox. This gives gives /test the oppurtunity to expand to test manager settings and guides, and makes it very clear that /vanillamode is just that, for original settings. I don't assume every player knows what all quirks are, but the intention is clear.

I'm in favour of seeing /test similar to how /plusonly and /latestversiononly behave, whereas /test gives a "you can only run /test if you enable /vanillamode first'" check. This gives gives /test the oppurtunity to expand to test manager settings and guides, but also keeps /vanillamode to just a clear intention for original settings. I don't assume every player knows what all quirks are, but the intention is clear. I would like to see /test as a server setting only, and /vanillamode for both SP and as a server setting, similar to /allowmouseaim, but instead defaults to true. (so you can always change your mode clientside and in SP and the server can disallow it to always be on or off, whatever the server decides).