A register is only worth as much as the document it is measured against. Without one, "curated" means whatever the curator felt like on the day, and the reader has no way to tell a careful selection from an accumulation.
So the criteria were published before the first entry, deliberately and in that order. What follows is the bar in prose: each criterion stated as what we actually check rather than what we would like to be true, then the categories excluded outright, then the rule that applies to our own publisher, then the part that is usually left out — how much of what is sent to us gets refused.
What this is, and who it is for
This describes version 1 of the entry bar. That phrasing matters and it is not a formality. The bar is versioned rather than quietly edited, every published version is archived rather than rewritten, and every entry in the register records which version it was admitted under. If you are reading this some time after it was written, treat it as a description of a version and read the versioned entry bar for the current text.
It is for three readers. The buyer who wants to know what an entry here is claiming. The operator deciding whether to submit. And the sceptic who assumes any listing site is selling something, which is a reasonable prior and one this document exists to answer.
One framing point before the criteria. This is a bar, not a recommendation. Clearing it means a business is real, independently owned, reachable, and was trading when someone looked. It says nothing whatsoever about whether the business is good, safe, solvent, competent or a sensible choice for you. We do not know those things and we do not claim to.
Criterion 1: it is a real thing you can buy, use, hire, read or visit today
Not a waitlist. Not a landing page with an email capture. Not a pre-launch announcement, a "coming soon", a concept, or a product that exists in a demo video.
What we actually check: that a visitor arriving from an entry page can do a specific, nameable thing. Buy this. Hire this person. Book this slot. Read this. Walk through this door during stated hours. If the reviewer cannot name the action in one short phrase, the criterion fails.
This excludes a surprising amount of genuinely interesting work, and it excludes it deliberately. A register of things that will exist soon is a register of intentions, and intentions decay faster than anything else on the internet.
Criterion 2: it is independently owned, and how we establish that
Not publicly listed, and not a subsidiary of a company that is.
What we actually check: something specific, recorded on the entry as the basis for the conclusion. That might be an ownership statement the business publishes itself, a company record, or a direct answer from the operator. What it is never is an inference from the size or polish of the website, because that inference is wrong in both directions constantly.
What happens when we cannot establish it: the answer is no, not a guess. The field that records how independence was established is left explicitly empty rather than filled with something plausible, and an entry cannot publish with it empty. This is enforced at build time rather than by anyone remembering.
Criterion 3: it is not primarily an advertising, affiliate, dropship or lead-gen vehicle
Carrying advertising does not disqualify a business. Having an affiliate programme does not disqualify a business. Plenty of good small operations pay for themselves that way.
What we actually check: whether the arrangement is the product. The question is what a visitor is being offered. If the answer is goods or services, and the advertising pays for the site, that is a business with advertising on it. If the answer is that the visitor's attention or enquiry is the thing being monetised, and the apparent offer exists to produce it, that is a different kind of site and it is out.
This is the criterion that most depends on judgement, and we should be honest that a reading can be wrong. It is made by a named person, it is recorded, and it is the one most likely to appear in a correction later.
Criterion 4: a named person examined it and can describe it in our own words
What we actually check: that an entry carries a reviewer's name and a review date, and that the description was written here rather than lifted.
The description is written by us. Operator marketing copy appears on an entry only as an attributed quote with a capture date, never as our description of what a business does. This is not a stylistic preference. The moment a register republishes marketing copy as its own text, it has become a distribution channel for claims it has not checked.
This criterion is the reason the register grows slowly, and it is the reason it can grow no faster than the hours available to look at things. That constraint is a feature and it is also, honestly, the one most likely to be under pressure later.
Criterion 5: it passed a dated liveness check by the published method
What we actually check: the procedure at our method, which has a first automated step and several subsequent steps read by a person, and which produces one of four outcomes rather than a pass or fail. How we check a business is still trading sets out the whole procedure including the parts that are slow, manual and occasionally wrong.
The binding between the check and the date is mechanical. Every date shown on an entry corresponds to a row in a check log. It cannot be typed in by hand, carried forward from a previous version, or refreshed by a deployment. If the log has no matching row, the build fails rather than publishing the entry.
That is a small piece of engineering and it is doing more work than anything else on this list. A date that a build can generate is not evidence, it is decoration.
Criterion 6: any claim we repeat is quoted and attributed, or omitted
What we actually check: that every factual statement on an entry is either something we verified with a method and a date, or something the business published, quoted with attribution and a capture date. There is no third category.
If a business says it ships within a week, we do not write "ships within a week". Either we quote them saying it, attributed and dated, or we leave it out. The reason is straightforward: an unattributed restatement transfers the claim's credibility from them to us, and we have not earned it.
The same rule closes off the most tempting shortcut in this whole category — writing a warm, confident entry description assembled from the operator's own words with the quotation marks removed.
The automatic exclusions, and why each one is out at launch
Whole categories are excluded regardless of how well an individual business would otherwise clear the six criteria:
- Adult commerce.
- Gambling, betting and prediction markets.
- Lending, credit, debt management and related financial services.
- Supplements, health products and anything making a health claim.
- Multi-level marketing and recruitment-driven income schemes.
- Crypto tokens, exchanges and trading services.
- Anything requiring a licence, registration or professional authorisation we cannot verify.
The reason is the same in every case and it is not a moral one. Each of these sits in a space where the checks that would make a listing responsible are checks we are not equipped to perform. We cannot verify a licence we have no access to. We cannot evaluate a health claim. We cannot assess whether a financial product is what it says it is. Listing something in these categories would mean publishing a dated claim we could not actually stand behind, which is the exact failure the rest of this document exists to prevent.
It is worth being explicit about what this exclusion is not. It is not an accusation against anyone operating in these categories, and it does not imply that a business in one of them is disreputable. It reflects our own limits.
Reopening any of these is a legal decision, taken with advice, and never an editorial one. No amount of a good business asking, and no persuasive individual case, moves a category out of that list. If one ever changes, it will change as a versioned amendment with the reasoning recorded, not as an exception granted quietly.
The disclosure rule for publisher-owned properties
Sites owned by this publisher may be listed when they meet the same entry criteria. Each such entry must prominently identify itself as Publisher-owned and explain the relationship. Ownership does not establish quality, independent testing, or an endorsement.
Publisher-owned entries receive no preferential position in browsing or discovery. Buying a product or service from us does not buy inclusion, faster review, or a better position. Who pays for this site explains the funding boundary.
The three-entry gate on category pages
A category page does not publish until it holds at least three entries that meet the bar and has an editorial introduction written by a person.
This is a publish condition enforced in the build rather than an intention. The reasoning: a category page with one entry is not a category, it is a page built to exist. Two is not much better. The pattern of publishing thin category pages in the hope that they fill up later is precisely how a register turns into a set of templated pages with a business name substituted in, and the gate is there because good intentions have a poor record against that particular drift.
The practical consequence is that the register will have visible holes — categories that clearly should exist and do not, because there are only two things in them so far. We would rather show nothing than show a page built to be arrived at.
The refusal rate, and what it means
Here is the number this section is supposed to contain, and here is why it is not here.
The register currently holds no published entries, so there is no refusal rate. Not a low one, not a provisional one, not an internal estimate. There is no denominator. Publishing a figure now would mean inventing it, and a fabricated number on the page that explains our standards would be a self-refuting document.
A conforming observed figure comes from a submission log — refusals over submissions, with the period stated. This edition supplies no such log, so the rate is not measured; the versioned entry bar defines the calculation.
A refusal rate is easy to misread. A high rate is not a boast about rigour; it may equally mean the submission guidance is unclear or attracts unsuitable submissions. A low one is not evidence of laxity; it may mean the bar is well explained. The number is interpretable only beside the patterns behind it, which is why what we turned down and why and the refusal file describe the method through anonymised patterns, never a named business.
The rate is not a target. A refusal rate managed toward a number has stopped measuring anything.
How the bar is versioned, and how you will know when it changes
This document describes version 1.
Every subsequent version is recorded in the method changelog with four things: what changed, why, what it invalidated, and which entries had to be re-reviewed as a result. Published versions are never rewritten in place, so an entry admitted under an older version can always be read against the text that admitted it.
The fourth item is the one to watch. A bar that gets stricter should trigger re-reviews. A bar that gets looser and triggers none is a bar that was loosened to admit something specific, and the changelog is structured so that you can see the difference from outside.
The bar applies at every re-check, not only at admission. Meeting it once does not keep an entry in the register. If a business changes such that it no longer clears a criterion, the entry is removed, and removal is not an accusation — the note records only that it no longer meets the bar. That is a separate thing from a business that has stopped trading, which becomes a tombstone rather than a removal, and what happens when a listed business disappears covers that sequence in full.
If you are an operator weighing all this up, how to submit a business sets out what actually happens to a submission and every reason we might come back with a no. It is free, it is slow, there is no priority, and refusal is common.
In Plain English
Six things must all be true before anything appears in the register:
- It exists now and you can do a nameable thing with it today.
- It is independently owned, established by something specific and recorded, or the answer is no.
- The advertising or affiliate arrangement is not the product — attention and enquiries are not what is on offer.
- A named person examined it and wrote the description here. Operator copy appears only as an attributed quote.
- It passed a dated check whose date is bound to a log row and cannot be generated by a build.
- Every claim is checked-and-dated or quoted-and-attributed. There is no third category.
Seven categories are excluded outright, because they need checks we cannot perform. Reopening one is a legal decision, never an editorial one.
Our own publisher's properties must meet the same bar and carry a prominent Publisher-owned disclosure; they receive no preferential placement. Category pages need three entries before they publish.
The refusal rate is not measured because this edition supplies no assessment log. No estimate is substituted.
Clearing all of this means a business is real and was trading when someone looked. It is not a rating, a recommendation or an endorsement — why we don't rate anyone explains what that costs you and why we accept the cost.
Read the versioned entry bar → /entry-bar/