Methodology: how these tools are built and checked

One tool, one file, no network

Each tool is a self-contained module: a manifest describing it, a model holding the arithmetic, and a UI that renders the model’s output. The model is pure functions over numbers — it cannot fetch, store or log anything, because it has no access to the network at all.

That is a structural guarantee rather than a promise. The one exception is the weather tool, which reads current conditions from Open-Meteo, a free and open weather API that needs no key. Every other tool computes its answer from what you typed, in the page, and keeps working with the network switched off — which is the easiest way to verify the claim yourself.

Where the numbers come from

Classifications are copied from their published source, not reinterpreted. Body-mass-index bands follow the WHO adult table; healthy weight ranges are the same formula inverted, not a separate dataset. Time-zone conversion uses the IANA time-zone database — the same one your operating system and every server you use already relies on — so daylight-saving transitions resolve the way they do everywhere else.

Everything else is standard arithmetic or a published definition: unit conversions use the exact defined factors (2.54 cm per inch, 0.45359237 kg per pound) rather than rounded approximations, and financial tools state their compounding assumption on the page instead of leaving it implicit.

How a result is checked before it ships

Two things are tested for each tool: that the arithmetic is right, and that it is right at the edges. Correctness is checked against a worked example with a known answer, computed by hand from the published formula.

Edges get their own tests, because that is where tools quietly break. Where a tool accepts two input paths to the same quantity — metric and imperial, decimal degrees and degrees-minutes-seconds, 12-hour and 24-hour time — the two paths are asserted to produce the same answer, which catches a conversion factor that has drifted. Inputs at and just past each category boundary are tested on both sides, since a threshold that belongs to the higher band will otherwise be off by one category at the exact cut point.

Rounding, and which way it goes

Intermediate values are never rounded; the truncation happens once, at the output. The triangle tools show angles to four decimals and lengths to six, because passing through inverse trigonometry amplifies error more than plain arithmetic on lengths does. Where a result collides with floating-point noise, the matrix tools snap a value to a short decimal only if it sits within a relative 1e-12 of that decimal, and treat anything below 1e-12 as an exact zero — so an integer matrix inverts to clean values while a genuinely high-precision result still returns unflattened.

Where a count is rounded, the direction matters more than the decimal place. Bag counts always round up, because buying one bag short stops the pour while one bag extra costs the price of a bag; position sizes always round down to whole lots, because rounding up would risk more than the percentage you set. Each direction is chosen so the error lands on the safe side, and each is stated on the tool page rather than left to be discovered.

What the tools refuse to compute

A calculator that always returns a number is not safer, only harder to catch. So a set of inputs is refused instead of approximated: an interior angle of 0 or 180 degrees, three sides where the sum of two equals the third, a stop price on the wrong side of the entry, matrix rows of differing length, a singular matrix, and a timecode that drop-frame numbering skips entirely. Refusing names the problem; approximating hides it.

One refusal is not about arithmetic at all. A pathological regular expression can backtrack exponentially, and JavaScript has no way to interrupt a running exec, so the regex tester runs matching on a worker thread and terminates that thread on timeout — killing it is the only hard interrupt the platform offers. Pattern length and test text are capped for the same reason: to keep a paste accident from freezing the page.

What these tools are not

They are calculators, not advisers. A BMI band is a screening label, not a diagnosis — it cannot tell muscle from fat, and it reads athletes, older adults and children poorly. A dosage or finance figure produced here is arithmetic on the numbers you supplied; it does not know your situation, your jurisdiction or your currency’s rules.

Where a tool has a known limitation, the limitation is stated on the tool’s own page rather than buried here. If you are making a decision that matters, use these results as one input and check the primary source.

Reporting a wrong answer

If a result looks wrong, the most useful message includes the inputs you used, the output you got, and the answer you expected. That is usually enough to find the cause in one pass. Corrections are the single most valuable thing a reader can send, and they are treated as a bug rather than an opinion.

Corrections and questions

Found an error, or want to know which formula a tool uses? Write to [email protected] with the inputs and the expected result — corrections are the most useful thing you can send.