Methodology

Two layers

The fact table holds two parallel slices of the UK fiscal universe. They are modelled separately and never cross-reconciled.

Managed expenditure

PESA Chapter 1. Departmental totals by DEL/AME × Resource/Capital. Reconciles to Total Managed Expenditure (~£1.2tn/year). Covers 2020-21 outturn through 2028/29-plan (2029-30 for CDEL).

Identifiable expenditure

PESA Chapter 5.2 (UK sub-function totals) and Chapter 10.1–10.4 (country × sub-function). 125-odd sub-functions across four countries. Covers identifiable spending only — non-identifiable items (most defence, most debt interest, EU transactions) are excluded, so this layer does not reconcile to TME. Covers 2019-20 to 2024-25, outturn only.

Real terms

Every fact carries nominal and real values. The GDP deflator comes from HMT’s GDP deflators at market prices, and money GDPseries — ONS L8GG for outturn (1955-56 → 2024-25), OBR March 2026 EFO for forecast (2025-26 → 2030-31). The forecast’s deflator level is chain-computed from the published percentage change column; HMT publishes only the pct-change for forecast years.

Base year: 2024-25 = 100.

Reconciliation checks

Ten automated checks verify data integrity on every ingest. The strongest are the workbook-totals family, which re-parse published “Total …” rows directly from the source workbook and compare to the model’s aggregated sums — a self-check against data the loader deliberately skipped.

• loader-integrity — CSV → DB round-trip
• workbook-totals — PESA Ch 1 total rows vs DB sums
• ch10-country-totals — PESA Ch 10 country grand totals
• ch5-function-totals — PESA Ch 5.2 function subtotals
• identifiable-residual — Σ country ≤ UK total per sub-function
• real-terms-consistency — valueReal = valueNominal × deflator ratio
• outturn-plan-labelling, vintage-integrity, del-sign-sanity, year-coverage — housekeeping invariants
• scenario-rollup — per-scenario arithmetic consistency

Sources

All sources listed here are under the Open Government Licence v3.0 unless noted otherwise. Each source row is tracked with its sha256 so a workbook replacement is caught by the reconciliation layer.

Known limitations

• ‘Programme’ is not a canonical UK concept. PESA Ch 10 sub-functions are the closest available grain at country level. Some departments (notably MoD) don’t expose sub-function detail, and Health is reported under HMT’s own classification rather than COFOG.
• AME is largely non-discretionary at line grain. The state pension, debt interest, and BoE APF losses are demand- or rate-driven. Modelling them as edit targets can invite nonsense; treat as policy parameters instead.
• Machinery-of-government changes. PESA 2025 restates historical data under current department codes. Earlier vintages used different groupings. Until a second vintage is ingested, the model has nothing to bridge.
• Identifiable residual. For ~10% of (sub-function × year) pairs, identifiable country totals marginally exceed the UK total — a consequence of how HMT classifies certain cross-cutting lines. Flagged as WARN in reconciliation, not corrected.
• Plans not forecasts. PESA plan-year figures are departmental expenditure limits, not the OBR’s central forecasts. For forecast deflators we use the OBR EFO; for spending levels we use HMT plans.