August 18, 2026
/
4 Minutes
Evidence Beats Claims
An application that says the product is loved scores lower than one that shows the reorder rate. That is deliberate, it is enforced in code, and it is why we ask for almost no documents.
We check the public record instead of asking for attachments
Most marketplaces make you upload documents to prove you exist. Documents are the easiest thing in the world to forge and the slowest thing in the world to read.
So we query the public record directly: GLEIF for legal entity identifiers, Companies House in the UK, Brønnøysund in Norway, recherche-entreprises in France, Zefix in Switzerland, and VIES for EU VAT registration. Six public registries, plus sanctions screening and the domain record — how long the domain has existed, whether mail is actually configured, whether the web presence matches the claim.
A legitimate producer passes this without sending a single file. That is the point: verification should cost the honest applicant nothing and cost everyone else everything.
The cap, not the penalty
Inside the score, an unevidenced claim is capped rather than punished. A strong assertion with nothing behind it cannot reach the top band, but it does not drag the sheet down either.
We cap rather than penalise because penalising rewards silence — applicants learn to say less, and a file that says less is a file we cannot evaluate. A cap rewards proof without punishing candour. Say the thing; show the thing if you can; say “not yet” if you cannot.
What counts as evidence
Links and numbers, and they count double. Sales to date with units and channel. Committed pre-orders. Named retailer interest. Third-party data we can pull ourselves. Anything with a date and a source.
What does not count: awards nobody can verify, “viral” without a link, “thousands of happy customers” without a platform, and projections of any kind. Pre-launch, verifiable traction outweighs eloquent projection every time the two conflict — that rule is written into the instructions the reviewer works from.
And then a person decides
Every automated check produces input. None of them produce a decision. A reviewer reads the file personally within three business days and signs the outcome.
The machine’s job is to make sure that person is reading something true.
That is the whole method. Apply to Y-Score — thirteen questions, or request access to THE CATALOG if you already know how to sell.