Methodology

Source hierarchy

Figures are accepted from primary government publications only, in this order of preference:

1. The revenue procedure or statute itself, as published in the Internal Revenue Bulletin.
2. Official IRS forms, instructions, and publications.
3. Official agency releases, such as the Social Security Administration's contribution and benefit base.
4. State revenue and taxation department publications, for state-level figures.

Secondary sources — news coverage, tax software blogs, other calculator sites — are never used as the basis for a published figure. They are occasionally useful for finding out that a change happened; they are not evidence of what the change was.

Transcription

Rates and thresholds are transcribed by hand from the source document into a dedicated data file, one file per tax year and topic. No figure is carried over from a previous year, and none is interpolated or extrapolated.

Every data file records where its contents came from: the source name, a link, the effective date, the review date, and a status of official, estimated, or draft. The calculator reads that metadata and displays it, so the provenance strip you see on each page is generated from the same record the calculation uses. They cannot drift apart.

When a figure is withheld

If a figure cannot be confirmed against a primary source, it is not published. Concretely, that means one of three things happens:

• The field is omitted from the calculation and the results panel explains what was left out and which direction it would move your answer. The refundable child tax credit is handled this way.
• The input becomes user-supplied. Sales tax rates and state income tax rates work like this — you enter a rate you can verify, and the output is labelled as based on your figure.
• The whole calculator stays unpublished and is listed as in development. Several are, for exactly this reason.

What does not happen is a plausible-looking number appearing without a source behind it.

Testing

The calculation engine is tested independently of the interface. The suite covers zero and negative income, every filing status, each bracket edge and the values immediately either side of it, very large incomes, capital loss limits, the Social Security wage base, Additional Medicare Tax thresholds, and the reverse sales tax calculation.

The most valuable tests are the bracket fixtures. Because Rev. Proc. 2025-32 publishes each band as a cumulative amount plus a marginal rate, the expected values could be taken straight from the IRS tables rather than from our own output. An error in any bracket boundary, or rounding drift anywhere in the chain, fails the suite immediately.

A second suite checks the pages themselves: that every asset path resolves, every label points at a real control, every ARIA reference exists, structured data parses, no element id is duplicated, every field a calculator reads exists in its form, and no page carries a duplicated title, description, or canonical URL.

Two further suites run every calculator the way a visitor would. One presses "Load example" on all fifty pages, renders the results panel, and rejects any output containing NaN, Infinity, undefined, or an empty shell. The other types a chosen set of figures into each calculator and checks the number it displays against an expectation derived by hand from the published figures and the standard closed-form formulas — never from this codebase's own output, because a test that captured our answers would agree with any bug we have. Each derivation is written out above its test so it can be checked without running anything.

The same suites cover the cases that break calculators in practice: a zero interest rate, a cleared term, an empty field, a payment too small to cover the interest, a contribution above a statutory limit, a debt that never clears, and a rate solver with no root to find. In each case the requirement is the same — an honest statement of what is missing or impossible, never a confident $0.00.

How the tax year is managed

Every figure that depends on a year lives in one file for that year and topic — federal rates, payroll parameters, estate and gift exclusions, retirement plan limits — under assets/js/tax-data/. Nothing is written into a calculation function and nothing is written into a page.

That matters more than it sounds. A limit repeated across three calculators is a limit that will eventually be updated in two of them, and the site would then disagree with itself while looking equally confident in both places. One file, one figure, one review date.

The review date shown on a page is taken from the data it actually uses, and where a page uses more than one file the oldest date wins — a page is only as current as its stalest figure. Pages doing pure time-value-of-money arithmetic have no such data and carry the site-wide review date instead.

Statutory figures against planning assumptions

A growing number of calculators here need inputs that no government publishes: an investment return, a rate of house price appreciation, a withdrawal rate in retirement, a PMI premium, a future inflation rate. These are treated as a different class of thing from a statutory limit, and the distinction is made visible rather than left to the reader to infer.

• Statutory figures — contribution limits, tax brackets, wage bases — come from the data files above, carry a source and a review date, and are applied without asking.
• Planning assumptions are always editable, are labelled as assumptions in the field hint, and are echoed back in the results panel so no result is presented as though it rested on published data.
• Quoted figures — a PMI rate, the rate a lender charges with and without points, a state unemployment rate — are left blank and asked for, because the honest default is none.

Where a limit binds, the calculator says so explicitly rather than silently truncating: a contribution above the elective deferral limit is shown as the amount requested, the limit that applies, the excess, and the figure actually used.

Rounding and precision

All money in the tax engine is handled as integer cents internally. Values are converted once on the way in and once on the way out, so intermediate results never accumulate floating-point error across bracket boundaries. Rates are stored as decimals rather than percentages so no calculation divides by 100 midway through.

The finance engines work differently and deliberately. Interest accrual is a continuous quantity with no legislated rounding point, and forcing it to whole cents every period introduces a drift that compounds over 360 payments. Amortization schedules therefore round only when a row is emitted, and the final payment absorbs whatever residue is left so the balance lands exactly on zero — which is also what a lender does.

Where a published worksheet rounds at a particular step, we round at that step. Where it does not, we carry full precision to the end.

Who is responsible for what is here

Calc365 is written, built, and maintained by one small independent team, publishing under the name Calc365. There is no third-party content agency, no syndicated feed, and no reviewer whose name appears here without having done the review.

That last point is worth stating plainly, because sites in this category routinely do the opposite. It is common to see a photograph, a credential, and a "reviewed by" byline attached to a page the named person has never opened. We are not going to invent one. Until a named specialist with verifiable credentials genuinely reviews these calculators, the accountable party is the publisher — and the publisher is reachable, corrigible, and named on every page.

Every calculator therefore carries three things instead of a byline: the tax year it applies to, the primary document its figures come from, and the date those figures were last checked against that document. Those are verifiable by you, which a photograph is not.

Correcting an error

Errors in a tax calculator are not a matter of opinion, and we would rather hear about one from you than leave it in place. Write to [email protected] or use the contact page, and if you can, name the page, the figure, and the official document you are working from.

What happens next:

• A wrong rate, threshold, or limit. Checked against the primary source the same week. If it is wrong it is fixed, the review date moves, and the change is listed in the update log below.
• A wrong formula or a broken calculation. Treated as urgent. If it cannot be fixed immediately, the affected calculator says so on the page rather than continuing to hand out a figure we do not trust.
• A missing case. Some omissions are deliberate and already documented under each calculator’s limitations. If yours is not, it is either added or added to the limitations, and we will tell you which.
• A disagreement about interpretation. Where the law is genuinely ambiguous we say so on the page instead of picking a reading and presenting it as settled.

We do not quietly change a figure and leave the page looking as though it always said that.

Update log

Substantive changes to what the calculators compute or to how the site handles your data. Cosmetic changes, wording, and deployments are not listed, and do not move any date on any page.

• 29 August 2026 — shared links stopped carrying figures in the page address. A shared scenario used to travel in the query string, which reaches server logs, referrer headers, and the advertising request. It now travels in the URL fragment and is removed from the address bar before any third-party script runs. Links shared under the old scheme still work. See the privacy policy.
• 29 August 2026 — analytics events rebuilt. The old "calculator completed" event fired on every keystroke and made the completion rate meaningless. Events now distinguish a start, a first usable result, and a recalculation, and every payload is restricted to a fixed vocabulary that has no field a figure could travel in.
• 29 August 2026 — one source for every date. The sitemap, the structured data, and the review line shown on the page are now generated from a single record per page, and are separately meaningful: first published, last substantively changed, and figures last reviewed.
• 29 August 2026 — mobile calculator order corrected. An empty results card no longer sits between the page title and the first field on a phone.
• 1 August 2026 — seven no-income-tax state paycheck calculators added, bringing the published set to seventeen states.
• 31 July 2026 — ten state paycheck calculators added and canonical URLs moved to their extensionless form, with permanent redirects from the old addresses.
• 29 July 2026 — state income tax calculator published covering the states whose 2026 figures were confirmed against the state revenue department.
• 24 July 2026 — estate and gift tax calculators published against the 2026 unified exclusion.

How the site is paid for

Calc365 carries advertising, supplied by Google AdSense. That is the whole of the revenue: there are no affiliate links, no lead generation, no referral fees from lenders or tax preparers, no sponsored calculators, and nothing behind a payment.

The reason to say so on the methodology page rather than burying it is that the alternative business models in this category shape the arithmetic. A refinance calculator paid per lead has a reason to make refinancing look good. This one does not, and you can check: every formula is described on its own page, and every rate traces to a government document you can open.

Advertising is placed so that it never sits between an input field, the Calculate button, and the result. If you ever find an ad breaking a calculation flow, that is a bug — report it.

Review cycle

Each page shows the date its figures were last reviewed. Reviews happen when the IRS publishes annual inflation adjustments, when legislation changes a rate or threshold, when the SSA announces a new wage base, and whenever a reader reports a discrepancy.

If you believe a figure here is wrong, please tell us and include the official source you are working from. Corrections backed by a primary source are made promptly.