How we collect, verify, and date every fact

What the Atlas is

The Atlas is a verified, dated map of the agent economy — the vendors, skills, harnesses, and frameworks people actually build and buy with. Every entry is checked against its primary sources, stamped with the date we last verified it, and carries a confidence level. When we don't know something, we say so rather than fill the gap with a guess.

Every fact carries its provenance

Each catalog entry records three things alongside the fact itself, and we show them on the page:

How we verify

We collect from primary sources — a vendor's own site and pricing page, GitHub repositories and release notes, official docs, and public discussion — then confirm each fact against the source before it goes in. The agent space changes weekly, so freshness is the point: we re-check entries on a schedule and move the last-verified date only when we've actually re-confirmed.

Stale beats wrong.An out-of-date fact we're transparent about is recoverable; a confident fact that's wrong is not. We never scrape sources that prohibit it.

Confidence and the review queue

Not every fact clears the same bar, so we don't treat them the same:

• High confidence — confirmed against a strong primary source and recently checked. Published.
• Uncertain — conflicting or thin sourcing. Held for human review before it appears, never auto-published.
• Low confidence— we can't stand behind it yet. Not published; the field reads unknown until we can verify it.

How we score — and what we will never do

Where we have enough real signal, an entry may carry an Atlas score: a transparent, computed composite built only from things we can actually measure — GitHub momentum (stars and release velocity over time), adoption signals where a source publishes them, security-audit status, and freshness. When a score is shown, the inputs are shown with it.

It is an editorial index, not a star rating, and it is never emitted as a star rating in our structured data. We do not invent ratings or review counts. We removed the fabricated star ratings the site used to generate, because fake data is worse than none — it's the one lie this project can't afford.

The firewall: the opinion is never for sale

We take vendor money — clearly-labeled placement, function sponsorship, and newsletter sponsorship — and it never moves a ranking. Verdicts are written without vendor input. A sponsored pick is always labeled as sponsored, walled off from the ranked verdict, and the scoring criteria are shown on the page and applied to everyone the same way, including the sponsor and including us.

Every paid relationship is listed on our disclosures page.

Found something wrong?

Tell us. Every entry can be corrected, and anyone listed can request changes or removal. Use the report link on the entry, or contact us directly — corrections are the fastest way the Atlas gets better.