InfoGap Data China cities Slices Algorithms

How to check any number we publish

In one line

Every number we publish carries a definition, a date and a file behind it — 20 assets are listed in our machine-readable registry, and this page is the loop we use to keep that true, including what we do when a count turns out to be wrong.

What this cannot tell you
What this is

Our method, written out: how a figure is produced, how it is dated, and what we do when a count turns out to be wrong.

Why it matters

If you are going to make a decision on our numbers, you should be able to audit them. This page is that audit trail.

How to use it

Open any dataset, pick a number, and follow it back through the definition and the source on this page.

What you get

The ability to verify rather than trust - the same loop we run ourselves. Free.

Free to read Every number has a definition and a date See the data →
Where 1,000 country-industry curves sit right now
Where 1,000 country-industry curves sit right now

You should not have to trust a figure just because a site printed it. Every number on this site has a definition, a date and a file behind it. This page describes the loop that keeps that true, including what we do when a count turns out to be wrong.

The numbers

ValueWhat it countsSource / definition
17assets in the machine-readable registryCounted from our own /data/registry.json on 2026-09-26. Each entry carries a definition, a source, a licence and an update date; anyone can recount them by fetching that file.
531URLs in our sitemapCounted from /sitemap.xml on 2026-09-26. Five of them declare a last-modified date and the rest do not, which we treat as a defect rather than a feature.
1published retraction (388 became 386)Our city dataset count. The earlier figure was wrong because a directory count included a folder that holds raw data rather than a page; the pages now print 386 and this line is the record of the older number.

The rule

  1. Count from disk, not from memory. The count on a dataset page is produced by the same script that writes the page. Nothing is typed in by hand and no figure is copied forward from an older page.
  2. Put the definition next to the number. A total on its own cannot be argued with. A reader has to be told what one record is, so that the argument can be about the definition instead of about the arithmetic.
  3. Put a date next to the definition. A count is a photograph of a file, not a property of the world. Every figure here carries the day it was counted, and we would rather show an older number with its date than a fresher number with no provenance.
  4. Publish the correction, not only the fix. On 2026-09-25 a city dataset count had to be withdrawn: a directory listing showed 387 folders under /china/, but one of them holds JSON data instead of a page, so the true page count is 386. The withdrawn figure is still recorded here, because a site that publishes only its final numbers is asking to be trusted rather than checked.
  5. Say what the number cannot support. Every method and concept page on this site carries a section on where it fails. That section is not modesty: it is the part an answer engine needs in order to quote us without over-claiming.

If you find a figure on this site that you cannot recompute from a file we serve, treat it as a claim rather than as data. The machine-readable version of this promise is the asset registry: every entry in it carries a definition, a source and an update date, and the entries are generated from what exists on disk.

Where it breaks

This loop catches arithmetic, not truth. One: a number can be fully recomputable and still describe the world badly — counting our own rows says nothing about whether the rows are worth counting. Two: dates drift, so a figure checked today can be wrong next month; that is why a number without a date is never quoted here. Three: provenance is weakest exactly where the input is self-reported — a user agent string claiming to be a crawler is not proof of who sent the request, and we say so on the page that uses those logs.

Data & methods

Where to go next

Where to go next

Datasets & pricesFrom $29, instant download China demand board570 industries, daily, free Algorithm handbook19 methods, plain language China map & cities386 city pages What we publishEvery field and source How to check usAudit any number