A calculator is only worth the arithmetic behind it, so here is how the arithmetic on this site is checked.
- 6,308 automated tests
- Every formula is tested against worked examples and its edge cases — zero, negative and out-of-range inputs, and the points where it should refuse to answer rather than guess. Nothing is deployed while one of them fails.
- 627 calculators that show their working
- Each result can be expanded into the steps that produced it, so you can check the arithmetic rather than take it on trust.
- 57 structural checks per calculator
- A validator runs before every deploy: that the formula exists, that the inputs it declares are the ones the formula reads, that the article covers what the page claims to answer, and that the page has at least four real questions answered on it.
- 3,643 questions answered in full
- Answered on the page itself, not linked away to somewhere else.
Across 627 calculators in 15 categories, with 674,953 words of explanation written for them. Figures counted from the source on November 2, 2026, not estimated.
Every formula is a separate, tested module
A calculator here is not a form wired to a script. The formula lives in its own file that receives numbers and returns numbers, with no access to the page, so the same file that runs in your browser is the file the tests run. That is what makes the tests meaningful: there is no second implementation that could disagree with it.
The tests cover the awkward cases, not the easy ones
Checking that a loan calculator handles a normal loan proves very little. The tests concentrate on the edges: a zero interest rate, a term of one month, an amount larger than any real loan, a body-fat measurement that cannot physically exist, a date before the one it is measured from. In many of those cases the right answer is to refuse and explain why, and the test asserts the refusal.
Several of these tests have caught real mistakes. A negative-equity calculator reported that no loan was ever underwater, because at month zero the balance equals the value and the comparison was the wrong way round. A bond calculator compared an effective annual return against a nominal semiannual one and so invented a gap that could not exist. Both were found by a test, not by a visitor.
Structural checks before anything is published
Separately from the formulas, a validator reads every calculator before a deploy and refuses it if the page and the formula have drifted apart: if an input the page offers is not one the formula reads, if the article does not actually address what the page claims to answer, or if there are fewer than four real questions answered on it. Nothing is deployed while the tests or the validator fail.
Where the formulas come from
Published equations are used by name rather than approximated, and the name is stated on the page: Mifflin-St Jeor for resting metabolic rate, CKD-EPI 2021 for kidney function, Du Bois and Mosteller for body surface area, Devine and Robinson for ideal weight. Where a standard body sets the thresholds, those are used as published rather than rounded for convenience, including the lower WHO BMI thresholds for South and East Asian populations.
Where two accepted methods disagree, the page says so and shows both instead of quietly picking one. Four body-surface-area equations do not agree with each other, and a calculator that hides that is less useful than one that shows the spread.
What this does not do
These are calculators, not advice. A BMI figure cannot tell muscle from fat. A tax calculator works from the rates and thresholds it was written against, and tax rules change. A health calculator does not know your history, and none of this replaces a doctor, an accountant or a surveyor. Where a calculation depends on an assumption, the page states the assumption rather than burying it.
If you find a figure you believe is wrong, please report it. A specific example — the inputs you used and the result you expected — is enough to write a test from, and a wrong answer is worth fixing properly rather than patching.