Budget Galaxy aggregates UK public spending and tax data from the official primary sources. Every number traces back to a named file we committed with a SHA256 hash.

• HM Treasury OSCAR II: ~55 departments, executive agencies and bodies. The 2023 OSCAR edition was reprocessed to correct a £114B NHS artifact that would otherwise double-count health spending. Same n/a-trust artifact also removed for 2020-2022.
• NHS: trust-level operating cost breakdowns for 206 provider trusts (118 acute, 45 mental health, 18 community, 15 specialist, 10 ambulance), with workforce and cost-type metadata where published. NHS commissioner (ICB) and provider (trust) values are netted against each other to avoid double-counting.

• HMRC tax receipts 2005-2024 (annual bulletin): full breakdown by tax type — Income Tax, NI, VAT, Corporation Tax, fuel duty, stamp taxes, IHT, CGT, APD, IPT, duties and more.
• Income Tax + NI bands for rUK and Scotland, fiscal years 2017-18 through 2024-25, from HMRC's "Rates and thresholds for employers" and Scottish Government publications.
• VAT & indirect tax estimation: ONS "Effects of Taxes and Benefits" decile consumption shares.

• 401 English local authorities from MHCLG Revenue Outturn (2017/18-2024/25), plus 32 Scottish councils from gov.scot for council tax bands.
• 19 English councils with supplier-level spend parsed from their own published registers and classified into MHCLG service categories. An additional 7 police and fire authorities are covered as separate entities.
• Population data: ONS mid-2023 local authority estimates for per-capita analysis.

• OBR Historical Public Finances Database: Public Sector Net Borrowing (PSNB) and Net Debt (PSND) from 2000-01 through 2022-23, with nominal GDP for % calculations.

The full data pipeline is in the project repository. Every number on this site can be traced back to a specific line in a specific source file. See the council spend source manifest and the tax data source manifest for the three-layer audit trail: live landing pages, Wayback Machine snapshots, and SHA256-hashed raw files committed to git.

Public spending data has gaps, inconsistencies, and quirks that most transparency tools hide. Budget Galaxy documents them here because they matter for how you read the numbers.

When additional countries are enabled (US / Germany / France / Brazil are next on the roadmap), each will get its own gap-disclosure section below.

When you click any node in Explore & Compare or Budget Galaxy, you see a contextual "About this programme" panel. The content comes from four layers of increasing quality, documented here honestly because quality varies.

• Tier A — Hand-curated (4,199 entries · 15,245 source citations). Written across multiple sprints of research against official publishers — DWP Stat-Xplore, HMRC Annual Report, NHS England TAC, ONS, IFS, OBR, NAO, House of Commons Library briefings, NHS provider Trust ARAs, council ARAs, COFOG aggregates. Each entry has a description, beneficiary count, legal basis, 8–12 key stats, recent-change notes, and 4–6 source citations as {publisher, title, url} dicts. Caveat: the underlying research was prepared by LLM agents using their training data (through early 2026). Numbers are cross-checked against our own tree JSON by scripts/verify_enrichment.py. Coverage of the 8–12 stats / 4–6 sources / 3–6 related-links contract floor: 86% / 41% / 71% as of 2026-04-27 — the 59% below source floor are early-sprint entries pending re-audit.
• Tier B — Imported (2,225 nodes). Pre-existing enrichment content in the repo: program_enrichment.json (779 compact entries, schema year / description / beneficiaries) and enrichment_top50.json (651 rich entries, adding legal basis / notes / OSCAR category). These cover 25% of the tree. Caveat: most of this content was LLM-drafted at an earlier stage and not re-audited. The 2026-04-20 audit flagged 11 top-level departments with incorrect founding years (Home Office listed as founded in 2024 vs actual 1782, MoJ as 2024 vs 2007, etc.). The frontend now suppresses any Established: 2024 field coming from this layer as a sentinel for "unknown" — but other factual details in these 2,225 entries are not individually verified.
• Tier C — Procedural (~1,500 council + council-service nodes). Generated at render time from our own data: _top_suppliers metadata joined with ONS mid-2023 local-authority populations (345 England + Wales authorities) to show population, spend per resident, tier-average comparator, top supplier, supplier concentration, and top purposes. 69% of councils in the tree match an ONS population row; the other 31% are police & crime commissioners, fire authorities and combined authorities — they don't have population rows by design (they serve multiple councils' populations combined). Those still get Tier D.
• Tier D — Contextual fallback (every remaining node). For any node that doesn't match A/B/C, the frontend generates a card showing the ancestor chain ("part of X, part of Y"), the node's rank among siblings, and the neighbours immediately above and below it in size. Purely from the tree; zero external data needed. This guarantees every single node in the UK tree shows some context — no bare "MEH" card anywhere.

• Scotland, Wales and Northern Ireland councils get Tier D (not C) because ONS publishes England & Wales population data only; SG/WG/NISRA equivalents are not yet loaded.
• Combined Authorities (GLA, Greater Manchester, West Midlands etc.) get Tier C via a heuristic but the "tier comparator" line uses the wrong reference group. Cosmetic, not factual.
• Most Tier A source citations point at a publisher + title but not a clickable URL. The publications are all real and findable, but users can't one-click.
• The Tier B 2,225-entry layer has not been individually fact-checked. A user may encounter a date or figure that's out of sync. Please file an issue — the verifier script (scripts/verify_enrichment.py) is kept in the repo to make each flag quick to reproduce.

The verifier is reproducible: py scripts/verify_enrichment.py writes data/uk/_enrichment_audit.json with per-entry flags for tree-match, scope discrepancy, missing URLs, and suspicious founding years. It runs in under a second.

Whenever Budget Galaxy cites a number, we want you to be able to prove it came from where we claim it came from — even if the publisher reorganises their site or pulls the page down. Four redundant layers per citation:

1. Live source URL — the publisher's own page (gov.uk, nhs.uk, ONS, etc). Fragile: UK government reorganises URLs constantly (DWP routes changed in 2024, BEIS split into DESNZ/DSIT/DBT in Feb 2023).
2. Wayback Machine snapshot — web.archive.org/web/<ts>/<url>. Permanent HTML capture taken when we cited the page. Click "archive ↗" next to any source.
3. Archive.org file mirror — for raw data files (CSVs, PDFs, ZIPs), we upload a permanent copy to the Internet Archive collection budget-galaxy-uk-councils-2024 via their S3 API. Currently used for the 161 council spend datasets (~2.1 GB).
4. Captured-at timestamp — ISO 8601 timestamp on every archived citation, so the audit is dated and reproducible.

Coverage 2026-04-27: council spend lookup 161/166 (97%) all four layers. Budget tree enrichment 4,199 entries / 15,245 citations — layer 2 (Wayback) being archived now via scripts/archive_enrichment_sources.js (~1 hour run with authenticated IA S3 keys, ~100 req/min). Layers 1 + 4 already at 100% in the JSON. Layer 3 only relevant for raw data files (not URL-based publisher pages).

Budget Galaxy is built and maintained by an independent developer. The project is open and every number is traceable to an official source.

The project exists because public spending data is technically published but practically unreadable. Central government datasets are scattered across official portals, tax receipts are in monthly ODS bulletins, local government data is fragmented across 400+ councils, and nothing connects your specific tax bill to the services it funds. Budget Galaxy connects all of it for the UK; more countries are being onboarded with the same depth.

The data, the pipeline, and the full source are in the project repository.