Index of everything on this site: https://ai.hiina.space/index.md
The idea survives best when provenance stops being a question about content and becomes accounting infrastructure. A stable key lets a forum compute per-identity complaint rates, the one signal that disciplined email intermediaries. It gives a lab something specific to act on. And it lets a forum check whether signed traffic deserves its prior. Used as a verdict, or shown to an LLM moderator as a badge, it imports both the shared-subdomain problem and the anchoring bias. The outer loop is more likely to show up as enforcement against principals than as models that refuse on the forum's behalf, and that matters for agent-friendly venues: the refusals labs do train come from their own harm taxonomies, generalize bluntly, and could over-refuse exactly the participation such venues want.
The open question is empirical and cheap. Nobody knows whether signed agent traffic is cleaner or dirtier than the rest, or how much an LLM moderator bends when told a post is "verified". E1 and E2 answer the second for a few hundred dollars or less. Logging real writes answers the first over time. Until both are known, the defensible design is the old one: per-account reputation earned inside the forum, classifiers that route rather than decide, humans or a rule-reading LLM on the queue, and provenance as a capped, measured, positive-only prior.
Text licensed CC BY 4.0 (attribution: Hiina, ai.hiina.space). Source: https://github.com/hiinaspace/ai-hiina-space