SaishutsuUK public spending, followed

Method

How the figures are made

Every number on this site is read from a published file, and every step from file to figure can be repeated.

What is counted as spending

HM Treasury's Public Expenditure Statistical Analyses (PESA) 2026, table 5.1: public sector expenditure on services by departmental group, outturn, for 2025-26. PESA's 2025-26 figures are provisional outturn. The groups are disjoint and add up to spending on services, which is the total used here. Total managed expenditure adds accounting adjustments, which are not payments to anyone and are left out.

What is counted as followed

  • Payment records. Every row of every "spending over £25,000" and "over £500" file collected, for payments dated in 2025-26, from central government departments, NHS England, NHS trusts and integrated care boards, English councils, the Scottish and Welsh Governments and one Welsh council. Each payment is put in the section of the body that paid it.
  • Benefits. The Department for Work and Pensions' and HMRC's benefit expenditure, counted by benefit, never by person. For 2025-26 these are the departments' forecast and budget figures, because their outturn is not published yet. Every headline is given with and without them.
  • Pay. Each body's staff costs, counted as followed to its staff as one total, because wages are never published person by person. For central government, HM Treasury's OSCAR database (2025-26, provisional outturn), body by body; for local government, PESA 2026 table 7.8, United Kingdom local government pay, which for 2025-26 is still HM Treasury's plan. Together they come to £314.0bn against PESA's £311.1bn of pay: OSCAR counts some pay, such as research staff, that PESA treats as capital. Each figure is kept as published. A few payments in spend files are also part of pay (to pension funds, staff-benefit schemes and payroll bureaux), so a small amount may be counted twice.
  • Debt interest. PESA's own figure for each section's public sector debt interest, counted as followed to the holders of the debt as one total. Holders are never published one by one, so this is as far as it can be followed.

The rule throughout: money is counted as followed as far as it can lawfully be followed, and the reason it stops there is stated. Nothing is estimated to get further.

A payment is identified when its payee is matched, by the model's rules or a person's review, to a registered company, charity or other organisation outside government. It is named only when the payee is named but not yet matched, and withheld when the publisher redacted the name or marked the payee as a private individual. Identified and withheld payments count as followed. Named-only payments do not: among them are payroll accounts at HMRC, NHS trusts, health boards and councils, which are public bodies. They are shown as the difference between "followed" and "at most". Payees a proposed, not yet reviewed, match places outside government are counted with the identified.

A payment to another public body is passed on, and is not counted here: PESA counts that money in the section of the body that receives and spends it. Payments whose payee a proposed, not yet reviewed, match names as a public body are held apart and not counted either way.

What is not added

  • Grants and contracts. A grant a department pays is already one of its payments; adding the grant register as well would count it twice.
  • Student loans. Lending is repaid, and is not part of total managed expenditure.
  • Rows that repeat an earlier row of the same sheet exactly ("suspected duplicates"). They are kept as published, reported on each page, and not counted.
  • Rows a publisher marked as duplicates or "exclude from publish".

Caveats

  • PESA's totals are net of income from sales; payment files are gross. A section's followed share can be overstated for that reason, and is shown as over 100% where the records exceed the total, rather than capped.
  • A payee named only may turn out to be a public body, which would move it out of the "at most" figure and into passed on; or an organisation outside government, which would move it into followed.
  • Payments are put in a year by their date. A few publishers' files carry months outside the year they are labelled with; those rows go to the year of their date.
  • Nothing is estimated, apportioned or extrapolated. A gap is only described with a figure where a published figure measures it.

Reproducing it

The model is Python and Postgres. Its raw files are pinned by sha256 in a manifest; the database can be rebuilt from them alone, and the figures here come from one export, exports/coverage.py, whose output is downloadable. The code is not published yet.