Methodology

A trust filter, not a long catalogue.

Know the problem. Define success. Pick the AI. Ship. Every listing is a written page with a fixed set of fields. The site is static. A page is published only after a person has edited it.

What we assume before a listing is useful

You know and love the problem.
You can define success criteria.
You are choosing AI that matches that problem. This is where listings help.
You will get it live.

We do not write handoff-to-engineering playbooks. We judge whether a tool deserves time, effort, and money for a person who will ship.

How a listing gets here

A listing is researched and written against the required fields.
An incomplete listing is not published. The required fields have to be filled in first.
The body argues the verdict in UK English, with sources, not adjectives.
A person edits the page before it is published.

What we judge

Time to live. Money to useful. Effort to own. Problem-fit. Lock-in. See the rubric for how those map to the four optional scores on a listing.

What counts as a source

Primary vendor docs, pricing pages, licences, and first-party changelogs. Second-hand round-ups are not sources. If the product and the docs have not been opened, the page is marked as an example, or it is not on the site.

Verdicts

A verdict is for a job in one section, not for a company. Yes means the tool deserves time, effort, and money for that problem. Yes, with caveats means the lane is narrow. No means skip the demo: wrong problem, too much lock-in, or it will not get you live.

Scores are optional

Scores are directional, out of ten, and only used when we have enough evidence. We would rather omit a score than invent precision. They are not lab measurements.

Affiliates

A listing says whether an affiliate programme is unknown, absent, or available and not joined. We do not publish rates.

Example listings

Example listings are marked on the page and in the catalogue. Each one says it is not a real review. Do not treat them as recommendations.