Every calculator here is backed by a test file that runs its real engine and records each check it makes. The coverage report counts those checks. A count on its own is a weak claim, so each case is filed under one of five kinds, and the kinds are not equal. Reading the mix is how you tell a well-evidenced calculator from a well-counted one.
The five kinds, strongest evidence first
- Independent cross-check — 6,601 cases
- The expected value does not come from our own engine. It is either a figure published elsewhere — an official table, a worked example, a statute, a standard textbook formula — or the same quantity recomputed by an independent route, written into the test file next to the case.
- Golden — 16,272 cases
- The expected value was worked out by hand, with the arithmetic written into the test file so a reader can re-derive it. A number produced by calling our own engine does not count as golden.
- Boundary — 8,842 cases
- Zero, negative, empty, very large, or degenerate inputs, and the rounding or threshold points where a formula changes behaviour.
- Property — 192,567 cases
- An identity, reconciliation, or monotonic relationship that must hold for any admissible input — for example that a schedule closes to a zero balance, or that a bigger deposit never lowers the result.
- Structural — 14,039 cases
- Output shape and type sanity: the field exists, the number is finite, no NaN, the array is the right length. Real checks, but the weakest evidence here.
Independent and golden cases are the ones that can catch a wrong formula, because their expected value comes from somewhere other than the code being tested: 22,873 of the 238,321 total. Property cases dominate the count — they are generated in loops and a single property can contribute thousands — so a calculator with a large total and no independent cases is telling you less than the number suggests. Structural cases are the weakest: they check that a field exists and is a finite number.
What the numbers do not prove
They do not prove the site is right. A test file is written by the same person who wrote the engine, so a misunderstanding shared by both survives every case. That is the honest limit of this page and it is stated on the report itself: it is not an independent audit. What the cases do prove is narrower and still useful — that the engine computes what the method section says it computes, that it has not drifted since the last run, and that the edges behave.
A count is also not a coverage claim about your particular input. A calculator with a thousand property cases may still have one unhandled combination; the cases say what has been checked, not that nothing else exists.
Checking a figure yourself
- Open the calculator's row on its category page. Each row's address ends in the calculator's slug, so the evidence for one calculator can be linked on its own.
- Read the case names shown under each suite. They are recorded verbatim by the test runner, so a case that says a published figure is reproduced names the figure.
- Where a calculator declares an outside authority, its page carries a Checked against line linking that source with the date it was read. Only a declared source appears there — an independent case on its own is not treated as an authority.
- The whole report is downloadable as machine-readable JSON: per calculator, per suite, per kind, plus the test file that produced each one.
How the report is produced
One script runs every suite, reads the structured artifact each one writes, and regenerates the report. No number on these pages is typed by hand; a figure that disagrees with the artifact fails the build rather than publishing. The report was last regenerated on 2026-09-22, covering 282 suites.
