Methodology: how our calculators are built and tested

How a calculator is made

It starts with a question people actually search for. We measure that demand and look at which inputs and questions the best existing calculators cover. The formula comes from a cited source: a law or regulation, a technical standard, a government agency, a professional body or the math literature.

The calculation logic only receives your entries and only returns the result, with no access to the internet, the clock or random numbers. The same logic runs in your browser and in our tests.

Test cases

Every calculator has at least four test cases whose answers don’t come from our own code: they come from worked examples in the source, from the text of a rule or from an independent calculation. On top of that, 300 random valid inputs make sure the calculator never crashes or returns an invalid result.

Every test case is rerun each time we publish. If a result doesn’t match, the page isn’t published, and a page that is already live is never replaced by a version with errors.

Independent review

Before a calculator is published, a second AI model, different from the one that wrote it, works out new examples from scratch. It only sees the description, the formulas and the sources, never the program code. The page is approved only if all of its results match the calculator, and those examples become permanent test cases.

Some calculators are also recomputed by an independent script written in a second programming language.

Rules for sources

We prefer primary sources: laws and regulations, government agencies and statistics offices, central banks, international organizations such as the WHO, and medical guidelines. Encyclopedias such as Wikipedia only supplement them, for example for math definitions. Every linked source has to be on a vetted list of allowed sites and is checked before publishing.

Facts that change over time, such as tax rates, wage thresholds or federal limits, carry a review date. When it comes up, that text is the first thing we update.

How we use AI

Texts and calculation logic are written with the help of AI tools (language models from Anthropic). We say so openly because it helps you judge the content. Publishing doesn’t rest on the model alone but on the test cases, the independent review and the automated checks described above. The person responsible for the content is named below.

Mistakes and corrections

Every calculator has a “Report a problem” button. Each report is recalculated with the values you entered. A confirmed fix goes through the same checks as a new calculator, and each page shows the date it was last updated.

The calculators give estimates based on the formulas shown. They are not a substitute for professional tax, financial, legal or medical advice.