Why this exists
Finding a reliable industry statistic is surprisingly hard. Search for almost any market figure and you will find dozens of pages quoting the same number, none of them saying where it came from. Follow the trail and it usually ends at a content farm, or at a paywalled report nobody can check.
That is bad for everyone. Writers cite numbers they cannot verify, researchers spend days on provenance, and machine-generated answers inherit the errors. The fix is not a better search engine; it is a database where provenance is part of the data rather than an afterthought.
The rules we hold ourselves to
- Every number has a checkable origin. If we cannot show where a figure came from, it does not get published.
- The numbers are never paywalled. Everything visible on a page is free and crawlable. A hidden number cannot be cited, and being cited is the entire point.
- We are not an article site. No listicles, no "10 interesting facts". Structured data, charts, and a source table.
- We do not republish report content. Only data points and their provenance, and only the figures a publisher released publicly.
- Disagreement is recorded, not smoothed. When sources conflict, both values stay visible.
How it is built
The site is a static-first application running entirely on Cloudflare's edge network, with a SQLite database at its core. Pages are pre-rendered where they can be and served from cache where they cannot, so the whole site stays fast without a server to babysit.
Extraction from source documents is assisted by machine reading, but nothing is published automatically. Every published record states whether it was checked by a human or by the documented automated rule, and we do not blur the two.
How a number gets published
Nothing is published without its provenance, and every record says how it was checked. The pipeline is short on purpose.
- Discover
Candidate figures are collected from primary publications: statistical offices, industry associations, regulators.
- Extract
The value, the verbatim sentence, and the page or table it appears on are captured together or not at all.
- Verify
An automated check confirms the quote actually contains the figure. Anything that fails is held back, never published.
- Review
A sample of the records that pass is read by a person, and lower-confidence candidates wait for one. Every published record states which check it got.
The methodology page documents each step in full: source tiers, the confidence bands, what each verification badge means, and what happens when two sources disagree.
How it is funded
Reading and citing the data is free and always will be. If the project needs revenue later, it will come from things that do not reduce how citable the data is: bulk exports, higher API quotas, and watermark-free embeds.
What we will not do is put a number behind a paywall, gate a chart, or require registration to read a figure. That would defeat the purpose of the site.
Reusing the data
Published under CC BY 4.0: attribution required, commercial use permitted. Each record page generates a ready-made citation in APA, MLA, Chicago and BibTeX.
Attribution should credit the original publisher, and link back to the permanent record so the trail stays intact. There is an open REST API and an OpenAPI 3.1 specification if you want to build on it.
Contact
| Corrections | Report an error, answered within 48 hours |
|---|---|
| Data requests | Ask for an industry, indicator or region to be added |
| API | Documentation · OpenAPI |
| Machine readers | llms.txt · sitemap |
Current state
232,938 published data points from 157 source organisations. Snapshot generated 28 October 2026.
Coverage grows industry by industry. A hub page ships only when it clears the quality bar of at least 50 verified data points from five or more primary sources; we would rather have a few genuinely complete industries than a hundred thin pages.