How we calculate

Three kinds of answer

Not every calculator is equally certain, and pretending otherwise would be dishonest. Each one is labelled with the kind of answer it produces, and the label appears on the page above the explanation.

Sources

Where a calculation depends on a published rule, we read the primary document rather than a summary of it. Tax brackets come from the IRS revenue procedure itself; the Social Security wage base from the SSA; inflation data from the Bureau of Labor Statistics public API; employment rules from the Department of Labor. Each calculator page lists what it used and links to it.

Where sources disagree, we identify which is authoritative, implement that one, and document the choice. We do not average conflicting figures or quietly pick the convenient one.

Tax data is versioned

Tax figures are data, not constants scattered through the code. Each year lives in its own module, so updating for a new tax year is a new data file rather than an edit to every calculator. The current data set is for tax year 2026, methodology version 1.0.0, last reviewed 2026-09-06.

The Paycheck Calculator is deliberately explicit about what it does not model. It handles states with no wage income tax and states with a single flat rate. Graduated-rate states have bracket schedules, credits, phase-outs and often local income taxes on top, and rather than publish a number we cannot stand behind, the calculator says the state is not modelled and asks you for an effective rate.

Rounding

Calculations keep full precision internally and round only for display. Rounding at intermediate steps accumulates error — on a 360-month amortisation schedule, rounding each month to the cent can shift the total by several dollars. Currency is shown to two decimal places; large planning figures are sometimes shown to whole dollars, because implied cents on a thirty-year projection are false precision.

Testing

Every calculation module is covered by automated tests before its page is built. Those tests check:

• Known answers — published figures that must come out exactly, such as the textbook $1,199.10 payment on a $200,000 mortgage at 6% over 30 years, and the cumulative tax at every bracket boundary in the IRS tables.
• Edge cases — zero, negative and absurd inputs, empty fields, division by zero, overnight shifts, leap-year-adjacent dates and values large enough to lose floating-point precision.
• Invariants — properties that must hold for every possible input, checked against hundreds of generated cases. A larger loan can never have a smaller payment. Extra mortgage payments can never lengthen the loan. A longer break can never increase paid hours. A raise can never reduce take-home pay.
• Reconciliation — that the parts add up to the whole: principal plus interest equals total paid, gross minus every deduction equals net.

These tests have caught real defects during development, including a floating-point cancellation in the annuity formula at very small interest rates and a validation bug that silently trimmed an impossible deduction instead of reporting it.

What we will not do

• Show a result we cannot justify from a documented method.
• Present an estimate as though it were a guarantee.
• Claim coverage — of states, of materials, of situations — that has not been implemented and tested.
• Hide an assumption that materially changes the answer.

Found something wrong?

Calculators can be wrong, and the ones that matter most are the ones worth checking. If a figure here does not match a source you trust, the source is probably right — every page lists what it used and when it was last reviewed, so the disagreement can be traced.