How we build and check these calculators

How a calculator is built

Each calculator starts as a pure function in lib/finance — inputs in, numbers out, no hidden state and no network calls. The formula is written down before the interface is, and the interface is only a way to feed that function. When two tools need the same math (amortization, bracket stacking, FICA), they call the same function rather than reimplementing it, so a fix in one place fixes it everywhere.

Tax rates, brackets, deductions, wage bases, and contribution limits are never written into the code. They live in versioned data files stamped with an explicit tax year, so an annual update is a data change with a reviewable diff — not a code change that could quietly alter behavior.

Every calculator ships a public methodology page — 27 of them, counting the mortgage sub-tools — spelling out the exact formula, the step-by-step calculation, the assumptions, and the sources. See, for example, how the mortgage calculator, the income tax calculator, or the paycheck calculator works. If we can't show the work, we don't publish the tool.

How the math is checked

The finance functions are covered by unit tests that assert hand-computed values, not snapshots. That distinction matters: a snapshot test only proves the output hasn't changed, so it will happily lock in a wrong answer forever. Our tests state the arithmetic in the test itself, so a test can only pass if the code agrees with a number a person worked out from the published tables.

Three real examples from the suite:

• Federal, $100,000 single, 2026. The test asserts a $16,100 standard deduction, $83,900 of taxable income, and $13,170 of tax — with the bracket arithmetic written out in the test: 10% × $12,400 + 12% × $38,000 + 22% × $33,500.
• Ohio's non-marginal table. Ohio's 2026 schedule (HB 96) is not a pure marginal table — it reads "$332 plus 2.75% of the excess over $26,050". The test pins $150,000 single to $332 + 2.75% × $123,950 = $3,740.63, and separately pins income below $26,050 to exactly $0, so the fixed table base can never be applied to someone who doesn't owe it.
• Massachusetts' millionaire surtax. The test rebuilds the whole calculation — 5% up to the indexed threshold, 9% above it — against the 2026 threshold of $1,107,750, so an incorrect or un-indexed threshold fails the build rather than shipping.

There are also structural tests that a single wrong figure can't sneak past: bracket slices must sum to the reported total, the state dataset must contain exactly 50 states, and every no-income-tax state must return $0.

Two linters run alongside the tests. One scores every state page against a completeness checklist and hard-fails if copy from one calculator bleeds onto another — if capital-gains wording ever appeared on a wage page, that is a build failure, not a proofreading task. The other crawls the sitemap and asserts head-tag invariants: the canonical URL must match og:url, the social card image must match the page's share image, and the meta title must match the page title. That second linter exists because a real bug shipped: pages with their own generated share image inherited a stale card image from the layout, so the wrong picture appeared everywhere it was shared. We wrote a check instead of resolving to be more careful. The unit suite and a full TypeScript type-check are run before a change ships, and the production build fails outright on any type error.

Where the data comes from

Primary sources only. Federal figures come from IRS Revenue Procedures and Notices and from the Social Security Administration. State figures come from each state's own Department of Revenue, Comptroller, or Tax Commission — its published rate schedules, estimated-tax forms, withholding booklets, or the statute itself. Economic data comes from the U.S. Census Bureau and FRED, and mortgage rates from the Freddie Mac Primary Mortgage Market Survey.

Current as-of dates, read directly from the data files that power the calculators:

• Federal, tax year 2026 — last verified against the cited sources on July 22, 2026. Sources: IRS Rev. Proc. 2025-32 (2026 inflation adjustments); SSA: 2026 Social Security wage base; IRS: 2026 retirement plan limits (Notice 2025-67); IRS Topic 409: Capital gains.
• State income tax — the baseline dataset was verified on July 15, 2026, and each state additionally carries its own verification date and change note. 45 states are published on 2026 figures; 5 are deliberately held on 2025 because at least one figure we'd need for 2026 has not been published yet. Those pages say so, on the page, rather than presenting last year's number as this year's.
• Mortgage rates — 30-year fixed average as of July 16, 2026, from Freddie Mac Primary Mortgage Market Survey (via FRED MORTGAGE30US). The value is committed to the repository rather than fetched at build time, so the rate a page shows and the as-of date it prints can never disagree.
• Median household income — 2024 observation year, from U.S. Census Bureau via FRED (real median household income), in inflation-adjusted dollars.

The policy on aggregators. Compilations like the Tax Foundation's annual state tables are genuinely useful — as an index. We use them to find out which states changed and where to look. We do not use them as the source of record. Every figure that reaches a calculator was read off the government document, and the state's own publication wins any disagreement.

What we do when sources disagree

They disagree more often than you'd expect, and the failure mode is almost never a typo — it's a cutoff date. Here is what checking every state against its own publication actually caught for tax year 2026.

Legislation that post-dated the compilation (nine states). The widely-cited 2026 state table was published in February 2026 and therefore cannot contain anything enacted after it. It was stale or wrong for Georgia, Utah, South Carolina, Arkansas, West Virginia, Hawaii, Vermont, Maine, and Massachusetts. Concretely:

• Georgia cut its flat rate from 5.19% to 4.99% under HB 463, signed May 11, 2026 and retroactive to January 1 — three months after the compilation's cutoff.
• Arkansas cut its top rate from 3.9% to 3.7% in a May 2026 extraordinary session, also retroactive to January 1. West Virginia cut every rate by roughly 5% under SB 392. Both cuts are absent from the compilation.
• Maine added a new 2% surcharge on high incomes, retroactive to January 1 — not in the table at all. Utah cut its flat rate to 4.45%, its fifth consecutive annual cut.
• Massachusetts is the subtle one: the compilation still shows a $1,083,150 surtax threshold, which is the 2025 figure. The indexed 2026 threshold is $1,107,750, and it appears in the Massachusetts DOR's own estimated-tax forms. Using the stale number would have over-taxed a narrow band of income by four percentage points.
• Hawaii was wrong for a different reason: Act 46 staggers its two levers on separate schedules — the brackets widen in one set of tax years and the standard deduction rises in another — so for 2026 the deduction nearly doubles while the brackets don't move at all. A table that carries both forward together gets it wrong.

When a state contradicts itself. Oregon publishes withholding figures and return figures that don't match. We use the return figure, because our calculators estimate what you actually owe for the year, not what your employer withholds each pay period — and we say so on the methodology page rather than silently picking one. Maryland had two Comptroller documents that disagreed; we resolved it by verifying digit-for-digit against the Comptroller's own 2026 payment-voucher worksheet, the document a taxpayer is actually handed.

When we can't resolve it, we don't guess. That is why 5 states are still on 2025 data. Arizona's rate is confirmed for 2026 but its standard deduction isn't; North Dakota's rates are confirmed but its indexed thresholds aren't. Publishing a confirmed rate against an unconfirmed threshold would produce a number that looks current and is wrong. A page that says "these are 2025 tables, next update when the state publishes 2026" is less impressive and more honest.

What we don't model

These are real limits, and knowing them is part of using the results correctly.

• Local and city income taxes. We model federal and state only. If you work in a city with its own income tax or a county with a local surcharge, your actual burden is higher than we show.
• Credits. No federal or state credits are applied — not the child tax credit, not earned income credits, not state-specific credits. Credits reduce tax owed, so results are conservative for anyone who qualifies for them.
• Deduction phase-outs. We apply each state's standard deduction as a flat amount. Several states taper it: South Carolina's new deduction phases out as income rises, and Wisconsin's slides toward zero. For higher incomes in those states we may show slightly less tax than you owe.
• State disability and paid-leave payroll deductions. Several states levy SDI or paid family-leave contributions on wages. Those are not in the paycheck model.
• Itemized deductions. Calculations use the standard deduction. If you itemize — mortgage interest, large charitable giving, high state and local taxes — your taxable income differs from ours.

Each tool's methodology page lists the assumptions specific to that tool. If you find something modeled incorrectly rather than merely omitted, that is a bug and we want it: report a correction.

Independence

Tallivo is free and supported by advertising, plus affiliate partnerships on a small number of clearly-labeled tools. Three commitments follow from that, and they are structural rather than promises:

• Advertisers do not touch the math. The calculators are pure functions of your inputs and the published tax tables. No advertiser, partner, or affiliate is an input to any result, and no result is ordered, ranked, weighted, or rounded to favor one.
• Your financial inputs never leave your browser. There is no backend that could receive them, so they cannot be sold as leads, matched to an advertiser, or used to change what you are shown. See privacy.
• Affiliate placements are labeled and separate. Where one exists it sits outside the result, is marked as such, and is not presented as a recommendation the calculator produced.

Who maintains it

Tallivo is built and maintained by the Tallivo team. We don't put a name or a credential on a page as a substitute for showing our work, because a byline isn't checkable and the work is. What we offer instead is verifiability: the formula for every tool is published, the assumptions are listed, each figure is attributed to the government document it came from, and every page states its tax year and the date the data behind it was last verified.

Nothing here is personalized advice. Tallivo is not a licensed financial, tax, or legal advisor, and these are planning estimates — see the disclaimer. For anything with real money at stake, take the numbers to a professional who can look at your whole situation.

Think something here is wrong? Tell us — with a source, and we'll check it against the primary document. More about the project on about.