Almost every listing site that ended up selling placement started out with a stated principle against it. That is not a cynical observation, it is a structural one. The pressures that produce paid inclusion are ordinary pressures: the register is small, growth is slow, someone offers money, and the first exception is always a reasonable one.
So rather than assert that we are different, this post does something more useful. It names each way this site could decay, and for each one it names the specific, externally observable change that would tell you it had happened. If you ever see one of these signals, you do not need to give us the benefit of the doubt. You have our own words.
What this is, and who it is for
This is written for the reader who has been burned by a site that was once worth reading. It is not a promise, and it is deliberately not framed as one. Promises are unfalsifiable. What follows is a set of tells, each of which you could check from outside, without access to anything of ours.
It is worth saying plainly at the top: publishing this prevents none of it. A public pre-commitment raises the cost of a quiet change; it does not make the change impossible. The point is to make sure that if we do go bad, we go bad loudly.
Failure 1: relaxing the entry bar to grow faster
How it would happen. The register stays small. Small looks like failure. Someone proposes that one criterion is too strict — perhaps the requirement that a named person examined the thing, because that is the expensive one. The bar drops, coverage grows, and nobody outside notices for months.
What you would observe. The entry bar is versioned, and every version change is recorded in the changelog with what it invalidated and which entries had to be re-reviewed. The signal is a version bump where a criterion is removed or softened and the changelog does not say which existing entries were re-checked against the new version. A bar that gets looser without a re-review is a bar that was loosened to admit something specific.
The second signal is faster and cruder: register growth accelerating sharply without a corresponding change in the published method. Growth in a hand-checked register is limited by hours worked. If it stops being limited by that, something else changed.
Failure 2: introducing a rating, a tier, or a badge
How it would happen. Not as a rating. It would arrive as something helpful: a "recommended for" label, an "established" marker, a sort order that puts some entries above others by default. Ordering is the mechanism by which a register becomes saleable, because once there is a top of the list there is something to sell.
What you would observe. Any field on an entry that expresses more-or-less rather than yes-or-no. Any default sort that is not neutral — alphabetical, random, or by check date. Any visual treatment applied to some entries and not others. Any badge at all.
Watch also for the softer version: a "featured" or "editor's pick" section that is described as editorial and is not explained anywhere. Why we don't rate anyone sets out what we give up by refusing this, including the things a reader genuinely loses.
Failure 3: letting a listed business pay us for anything at all
How it would happen. In stages, and each stage would sound fine. The five models we could switch to describes the destinations; this is the road. First, sponsorship of something adjacent to the register but not in it. Then a paid service sold to operators — a review, an audit, a consultation — which is not payment for a listing but is payment from a listee. Then the boundary erodes because the people making listing decisions are also the people invoicing.
What you would observe. Any commercial relationship between this site and a business in the register, in either direction. Any paid product sold to operators where the buyer's listing status is not irrelevant on its face. Any change to who pays for this site that removes an item from the list of things we do not take money for, rather than adding one.
Also watch for a specific and quite legible tell: a paid tier for faster consideration. Expedited review is paid inclusion with the receipt filed under a different heading. There is no paid queue, so there is nothing to jump.
Failure 4: publishing pages built from a template with a name swapped in
How it would happen. Someone notices that certain page shapes attract traffic reliably. A category crossed with a place. A trade crossed with a year. The template is written once and populated many times. Each page looks purpose-built. Collectively they are a machine.
What you would observe. The appearance of any URL pattern that combines two variables — a category and a location, a service and a town, an adjective and a noun. The appearance of pages that read identically apart from the substituted term. A sudden increase in page count that is not an increase in entry count.
This one has a hard structural rule attached: no page here is generated by combining a category with a place. The related guardrail is the three-entry gate — a category page does not publish until it holds at least three entries that meet the bar and has an introduction written by a person. A category page with one entry is a template page with better manners. The full list of formats we refuse is at what we refuse to publish.
Failure 5: quietly dropping entries instead of tombstoning them
How it would happen. Tombstones are unattractive. They make the register look like it is shrinking, and they are the visible record of businesses that did not make it. It would be very easy to delete instead — the entry simply stops existing, the register looks healthy, and nobody can tell the difference from outside.
What you would observe. This is the hardest one to detect, which is precisely why it is worth naming. The signal is an entry that was present and is now absent with no corresponding record. Deletions that leave no trace are the mechanism by which a register's history becomes unauditable.
Two caveats, both honest. An operator can ask for removal at any time, for any reason or none, and we honour it without announcement — that is a privacy commitment and it does mean some absences are legitimately unexplained. And a business that no longer meets the bar is removed rather than tombstoned, because a tombstone means "we confirmed it stopped trading" and nothing else. So absence alone is not proof of anything. A pattern of absences alongside a shrinking tombstone index would be.
Failure 6: hiding our ownership or buying preferential treatment
A publisher-owned site may be listed, but the relationship must be prominent on the entry. It must meet the same entry criteria, and ownership must never be presented as an independent review or a reason to move it ahead of other entries.
The observable failure is an undisclosed relationship, a fabricated review, or special placement that a reader cannot distinguish from the ordinary selection process. The entry bar and funding policy apply to our own sites as well.
What we have built to make each one harder
Some of these guardrails are enforced by the build. Some rely on judgement. It matters which is which, and the honest split is uneven.
Enforced mechanically:
- A displayed check date must correspond to a row in the check log. A date cannot be typed in, carried forward or refreshed by a deployment. Without a matching row the build fails rather than publishing the entry.
- Unknown values are
null. They are never defaulted, inferred or filled with a plausible value, and this is prevented at build time rather than by convention. - Schema validation rejects rating, score, rank, star and tier field names outright. They cannot be added to a record without changing the schema, and a schema change is visible.
- The three-entry gate is a publish condition on category pages, not a guideline.
Dependent on judgement:
- Whether a business is primarily an advertising or lead-generation vehicle. That is a reading, made by a person, and a person can be talked round.
- Whether a change to the method is a clarification or a relaxation. The changelog records the change; it cannot make us honest about how we characterise it.
- Whether an entry description has drifted from neutral into promotional. Nothing catches this except a second reader.
- Whether a commercial opportunity is genuinely outside the register or merely adjacent to it. This is the boundary that erodes first in most publications, and no build step protects it.
We would rather say which is which than let the existence of some automated checks imply that everything is covered. It is not, and no guardrail here is impossible to circumvent by someone who has decided to.
How to hold us to this
Report it. Anything factual that is wrong, stale or misattributed goes to our correction log. Reports are logged whether or not they turn out to be right, and the log records how each error was caught — by us, by a reader, or by the operator. A log in which every error was self-caught is either a very careful publication or an incomplete one, and the column exists so you can tell which.
Expect a note, not a silent edit. Changing a published fact without a visible note is the specific behaviour the correction log exists to make impossible. If you find a page that changed without one, that is itself the report.
No service window is claimed. This article supplies no dated queue log from which to calculate a median. Any implementation that states a window must publish the observed median and log its misses; an invented target is not evidence.
Read the versions, not the current page. Our entry bar in full describes a version. Versions are archived rather than rewritten in place, so the useful audit is not "what does the bar say now" but "what changed, when, and what was re-checked as a result."
Check the money page against the site. Who pays for this site lists every route by which money could reach a listing decision. If the site ever does something that page does not describe, the page is out of date or wrong, and either way that is worth telling us about.
In Plain English
Six ways this site could turn into the thing it criticises, and the tell for each:
- Bar drops to grow faster — a version bump that removes a criterion without saying which entries were re-checked.
- A rating appears — any field that says more-or-less rather than in-or-out, any badge, any non-neutral default sort.
- A listed business pays us — any commercial relationship with anyone in the register, in either direction, including a paid fast lane.
- Templated pages appear — a URL pattern crossing a category with a place, or page count rising faster than entry count.
- Entries vanish quietly — absences with no record, alongside a tombstone index that stops growing.
- Our own products get listed — an entry for a publisher-owned property, or an exception clause appearing in the exclusion rule.
Publishing this does not prevent any of it. It just means that if it happens, you were told what to look for first.
Read what we refuse to publish → /the-line/