Alpha: an early version. Promise cards are checked by AI Journalist, our automated reviewer; human editor review comes next. Become an editor

Public Ledger

How the numbers are made

Every number on this site comes from one of 24 official sources, or is worked out from them, and says which. Official forecasts are shown as their makers publish them; our own model results are always ranges.

16sources read by our nightly data job
21data editions on file, each kept as published
2 Oct 2026the newest edition was published
8forecasts checked against what happened so far

Every number has a label

Hover over or focus a number anywhere on the site to see its label, how it was made and a link to its source.

  • SourcedTaken from an official release, linked.
  • EstimateDerived or scaled from official figures; the method is shown.
  • ModelledProduced by a model, ours or PolicyEngine's, and shown as a range.
  • To checkA working figure not yet checked against an official source or by an economist. Please don't quote it yet.
  • Balancing figureA residual that makes the statement balance. It must be broken down before launch.

Ranges, not single numbers

  • Every sandbox result shows low, central and high.
  • Tax levers use HMRC's costings per unit of change. They include how taxpayers respond, but not knock-on effects on the wider economy.
  • Capital gains tax works in steps. HMRC says its figures for it cannot be scaled, so the sandbox offers only the rises HMRC costs and uses HMRC's own figure for each: never a figure in between, and nothing beyond the largest.
  • Where a source gives a central figure only, the range is an editorial band around it, and the method note says so.
  • Effects on GDP use spending multipliers of 0.5 to 1.0 and tax multipliers of 0.3 to 0.3: wide on purpose, and labelled as rules of thumb.
  • Bank Rate is set by the Bank of England. The sandbox lets you move it, labelled as such.
  • Years after the latest outturn show the OBR's forecast as the OBR publishes it.
Who gains and loses: microsimulation

The “Who gains and loses” panel uses PolicyEngine UK, an open-source model of the UK tax and benefit system. It applies your tax and benefit changes to every household in a large survey that represents the country (the Family Resources Survey, reweighted by PolicyEngine), works out each household's income before and after, and adds them up by income decile and region.

This is a static microsimulation: everyone keeps the same job, hours and spending, so nobody changes their behaviour. PolicyEngine gives one figure for each result; we show it with a ±10% range, labelled Modelled, and name the model and data versions it came from.

Our quick estimate and PolicyEngine can differ. The quick estimate uses HMRC's costings, which include how taxpayers respond; PolicyEngine's is static and starts from survey data a year or two old, projected forward. When the two are more than 15% apart, the panel shows both and says why. Disagreement is information, not an error.

The example households in “People like me” are the same for everyone. Their earnings and spending are fixed: spending is the Office for National Statistics' average for a similar household (Family spending in the UK, April 2024 to March 2025), so changes to VAT and fuel duty reach them. PolicyEngine charges VAT at the standard rate on half of that spending and scales it up to match total VAT receipts, which counts VAT that businesses pass on in prices.

It runs on PolicyEngine's free public service, only when you ask. Only the scenario is sent; your salary, household and region stay on your device.

Who checks the cards

AI Journalist is our automated reviewer: it re-reads every card against its live sources (the quote word for word, dates, evidence, status, cost, neutral wording, legal risk). Human editors review next.

Whatever a review finds is fixed in public: the card shows a dated correction with the old and new value. Every party's cards get the same checks.

Become a volunteer editor

Checked against what happened

Every forecast the site shows is recorded on the day it is shown and never changed. When the official outturn arrives, it is scored: inside the range, or a miss above or below.

See forecasts against outturn · Get the data (JSON or CSV)

Changelog of data editions

Every edition of every source the data holds, newest first. A new edition arrives through the nightly data job as a reviewed change, and older editions stay on file. Latest data build: 9 Oct 2026, 11,623 numbers from 16 sources, with no errors in its checks.

Show all 21 editions

Also as data: /api/v1/vintages (JSON) and CSV. Engine T0-0.1.0.

Reusing our work

What Public Ledger writes, such as status notes, cost notes, summaries and these method pages, is licensed under Creative Commons Attribution 4.0: reuse it freely, crediting Public Ledger with a link to the page it comes from. Official figures keep their own licences, mostly the Open Government Licence v3.0; quotes from Parliament are under the Open Parliament Licence; other quotes are short extracts whose rights stay with the speaker. Every source is named next to its number and in the open data API.

The site’s code, its data pipeline and every promise file are open source on GitHub, under the GNU Affero General Public License v3.0: anyone may run, study and change it, and anyone who runs a changed copy as a website must publish their changes too. Every edit to a promise card is a commit you can read there.

Also cited

Announcements and records used for single figures, such as a costed measure or a contract.