How our AI editorial works
The intelligence on RKLB.FYI is written by AI agents, checked against the data, and reviewed by people before it goes live. Here is exactly how that pipeline runs.
This content is AI-generated
The briefings, market digests, and risk-signal timelines you see across our intelligence pages are drafted by AI agents, not written by hand. We say so plainly: wherever a surface was machine-generated, it carries an "AI-generated" disclosure bar at the top showing when it was made, which model produced it, and whether a person reviewed it.
This page is our disclosure under the EU AI Act (Article 50), which requires that people be told when they are reading AI-generated content.
Domain desks do the writing
Each kind of analysis is owned by a dedicated agent "desk" — a market digest desk, a briefing desk, a risk-signal desk, and others. A desk gathers the underlying facts first (stock data, SEC filings, launch records, news), then asks its agents to synthesize that evidence into plain-language analysis. Desks only ever work from data we have already collected; they do not free-associate.
Validators check every claim
Before anything can be published, automated validators grade the draft. They confirm that every factual claim is tied to a real source in our inventory, that the numbers in the headline match the data we computed (not numbers the model invented), and that the confidence on each claim clears our threshold. Citations are checked for provenance. If a check fails, the draft cannot auto-publish — it is held for a person.
Our guardrails are designed to fail closed: when in doubt, content is held back rather than pushed out.
People stay in the loop
Drafts that need a human land in an editorial review queue, where an editor can approve, revise, or reject them. When a person has reviewed a surface, the disclosure bar says so ("reviewed by Editorial team"). When no person has reviewed it, we say nothing — we never claim a human reviewer who did not exist. Honest disclosure matters more than looking polished.
Higher-stakes or lower-confidence content always routes to a person. Auto-publishing is off by default and only enabled, per desk, once a surface has proven itself.
Our promise on quality
We would rather publish less than publish something we cannot stand behind. Every published surface is sourced, validated, and — where required — reviewed. If you ever find something that looks wrong, it is a bug to us, not an acceptable margin of error.