Hi all 
I’m Guido, longtime BlueSky user, build some smaller apps for BlueSky/ATproto but now it’s time for a real challege… 
I’ve built communities and implemented community platforms for 20+ years. Thinking about community platforms in light of recent trends in data ownership, sovereign software and combined with the ATProtocol… I see an opportunity to build something cool.
So I’m building Barazo, a forum AppView on AT Protocol. Think Discourse/Flarum alternative, but with portable identity and user-owned data at it’s core. The ease of use of signing in once (like Discord or Slack), but decentralized and under your own control.
The forum should of course be able to stand completely on its own. But if/when it gets a decent amount of traction, you could really start to benefit from the network effect that ATproto offers. For example: you can use cross-forum reputation, benefit from global labels etc. etc. In theory, you could even start a new forum with content that is already existing on the ATProtocol.
I’m still early in development and this is my first time designing lexicons, so I wanted to share what I have before I paint myself into a corner. Reading the Skylights/Bookhive situation was enough to convince me that coordinating early beats discovering incompatibilities later.
Current lexicons
5 records under forum.barazo.*. Full schemas in barazo-lexicons.
forum.barazo.topic.post
Forum threads. Title, markdown body, community DID, category, up to 5 tags, optional self-labels.
{
"required": ["title", "content", "community", "category", "createdAt"],
"properties": {
"title": { "type": "string", "maxGraphemes": 200 },
"content": { "type": "string", "maxLength": 100000, "description": "Markdown body" },
"contentFormat": { "type": "string", "enum": ["markdown"] },
"community": { "type": "string", "format": "did" },
"category": { "type": "string", "maxGraphemes": 64 },
"tags": { "type": "array", "maxLength": 5, "items": { "maxGraphemes": 30 } },
"labels": { "refs": ["com.atproto.label.defs#selfLabels"] },
"createdAt": { "type": "string", "format": "datetime" }
}
}
forum.barazo.topic.reply
Threaded replies. Root and parent are both strong refs – for top-level replies, parent == root.
{
“required”: [“content”, “root”, “parent”, “community”, “createdAt”],
“properties”: {
“content”: { “type”: “string”, “maxLength”: 50000 },
“root”: { “ref”: “com.atproto.repo.strongRef”, “description”: “Original topic” },
“parent”: { “ref”: “com.atproto.repo.strongRef”, “description”: “Direct parent (topic or reply)” },
“community”: { “type”: “string”, “format”: “did” },
“labels”: { “refs”: [“com.atproto.label.defs#selfLabels”] },
“createdAt”: { “type”: “string”, “format”: “datetime” }
}
}
forum.barazo.interaction.reaction
This is the one I’m most unsure about. Reactions are not hardcoded: the type field is a free string, and community admins configure which types are available. One community might only have “like”; another might have “agree”, “disagree”, and “insightful”.
{
“required”: [“subject”, “type”, “community”, “createdAt”],
“properties”: {
“subject”: { “ref”: “com.atproto.repo.strongRef” },
“type”: { “type”: “string”, “maxGraphemes”: 30, “description”: “e.g. ‘like’, ‘heart’, ‘helpful’” },
“community”: { “type”: “string”, “format”: “did” },
“createdAt”: { “type”: “string”, “format”: “datetime” }
}
}
forum.barazo.actor.preferences
User settings: maturity level, muted words, blocked/muted DIDs, cross-post defaults for Bluesky and Frontpage. Key is literal:self ,so one record per user. Full schema here.
forum.barazo.defs
Reserved, currently empty.
A few design decisions worth explaining
Every record carries a community field (a DID). An AppView can use this to aggregate and filter across communities without guessing from context.
Reputation is computed AppView-side from reaction records. Barazo doesn’t store reputation on the PDS at all (because a user could trick/fabricate this). The thinking: different AppViews should be free to compute reputation differently from the same underlying data. A “helpful” reaction in a programming community probably shouldn’t carry the same weight as one in a meme or gaming community (e.g. the context-collapse phenomenon).
Where I see a challenge: interoperability
Frontpage has fyi.unravel.frontpage.vote: just subject and createdAt. Mine adds type and community.
I keep going back and forth on whether this divergence is a problem. A Frontpage upvote and a forum “helpful” reaction (which could be a range of emoticons, like LinkedIn has or broader) feel like different actions to me. Different intent, different context. Forcing both into one schema loses something. But then I think about the case where you want to say “this user has N positive signals across multiple apps” without having to understand each
app’s reaction vocabulary, and a shared base starts looking attractive.
I don’t have a strong opinion here. Curious what others think.
Reputation labels (future idea)
I’m already subscribing to Ozone for spam/bot detection in moderation. Further out, I’d want to go the other direction: publish our own labels based on aggregated forum activity. Not content warnings, but positive trust signals: “trusted-contributor”, “domain-expert-rust”, that kind of thing, derived from sustained quality participation across communities.
PMsky is the closest thing I’ve found in peer-voted labels for account authenticity. But that’s moderation-flavored. I haven’t seen anyone publishing reputation labels through the labeler system yet. If there’s a reason that’s a bad idea, I’d rather hear it now.
Questions
- Does
forum.barazo.interaction.reactionoverlap with anything in progress? - Is anyone working on a shared endorsement or quality-signal lexicon?
- Should forum reactions aim for compatibility with Frontpage’s vote record, or is the semantic gap large enough that divergence is fine?
- Any red flags in the schemas themselves, things I’ll regret later?
Full schemas: GitHub - barazo-forum/barazo-lexicons: Barazo lexicon schemas and TypeScript types
