What we haven't proved is that you can have social media run by a publicly traded, ad-revenue-funded company without all sorts of harmful effects (with the main interface being smartphones with push notifications). That's where all the nasty "engagement" effects come from - trying to drive eyeballs to ads to improve revenue. It's very much a zero-sum game - every pair of eyeballs has 24 hours in the day, so the goal is to command their attention for as many of those hours as possible. That's where the evil creeps in.
This is just a theory but my fear is that social media that's optimized for "healthy usage" probably looked more like the forum and blog culture that social media killed.
Optimizing for engagement means you basically have a genetic algorithm on your hands for surfacing the content with the most "viral-potential." Eventually that stuff eats the healthy parts of the internet because people inevitably talk about the viral stuff that's happening, which means your healthy-use forum is nonetheless revolving around the conversation in the viral centers.
From there it's a matter of time before people start going directly to the viral source to keep up with the context and conversation. And once they're there, because it optimizes for engagement, it crowds out their use of everything else.
So there's a natural selective pressure here. Optimizing for engagement/addiction gets you a network effect that leads to overshadowing any other type of socializing. Unless there's some mechanism to actively select against virality and engagement they will naturally rise to the top even independent of ad-impression incentives.
Basically anything without crowdsourced voting that affects placement of the message. That could be obvious things like reddit posts, and other things like online reviews. Effortless "likes" and "upvotes" produce the worst feeds.
https://road.cc/content/tech-news/271757-strava-has-brought-...
I only see my friends' stuff on Instagram and it's lovely. Though they've recently changed to stick random stuff underneath the last post from a friend which does diminish the point somewhat.
In addition, how does one define "engagement" and "healthy usage"? Way back when, people spent much of their free time on a landline. Does one go on to blame AT&T for "optimizing for engagement" by providing a service that could keep people connected to any person in the country for hours upon hours while the phone company made millions minute-by-minute? Ultimately this depends on the person and his/her wants/needs (however "unhealthy" you think they might be) and those can only be determined by oneself regardless of how much data someone else collects.
Your AB experiment would only work if you had sentient gingerbread men as a control group. Even then, I doubt it would say or change much about what people do on a person-to-person basis.