Oeh, hands up for @erlend.sh following me 
Your four criteria (non-AI content, DNS, public track record, social graph) are pretty much where I think Iām going with this as well: observable evidence, human conclusion. No āgui.do has trust score 73, proceed with caution.ā
But thereās an irony I keep running into: you canāt really score bots, because the moment they get a negative signal they just create a new account. Plus they are getting really good at mimicking human behavior at the single-post level. So the only way (at least as far as I can tell, please correct me) to filter bots out is to give positive signals to humans about their profile (not single posts)ā¦ which means you end up scoring humans anyway, just to weed out the bots.
At least until the bots figure that one out too? But at least with profile-level scoring that also takes profile-age into account, gaming it becomes much harder (and hopefully economically unfeasible).
And thatās actually also what your @erlend.sh point illustrates: you assign trust to me because someone you trust already follows me. I got lucky, but most newcomers wonāt have that, and right now on every new platform youād start from scratch. Thatās fine if you want to, but shouldnāt be necessary. If the trust signal runs in the background (not as a public score, but as infrastructure that helps mods and the system decide youāre probably a real person), your DID could carry that across ATproto apps.
Your Glowrm post gets at a related concern where a cross-app reputation that follows you everywhere is dangerous. But imho thereās a difference between āthis account is part of a coordinated spam clusterā (useful signal for everyone) and āthis user got banned from a dating appā (nobody elseās business).
On Barazo: no trust problems yet, itās still in alpha. But Iāve been a community manager and forum mod/admin for 25+ years and based on that Iād rather build defenses before the spam wave hits than after
To be clear: nothing wrong with humans checking out who someone is and forming their own judgment, thatās what you did with my profile and it worked. But when your feed is 80% bot slop, you need something automated to help you trust that what youāre reading is from real people. I donāt have a good answer for how to do that without some form of hiding or deprioritizing content (again: let me know if you do 
). The question is how to do it without false positives hurting real newcomers.