We build entries from public sources, credit each one, and publish only when an entry has complete facts, text we wrote and links to related pages. We read documentation and repositories rather than run a lab, and we say so on every page where it matters. Ratings come from member reviews, never from payment. Checked 31 October 2026.
The short version
Question | Answer |
|---|---|
Where does the data come from? | The open web: vendor documentation, pricing pages, GitHub repositories, registries. Each entry names its source. |
Do you run hands-on tests? | Not unless a page says it did. “Our take” sections are built from documented facts and the comparison, and they never claim a test we did not run. |
What is a rating here? | An average of member reviews, shown only when real reviews exist. See how ratings work. |
Can a provider pay to change a verdict? | No. Payment buys a labelled placement, never a rating, a verdict or an editorial position. See how we make money. |
What happens to thin entries? | They stay unpublished drafts. Nothing is hidden with a noindex tag. |
How an entry is made
- Collect. Entries for agents, add-ons, models and hosts start from open data: a repository, a registry listing, a vendor page. The source URL and name are stored with the entry and shown to readers.
- Credit. An imported or submitted entry carries its attribution. A community submission is credited to the submitter’s GitHub account.
- Draft. Everything starts as an unpublished draft.
- Check against the source. An editor compares the facts with the primary page: price, plan names, license, install command, supported platforms. A fact the source does not state is marked as unverified rather than guessed.
- Add our note. We write a short editorial note on the entry, and for larger topics a full written page.
- Run the gate. The entry publishes only if it passes the checks below.
The publishing gate
The gate is a script in our repository, so the rule is the same for every entry. An entry is publishable when all of these hold:
Check | What it requires |
|---|---|
Facts complete | The fields that define that kind of entry are filled. For an agent that means name, kind, vendor, website, pricing model, a pricing check date, license and source URL. Add-ons, models and hosts have their own lists. |
Text we wrote | A minimum amount of our own text that no other entry shares. The minimum is lower for directory entries and 2,000 words for written guides and comparisons. |
Related links | The entry links to enough related pages that a reader can keep going. |
Fresh | It was verified within the last 90 days. |
An entry that fails stays a draft. We do not publish a thin page and tell search engines to ignore it. We do not publish it at all until it earns a place.
What “tested” means on this site
We want to be plain about this because it is where review sites most often overstate. Most of our pages are built from documentation, repositories, pricing pages and changelogs. That is research, and a careful one, but it is not a benchmark run.
- When a page says we ran something, we ran it, and the page says what and when.
- When a page quotes a score, the score comes from a published source and the page names it.
- Vendor claims are labelled as vendor claims. “98% fewer errors” from a landing page is a claim, not a finding.
- If we cannot confirm a figure on its primary page, it appears as an open question, not a number.
Dates and freshness
Every written page shows the date its facts were last checked against the sources. Prices and plan names change often, so treat an old date as a reason to check the vendor’s page. Readers can report a stale fact through the corrections policy.
Neutral by design
Providers pay for placement, and their links run through /go/ with a sponsored label. Member links carry a user-generated-content label. Neither changes the order of editorial lists. The full rules are in the editorial policy and the affiliate disclosure.
Questions we get
Can I ask for my product to be tested? You can submit it. We check it against its sources. We do not promise a hands-on test.
Do you rank by popularity? Some hubs sort by GitHub stars or by votes when you pick that sort. The default order is stated on the hub.
Where do I report a mistake? Follow the corrections policy.
