How we test
Every release runs 1358 automated tests, and the site is only published when all of them pass. This page is generated from the last run, on October 25, 2026.
Published examples
The strongest check on a tax calculator is the tax authority's own worked examples. Where the IRS or a state publishes sample calculations, wage-bracket tables or employer test cases, each one is a test: the engine must produce the same figure to the cent, or to the rounding the publication itself uses. These 465 tests cover:
| Area | Passed |
|---|---|
| Federal withholding (IRS Publication 15-T) | 8 of 8 |
| Federal annual tax and credits | 12 of 12 |
| Local income taxes | 77 of 77 |
| Home and money formulas | 17 of 17 |
State withholding
44 states and the District of Columbia's rules each have their own set of examples taken from the state's withholding guide. The other 7, Florida, Nevada, New Hampshire, South Dakota, Tennessee, Texas, Wyoming, have no income tax on wages to withhold; their payroll levies, where they have any, are tested with the federal and local suites.
| State | Passed |
|---|---|
| Alaska | 2 of 2 |
| Alabama | 9 of 9 |
| Arkansas | 7 of 7 |
| Arizona | 6 of 6 |
| California | 13 of 13 |
| Colorado | 7 of 7 |
| Connecticut | 8 of 8 |
| The District of Columbia | 6 of 6 |
| Delaware | 9 of 9 |
| Georgia | 6 of 6 |
| Hawaii | 6 of 6 |
| Iowa | 13 of 13 |
| Idaho | 7 of 7 |
| Illinois | 5 of 5 |
| Indiana | 7 of 7 |
| Kansas | 9 of 9 |
| Kentucky | 8 of 8 |
| Louisiana | 8 of 8 |
| Massachusetts | 11 of 11 |
| Maryland | 8 of 8 |
| Maine | 8 of 8 |
| Michigan | 5 of 5 |
| Minnesota | 9 of 9 |
| Missouri | 7 of 7 |
| Mississippi | 6 of 6 |
| Montana | 7 of 7 |
| North Carolina | 8 of 8 |
| North Dakota | 6 of 6 |
| Nebraska | 7 of 7 |
| New Jersey | 11 of 11 |
| New Mexico | 8 of 8 |
| New York | 15 of 15 |
| Ohio | 6 of 6 |
| Oklahoma | 8 of 8 |
| Oregon | 15 of 15 |
| Pennsylvania | 5 of 5 |
| Rhode Island | 6 of 6 |
| South Carolina | 8 of 8 |
| Utah | 12 of 12 |
| Virginia | 8 of 8 |
| Vermont | 8 of 8 |
| Washington | 7 of 7 |
| Wisconsin | 10 of 10 |
| West Virginia | 6 of 6 |
A second opinion: PolicyEngine-US
Published examples cover the common cases. To test the rest, 3000 randomly generated households, with different wages, filing statuses, children and deductions, are run through both our federal engine and PolicyEngine-US, an independent open-source model of the tax code. Income tax, the child tax credit and the earned income credit must agree within one dollar, or within the rounding the IRS tables themselves allow. Every remaining difference has to be explained and recorded before a release can ship. The last run passed 4 of 4 comparison checks.
Properties that must always hold
Some rules can be checked on thousands of random inputs at once: every line of a paycheck plus take-home pay adds up to gross pay to the cent; federal withholding never falls as salary rises; a loan's principal payments always add up to the loan; an extra mortgage payment never increases the total interest; a higher income never lowers the price you can afford. These property tests run on every release alongside the examples.
Checked against an independent calculator
Before launch we ran a set of paychecks covering every state and the District of Columbia through a widely used commercial payroll calculator and compared every line. Federal income tax, Social Security and Medicare matched in every case. Where a state figure differed, we traced the difference to its cause: a few cents of table rounding, a state that lets employers round to the whole dollar, or a rule we read from the state's own guide differently. Three things it turned up (Idaho's whole-dollar rounding, New Mexico's workers' compensation fee and Oregon's Workers' Benefit Fund) were confirmed with the state and added.
The pages themselves
The build also checks every page before it can be published: each figure on a page must come from the engine (no number is typed by hand), every page must meet WCAG 2.2 AA in an automated accessibility audit in both light and dark mode, and each page must stay within its size budget so it loads quickly on a phone.
What testing cannot promise
Tests prove the engine follows the rules as we read them. They cannot prove an employer's payroll system does the same, and payroll systems do differ in rounding and timing. If your paycheck and our figure disagree by more than a few dollars, please tell us: the corrections policy explains what happens next, and every fix is listed in the changelog.