In 2010, Carmen Reinhart and Kenneth Rogoff published a study on public debt. Both were well-known economists. The finding was short: in countries whose debt passed 90 percent of national income, growth stalled. The number spread fast. It was cited again and again in debates on debt.
In 2013, three researchers at the University of Massachusetts opened the study’s data file. They found three problems: a spreadsheet formula had left five countries out, some years had been excluded, and countries were weighted in an unusual way. With all three corrected, average growth came to plus 2.2 percent, not minus 0.1 percent. The authors acknowledged the formula error and stood by their overall conclusion.
Today the story is mostly told as an Excel error. Yet most of the gap came from the other two problems.
Repetition is not verification.
The more a number is repeated, the less it is checked. Each new source leans on the one before. Few go back to the start of the chain.
The same happens with everyday information. A tax rate, a fee, a percentage in a regulation. Dozens of pages explain them. Most quote each other; some were written years ago.
Checking a number at its source.
- Go to the original document. The law, the regulation or the institution’s own table. Sometimes the table sits in an annex, not in the text.
- Check the date. A secondary source can be accurate and still out of date.
- Test against a known result. Can you reproduce an amount the institution has published? If not, the method is missing something.
Every number a brand publishes speaks for it.
A rate on your website, a measurement in your catalogue, a market size in your deck. Readers take it as your word. If it turns out wrong, more than the number gets corrected; trust takes the hit too.
It matters more in the age of AI.
AI tools learn from these same pages. They readily repeat a number that is repeated often. As content speeds up, so do the errors.
We speed up content with AI; we check every number against its original document. We show the source on the page. The final word stays with the team.
We built our free calculator site, Hesapla 7/24, the same way. For civil servants’ travel allowance, an old source put the lodging rate at 50 percent. Turkey’s 2026 Budget Law says 60 percent for the first ten days, in a footnote to its schedule. Our travel allowance calculator therefore works from the schedule itself.
Where to start?
Look at the pages on your site that carry numbers:
- Does each number show its source?
- Is that source the institution that first published it, or a page quoting it?
- When was the number last checked?
If you want your site’s content built fast and with clear sources, let’s talk. We start with a short intro call.