Every figure on the budget page comes from two things: the 2026-27 rules, written down with their sources below, and a made-up population of households weighted to match published UK totals. Costings are checked against HMRC's own; the method, the checks, and where they fall short are written up in full in COSTING.md and POPULATION.md. The code is open source, under the MIT licence, on GitHub.
There's no list of every household in the UK, and the survey data that comes closest isn't open. So the simulator uses about 32,600 made-up households (39,600 families, in the benefits sense of a single adult or a couple and their children). Each is built to look like a kind of household the UK really has: ages, pay, pensions, self-employment, savings, rent, children, health conditions, which nation and region.
Each household then gets a weight: how many real households it stands for. The weights are adjusted, as little as possible, until the totals match about 800 published figures at once, from HMRC (who pays income tax and how much, by income and region), DWP (who gets Universal Credit), the Family Resources Survey's published tables (households by type, tenure and income), and the OBR. Published sources don't always agree with each other, so some can't be met exactly; the checks below show how close it gets. These are not real people.
Every household is run through the 2026-27 rules twice: as they are, and with your changes. The difference in tax paid and benefits received, times each household's weight, added up, is what the change raises or costs in a year. That's the "static" figure: it assumes nobody earns, claims or arranges their affairs any differently. Each change is also costed on its own; changes overlap (a tax rise reduces Universal Credit, which is worked out on pay after tax), so the parts don't quite add up to the whole, and the difference is shown as the interaction.
Winners and losers compare each household's income after tax and benefits, before and after, ranked by income adjusted for household size, the way DWP's official statistics do it. A change of less than £1 a week counts as no real change.
When a tax on extra income goes up, people tend to report less of it: they work a little less, pay more into pensions, take income in other forms or at other times. The second headline figure allows for that, using the Scottish Fiscal Commission's published estimates of how strongly taxable income responds, which rise with income (people on higher incomes have more ways to respond). The low and high figures use half and one and a half times those estimates, because nobody knows them precisely. Benefit changes and employer National Insurance are left static.
The starting point is the OBR's forecast published with Budget 2026 (28 October 2026). Your changes are assumed to start next April. Their cost is carried forward each year in line with the OBR's forecast for the tax or benefit it sits in, and the extra borrowing adds to debt and to interest on the debt, at the rates the OBR publishes. That's then checked against the two fiscal rules in the Charter for Budget Responsibility: the day-to-day budget in surplus in 2029-30, and debt falling as a share of the economy in 2029-30.
Official costings leave the economy as forecast, and so do the main figures here. A second, rougher line allows for the budget's effect on the size of the economy using the OBR's published fiscal multipliers, and what that does to tax and spending. It's shown separately, with a range, because those effects are much less certain.
Every costing gets a grade: A for an official costing by the OBR or HMRC, B for this kind of model of made-up households weighted to published totals, C for figures resting on general evidence from elsewhere, D for an informed guess. The page's own costings are B; the effect on the economy is C.
Each change also says how well that kind of change checks against published costings, mostly HMRC's "ready reckoner" of standard tax changes: close (within about 6% of HMRC), fair (within about 15%, or HMRC's figure inside our range), or known gap (a known difference; treat as rough). Removing a tax altogether is always a known gap: it's far beyond the small changes anyone publishes checks for.
Made-up households, not real people. About 39,000 synthetic families, weighted to match published UK totals for 2026-27. These are this year's rules applied to them, next to the official figures.
Loading the population…
Every rate and threshold the calculator uses, with where it comes from.