Dereverb vs Karaoke: which first?

[verbal hinge]

After extracting vocals from a song, should you use dereverb or karaoke models first? I've heard from some that, for lead vocals at least, dereverb should be done before karaoke (separating leads from bgvs). Is this the correct order to get the cleanest vocals (to use for AI)? Because I've also heard some people do it the other way too (karaoke before dereverb). And what if I'm aiming to extract background vocals instead of the lead? Dereverbing takes out so much bgvs sometimes so it'd make sense to karaoke before dereverb for bgv's at least right? 🧐