A tax calculator you can argue with
Most calculators give you a number and no way to check it. This one shows the bands, the rates, the authority and the sentence each figure was read from, so that when you disagree you can point at the line.
The claim is narrow on purpose. It is not that these figures are right. It is that every one of them names where it came from, and that anything we could not source is refused rather than estimated.
Nobody qualified has reviewed this site
Written by Usain Olivard. No qualified adviser has reviewed this site and it does not claim one.
What that means in practice: the arithmetic is checked against worked examples the tax authorities published themselves, and every rate is compared against a second body that compiled the same law. What it is not checked against is the judgement of somebody who files returns for a living. Those are different assurances and only one of them is on offer here.
7 of 995 rules have reached verified status. The rest have been read once, by machine, from the primary document. That is stated on every row of the source of record rather than averaged into a badge.
Who it is for
Weighing an offer abroad
A salary in another country is not comparable to one at home until tax and contributions are taken off both. Every jurisdiction here is computed on the same assumptions, which is what makes two figures comparable at all: numbers collected from several national calculators are each right and together meaningless.
Comparisons are made at the same position in each country’s own wage distribution rather than at the same converted salary, so the answer does not move with the exchange rate.
Checking a payslip
272 UK salaries have their own page showing income tax and National Insurance separately, band by band, at the exact figure rather than at a rounded neighbour. If your payslip disagrees, the page shows which line to argue about.
National Insurance runs on its own thresholds, not the income tax ones. Modelling it against the income tax thresholds is the most common error in take-home calculators and it is worth several hundred pounds a year.
Filing from abroad
10 guides cover the thresholds an American abroad files against, each led by the figure and the statute it comes from. None of them asks how much tax you owe. Each asks whether a rule applies to you, which is the question that actually decides a filing obligation.
Every figure in them is a rule in the same store the calculators read, so a guide cannot quietly quote a different number from the one a page computes.
What we are not
Four things this site cannot do, stated here rather than in a footer, because a limitation a reader has to go looking for is a limitation being hidden.
The output is an estimate for a straightforward case, built from published rules. It cannot see your reliefs, your other income, your residence history or the timing of a move, and any one of those can move a real answer further than the headline rate does.
Nothing here submits anything to anyone. Where a guide states a filing obligation it names the form and the statute, and stops there.
31 jurisdictions are refused a calculator, and the refusal is published with the size of the miss against each one rather than left as a gap in a list.
7 of 995 rules are verified, most of the store has been read once rather than twice, and the corrections log is short because the site is young rather than because it is careful.
How it stays free
Advertising, and in time affiliate links to services we would use ourselves. Neither will ever sit inside a source block or change a figure, and no advertiser sees anything you type, because nothing you type leaves your device to begin with.
There is no account, no email capture and no saved calculation. The rules ship with the page and the arithmetic runs in your browser, so there is no salary endpoint to log against. That is a consequence of the build rather than a promise about our conduct, which is the only kind of privacy claim worth making.
The whole ruleset is published under CC BY 4.0 and can be downloaded as one file. If this site disappeared tomorrow the figures behind it would still be checkable, which is not true of any calculator that keeps its rates in a database it does not show you.
The fastest thing you can send is the official clause you think we misread. It goes on the corrections log with the date, the page, the wrong figure printed in full and how it surfaced, whether or not it turns out we were wrong.
You do not need to be right, and you do not need to be polite about it. A reader who says a figure looks wrong at a particular salary and nothing else is still more useful than a check we did not run.
Ruleset sha256:fe993d9d072cbcac · oldest reading 06 Sept 2026 · 9 authorities
How a figure gets from a document to a page
The chain has five links and each one is recorded, because a figure whose history you cannot reconstruct is a figure you are taking on trust. This is what sits behind the number when a page shows you one.
An authority publishes a rate, a threshold or a schedule. That publication is the only acceptable origin: not a summary, not a news article, not another calculator. Where an authority offers both a machine-readable feed and a human page, the feed is read and the page is linked, because the feed is what can be checked again next year without a person reading it.
The exact text containing the figure is copied out and stored, along with a hash of it. Government pages move, get rewritten and quietly change. The hash is how a dead link stays checkable: it proves what the page said on the day it was read, which is the thing a disagreement actually turns on.
The figure becomes a record carrying its own tax year, the window it governs, the authority, the clause locator inside the document, the retrieval date and the archived sentence. Money is held as an integer count of the smallest unit, never as a decimal, because a decimal forces a tolerance on every test and a tolerance hides real disagreement.
The rule is compared against an independent compilation of the same law for the same year, and the arithmetic built on it is run against worked examples the authority published itself. Both have to agree before a rule can claim to be verified, and the second reader has already been right where we were wrong.
Only then can a page state the figure, and it states it by computing from the rule rather than by repeating a number typed beside it. That is why the counts, dates and statuses on this site change when the store changes: there is no second copy of any of them to fall out of step.
What we mean by verified
Three states, and the distance between them is the whole argument. A status here names how many independent readers have looked at a figure. It is not a grade we award ourselves for effort, and a figure cannot move up a level because somebody feels confident about it.
One reader, once
Read from the primary document by machine, with the sentence archived against a hash. Nothing has confirmed it independently. 988 of the 995 rules are here, and a calculator will still use them: the page carrying the result says so rather than implying a review nobody has run.
Two further things are true
A golden test checks the arithmetic against a worked example the authority published itself, and a separate body that compiled the same law agrees for the same tax year. Both, for the same rule, or the status does not change. 7 of 995 have cleared it.
Two readers, and they disagree
The document and the second compiler give different answers for the same year and nobody has adjudicated it. The figure stays on the page with the status attached rather than disappearing while it is worked out, and it ends as a correction or a recorded decision with a date against it.
The neutral grey on the first of those is deliberate. A pale green badge for “we read this once” is exactly the overclaim the whole system exists to avoid, and the brand green is a call to action here and never a success indicator.
Why we refuse countries
Published tax schedules do not always say what they mean. A surtax might apply to tax due, to taxable income or to gross earnings, and the series that publishes it often does not say which. Social security contributions might reduce the income tax base or sit outside it. A credit might be applied against tax or against income. Each of those is a fork, and taking the wrong branch produces a number that looks entirely reasonable and is wrong by several percentage points.
So every reading the data permits is built and computed, and each is measured against the average rates a separate body publishes for the same person at three different incomes. A reading has to land inside 0.5 of a percentage point at all three, not on average across them, because an average hides the income where the reading actually disagrees.
31 jurisdictions currently fail that test and get no calculator, however close they came. The worst misses by 16.19 percentage points. The closest is inside a single point and is still refused, because a threshold that bends for a near miss is not a threshold. Each refusal is published with the reading that was tried and the size of the gap, so the absence is a measurement rather than a silence.
Refusing is not free. It means a reader looking for one of those countries finds no calculator here, and goes somewhere that will give them a number without telling them how confident it is. That is a real cost and it is paid deliberately.
The partial answer is a reference page. Several refused jurisdictions have one: the statutory schedule as published, the outcome the OECD measures for the same person, and a plain statement that we do not compute it. It carries no figure of ours at all, which is precisely why it can exist where a calculator cannot.
A jurisdiction also loses its calculator the day it stops reconciling. That has already happened, and the misses are recorded rather than quietly dropped. A coverage list that only ever grows is a coverage list nobody is checking.
Rules the writing has to pass, not just the arithmetic
A number on a page must have been computed by the page
Prose cannot state a figure the engine did not produce. A worked example that drifts from the calculator beside it is the failure this rule exists to make impossible, and it has been caught happening.
A page states its own scope from the data
Coverage counts, reading dates and status tallies are read from the store at publish time. A page that describes itself in hand-written copy is the first thing to go stale, and on a site whose whole claim is provenance a stale claim is the specific failure to avoid.
Limits come before capabilities
Every calculator page names what it does not model, in the same weight of type as what it does. An assumption in a footnote under a result is an assumption designed not to be read.
A label is the claim
Calling a figure take-home pay when social security is not modelled is wrong even if a disclaimer sits underneath it. Where contributions are not modelled the page says income tax only, in the label, where the number is.
No figure without a source
Nothing reaches a published page except through a rule carrying its authority, clause, date and archived sentence. This is enforced at publish time rather than left to discipline.
A mistake is published in full
Corrections carry the wrong figure, the right one, the money between them and how it surfaced. A correction that hides what it corrected is a press release with a date on it.
Questions worth asking before you trust a calculator
Who writes this?
Usain Olivard writes it, and is not a qualified tax adviser, and no qualified adviser has reviewed the site. That is stated here rather than left for a reader to work out, because a calculator that lets you infer expertise it does not have is doing something worse than being wrong.
How do I know a figure is not made up?
Every one of the 995 rules behind this site carries the authority that published it, the clause inside the document, the day it was read, and the sentence it was read from with a hash of that text. All of it is on the source of record and downloadable as one file. No figure reaches a page except through a rule that carries all of those, which is enforced rather than intended.
Why do you refuse to compute so many countries?
31 jurisdictions are refused a calculator because no reading of the published schedule reproduces the rates the OECD publishes for the same person, to within 0.5 of a percentage point at three different incomes. The worst measured miss is 16.19 percentage points. Publishing those would mean guessing at something the data does not state, and a confident wrong number is worse than an absent one.
What does this cost me?
Nothing, and there is no account to make. It is funded by advertising and, in time, by affiliate links to services we would use ourselves. Neither will ever sit inside a source block or change a figure. If that stops being true, this answer changes before anything else does.
Does anything I type leave my device?
No. Every calculation runs in your browser against rules shipped with the page. There is no salary endpoint to call, so there is nothing to log, and that is a consequence of how the site is built rather than a promise about what we do with what we collect.
What happens when you get something wrong?
It goes on the corrections log with the date, the page, the wrong figure printed in full, the right one, and how it surfaced. 2 are logged. The wrong number stays visible. A correction that hides what it corrected is a press release.