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

Public Ledger

Forecasts against what happened

175 forecasts recorded since 8 Oct 2026. None can be checked yet: their years have not ended, or the official figures are not out. The first outturn is due in Apr 2027.

Every forecast this site shows is recorded on the day it is shown and never changed. When the official figure for that year comes out, we check it: inside the range is a hit; anything else is a miss above or below, measured from the nearer edge.

None yethit rateNo forecast with a range has been checked yet.
0missed aboveThe outturn came in higher.
0missed belowThe outturn came in lower.
175waiting for the outturnThe next 6 are due in Apr 2027.

Earlier official forecasts, checked

Before this site recorded anything, the OBR and the ONS had forecast years that have now ended. We check those too, to show how close official forecasts come. They are theirs, not ours. They are single numbers, so they are not in the hit rate.

8 checked. The outturn came in higher in 3 and lower in 5; the biggest gap was 2.0% (deaths, year to mid-2025).

List of misses

Biggest first, by the size of the miss as a share of the forecast.

Outturn used: ONS MYEB-2026-10-01, ONS PSF-2026-09. Scores are worked out again when a new release comes in.

What is recorded

183 forecasts so far: 175 the site shows and 8 earlier official ones. Ours are listed first and labelled as ours; the rest are the OBR's and the ONS's.

Show every series (17)
  • Spending: Culture & sportEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • Spending: DefenceEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • Spending: EducationEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • Spending: HealthEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • Spending: Housing & environmentEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • Spending: Police, courts, prisonsEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • Spending: Running governmentEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • Spending: Social protectionEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • Spending: Transport & economyEstimatePublic Ledger2026-27 to 2030-31, single numbers5 waiting; next due Jul 2027
  • BorrowingSourcedOBR2025-26 to 2030-31, single numbers1 of 6 checked; next due Apr 2027
  • Debt interestSourcedOBR2025-26 to 2030-31, single numbers1 of 6 checked; next due Apr 2027
  • IncomeSourcedOBR2025-26 to 2030-31, single numbers1 of 6 checked; next due Apr 2027
  • Public debtSourcedOBR2025-26 to 2030-31, single numbers1 of 6 checked; next due Apr 2027
  • Public debt, % of GDPSourcedOBR2025-26 to 2030-31, single numbers1 of 6 checked; next due Apr 2027
  • Total spendingSourcedOBR2025-26 to 2030-31, single numbers1 of 6 checked; next due Apr 2027
  • BirthsSourcedONSyear to mid-2025 to year to mid-2075, ranges1 of 51 checked; next due Sep 2027
  • DeathsSourcedONSyear to mid-2025 to year to mid-2075, ranges1 of 51 checked; next due Sep 2027

How the check works

The official figures come from the ONS public finances (then the OBR's outturn), HM Treasury's spending statistics and the ONS population estimates. A job runs with every nightly data refresh; a score only changes when a new release does.

Ranges come from whoever made the forecast: the ONS's published variants on the population page, our own model where it makes one. Where a forecast is a single number, we show it as one and say how far off it was, rather than invent a range.

What is left out, and why

Promise costs and sandbox results are what-ifs: what a policy would cost, or what a change would do, on official costings. No official figure measures a what-if, so they are not scored. The OBR's tax-by-tax forecasts are left out too, because the OBR and the ONS draw some taxes differently, and a gap would be a difference of definition, not of forecast.

The data: /api/v1/forecasts (JSON) and CSV.

Questions

Does Public Ledger make forecasts?

Mostly it shows other people's: the OBR's forecasts of the public finances and the ONS's population projections. Our own are few and simple, such as splitting the OBR's total spending by function in years with no published split. Sandbox results and promise costs are what-ifs, not forecasts.

What counts as a hit?

A forecast with a range hits when the official outturn lands inside it, edges included. Otherwise it misses above or below, and the size of the miss is measured from the nearer edge of the range.

Why are the OBR's forecasts never hits?

The OBR publishes single numbers in the tables we use, so they can only hit by matching exactly. We do not make up a range for them; instead we show how far off each one was. The hit rate counts forecasts with a range only.

Why are promise costs not checked?

A cost on a promise card is a costing of a policy, not a forecast of an official figure. No outturn measures what a single promise cost, so there is nothing fair to score it against yet.

Can a score change?

Yes. Official outturn is revised, so every score is worked out again when a new release comes in, and the table names the release it used. The forecasts themselves never change once recorded.

How many forecasts are waiting?

175 are waiting for their outturn. The next are due in Apr 2027.