How we calculate

Where a standard formula exists, we use it and name it. The mortgage and loan calculators use the standard annuity formula. Metabolic rate uses Mifflin–St Jeor, with Harris–Benedict shown alongside for comparison because the two disagree and the difference is worth seeing. Body composition uses the US Navy circumference method, with its published accuracy stated.

Where more than one convention exists, the calculator asks rather than choosing silently. Sample versus population standard deviation is the clearest example: the two divisors give genuinely different answers, and a calculator that picks for you is wrong roughly half the time.

Money is rounded to the currency’s minor unit — two decimal places for most currencies, zero for the yen — using round-half-away-from-zero.

This is less trivial than it sounds. The obvious implementation, Math.round(x * 100) / 100, returns 1.00 for an input of 1.005, because 1.005 × 100 is 100.49999999999999 in binary floating point. Our rounding function corrects for that representation error, so 1.005 rounds to 1.01 as an accountant would expect.

Amortisation schedules round each row to the cent as it is produced, exactly as a loan servicer does, and adjust the final payment so the balance reaches precisely zero. This is why total interest can differ by a few cents from the shortcut of payment × number of payments − principal. That shortcut is not what anyone actually pays.

Long series — a 360-month schedule — are summed with compensated summation rather than a plain running total, so accumulated floating-point drift cannot reach the displayed figure.

All date arithmetic is performed in UTC. Date-only calculations done in local time break across daylight saving transitions, producing off-by-one-day results for anyone in an observing time zone — a class of bug that simply cannot occur if you never use local time.

Calendar units are treated as calendar units. Adding one month to 31 January gives 28 or 29 February, not 3 March. Age is counted as complete years, then complete months, then remaining days from the anniversary — which is why it is not the same as total days divided by 365.

Every calculator is covered by an automated suite that verifies it:

• returns a valid result from its own defaults, so nobody meets a broken page;
• never renders NaN, Infinity, undefined or null anywhere in its output;
• keeps every table row aligned with its column headers and every chart series aligned with its labels;
• keeps breakdown percentages within 0–100;
• throws a specific, actionable error rather than crashing on empty or negative input;
• ships with a worked example whose inputs actually produce the stated result.

Beyond that, the shared financial, date and cryptographic functions have their own tests against independently derived values and published test vectors.

When input cannot be used, the calculator says what is wrong and what to do about it. “A payment of $100 does not cover the $114.50 of interest charged each month, so the balance would never fall” is useful. “Invalid input” is not, and the test suite specifically rejects it.

Each of the 278 calculators carries the date its formula and content were last reviewed, shown at the bottom of its page. Reviewing means re-checking the formula against its source, confirming the worked example still computes, and re-reading the assumptions and limitations for anything that has become misleading.

Finance, health and tax calculators are the priority for re-review, because their inputs — rates, thresholds, clinical guidance — change over time in ways that arithmetic does not.

They are estimates for information. They are not professional financial, medical, legal or tax advice, and no result here should be the only basis for a consequential decision.

Health calculators in particular describe population averages. BMI cannot distinguish muscle from fat; metabolic rate equations carry roughly ten percent error; body fat from tape measurements is accurate to within a few percentage points at best. Each of those limits is stated on the calculator that has it.

Please tell us. Include what you entered and what you expected. A reproducible report is genuinely valuable and we would rather fix an error than not know about it.