The case file, in plain text
Receipts was a free evidence report for any product website. Live from build to retirement, it did exactly one thing.
You pasted a URL. It checked what the site claimed against the trail those claims should have left. Real testimonials come from people you can find. Real traction gets mentioned somewhere other than the site itself. Real companies have legal pages that load and a checkout that works.
A minute later you got a report you could share. Every line showed what was found, what wasn't, and the raw data behind it. It never called anything fake. The strongest thing it would ever say was claims outpace evidence.
The problem. Launch feeds are full of sites whose claims can't be told apart from their design budget. Every “trust checker” before this one answered with a verdict — legit or scam — which is exactly the overclaim it was supposed to catch.
The bet. Refuse verdicts entirely. Score only the gap between what a site says and the public footprint those statements should have left: registry records, archive captures, certificate transparency logs, findable testimonial authors, independent mentions, deliverable addresses, checkouts that resolve. Evidence found, evidence not found — never a character judgment.
The build. One scan fanned out across five check families — domain history, social proof, footprint, fingerprints, operational reality — each check reporting evidence with its confidence, and every unreachable record explicitly counted for nothing against the site. The engine wrote its findings into permanent, dated, sharable reports; a language layer enforced the no-verdicts rule down to the word level. Next.js App Router, Postgres, a live scan pipeline with RDAP, archive, and search corroboration.
What survives. The scanner is retired, but 8 real receipts it printed are frozen on file exactly as the engine wrote them, scoring from 57 to 100 out of 100 — the site you are reading serves them statically, byte for byte. Read them at recent, or read how every number was earned at /methodology.
One scan, five trails.
Domain history checked when the site really came into existence. Social proof checked whether the people in the testimonials exist. Footprint checked whether the traction claims left any mark on the wider web. Fingerprints noted how the site was built, but never held that against it on its own. Operational reality checked the boring proof a business is real: working legal pages, an address that receives mail, a checkout that resolves.
When a check couldn't run, it said so, and that counted for nothing against the site. That rule outlived the scanner: it is visible in every frozen report.
Receipts was designed and built by Lateisha at Schmade, who also built Sensei and Meerkat. The full methodology, the language rules, and what the scanner couldn't do are preserved at /methodology. Reach a human at hello@getreceipts.dev, or see what came next at schmade.com.