A number is worth what its source is worth

A rate is read from the authority that publishes it, not from a summary of it and not from another calculator. What gets stored is the number, the document it sits in, the clause inside that document, the day it was read, the sentence it was read from, and a hash of that sentence as it stood that day. A figure missing any of those is not incomplete data that renders anyway. It is data this site cannot compute with, so there is no path by which an unattributed number reaches a page.

The calculation then reads those rules through a lens that records every rule it touched. The citations on a page are generated from that record rather than written beside it, which means a page cannot cite a source the arithmetic did not use, and cannot use one it does not show you. That is a stronger promise than a list of references at the bottom, because a list can go stale while the arithmetic moves on and nothing about the page looks different.

Before any of it is published, the result is checked against an outcome somebody else published: the average rates actually taken from a single person at three points in that country’s own wage distribution. Agreeing with that is not a matter of copying the same table twice, because the schedules and the outcome are published separately and by different processes. A jurisdiction reproduces those rates within 0.5 of a percentage point at all three incomes or it has no calculator here.

Nothing about that is unusual engineering. It is unusual only in that most calculators do not do it, which is why so many of them quietly serve a figure from a tax year that ended eighteen months ago under a heading that still says current. The methodology page states the calculation order, the worked example and every check, and the source of record lists every document behind every figure.

27 jurisdictions have no calculator here. That is not a coverage gap being worked up to quietly: it is the same check, applied to the countries where it fails, and 24 of them failed by an amount that is printed next to their name. The amount is the useful part. It says how far out a page here would have been, which is a thing no site publishing all of them can tell you about any of them.

A jurisdiction whose rules were transcribed perfectly and assembled incorrectly produces a clean, plausible, wrong number. That is worse than an obvious error and much harder to notice, because nothing about the page looks uncertain. The reader has no way to tell it apart from a right one, and neither does a search engine. Refusing to publish it is the only response that leaves the reader better off than guessing.

The list is recomputed on every build, so a jurisdiction can leave it as well as join it, and two have already lost pages they previously had. A refusal list that only ever grows is a list nobody is checking. Every jurisdiction is listed with its reason, and the ones that are computed carry the size of the agreement rather than a badge asserting it.

It does not give advice and it does not model your situation. Every figure assumes a single person on employment income with no dependants and no reliefs beyond those stated, because that is the only shape that can be computed identically across jurisdictions and compared honestly. Dependants, joint filing, pension arrangements, student loans, salary sacrifice and the timing of a move all change the answer, several of them by more than the gap between two countries usually is.

It does not model treaty relief, foreign tax credits, remittance rules, self-employment or two countries taxing the same income. It does not use a cost-of-living index, because none has been sourced to the standard everything else here is held to, and the city you live in moves that further than the tax system does. Employer contributions are left out everywhere: they never appear in your gross pay, so including them would move a country up or down for a reason that has nothing to do with what reaches your account.

Where employee social security is a real deduction and is not modelled, the page does not call its figure take-home pay. The label on a number is the claim being made, and a disclaimer underneath does not repair a label that overstates what was computed. Everything named here is a known gap rather than an oversight, and each is repeated on the pages it affects rather than only on this one.