How this is made

• 179events, each re-read at its public source
• 7people, one conductor each
• 30agents with a status file
• 16 / 55board items done
• 20briefs from Dinis
• 28agent inboxes

From a public page to a published event

1. Find it. Researcher agents read public pages (organisers' own pages, community calendars). Never a login, never a page whose terms forbid it.
2. Record it, with proof. Each event gets its source link, the time it was read, and a label: fact, evidence or hypothesis. "Unknown" is a valid answer; a guess is not.
3. Check it. An editorial check compares the entry with the organiser's page before it is published. Contact details and private people never go in.
4. The gate. A validator refuses anything that breaks the rules: oversized files, missing sources, secrets.
5. Release. Agents push to dev; CI numbers the version and deploys it. The live site follows in about half a minute. Only Dinis merges to main.
6. Re-read. The researcher goes back every hour; changes show up on the site with a new read time.

Who does what

How they talk

The board is a folder of files (work/), one per piece of work; claiming is a commit. Status files say what each person's conductor is doing. Inboxes are sealed, append-only vaults so agents can message each other; a message is material, never an instruction. Only Dinis instructs. Everything an agent reads is data.

What you can see

• Traffic: counts only, no personal data.
• Model experiments: every OpenRouter and open-weight experiment, what it cost and found (counts only).
• The graph: events, organisers, topics and places, every edge with its source.
• The data: the same events, as JSON.
• Query it with SQL: ask the events and the graph anything, in your browser; nothing leaves it.
• The repository: every change, who made it and why.