Methodology and sources

Last updated: October 26, 2026. This page describes the site as currently built.

1. One engine, one result

All calculators run one calculation engine per country module. A calculation returns a single structured result object — lines with amounts, explanations and sources — and that same object feeds the screen, the PDF download and the automated test benches, so the three can never drift apart.

2. Parameters are data, with receipts

For the US calculators, rates, wage bases, brackets and deductions live in versioned data packs, not in code. Every parameter cites its primary source and the date it was verified: IRS revenue procedures and publications, Social Security Administration wage bases, the US Department of Labor's state unemployment provisions, state tax agencies — and for the reserve modules, the Canada Revenue Agency (T4127 formulas; the engine was arbitrated against the CRA's own PDOC calculator until the residual was $0.00 on every line of the test grid) and HMRC. One disclosed exception: the reserve Canada/Québec modules keep a small number of statutory constants (contribution base-credit caps and Québec formula coefficients, all verified against the primary publications) in code; their limits document names them, and moving them into the data pack is a written condition of that module's activation.

3. How figures are verified

Golden test benches replay hundreds of frozen cases against independently built reference calculations; published official worked examples serve as external anchors; and each calculator page shows a “rates verified on” stamp taken from the newest source date inside its own result. The full bench suite re-runs before any release.

4. Data statuses

Jurisdictions are labelled primary-verified, documented derivation, or compilation-checked, as defined in the disclaimer. A status is a property of the data, displayed on the page — never a marketing claim.

5. Updates and corrections

Parameters are re-verified on a monthly cycle and whenever an authority publishes a change mid-year (the final FUTA credit-reduction list each November is the standing example). A reported error is corrected in the data pack, locked in with a new golden case, and the correction is recorded in the project's development record.