UK source packages: ONS WAS Round 8 household wealth distribution (total wealth, components, percentile aggregates)

A 2026-08-28 investigation of the microcosm UK wealth surface (PolicyEngine/microcosm#796 / #665 context) found that the national calibration is structurally blind to the household wealth distribution:

• The ledger carries zero WAS facts. The source census across all 408 compiled UK references is hmrc/dwp/ons(population + land)/obr/slc/scotgov/voa/isc — WAS exists in the build only as licensed donor microdata for the was_wealth QRF stage, never as facts.
• Only 3 of 361 bound targets touch any wealth column — the ONS national balance sheet land aggregates (ons.e44r, ons.e44n, ons.e43b) — all single national £ sums. Ten of the thirteen was_wealth columns have no target of any kind; there is no total-wealth, band, decile, or count-above-threshold target. (The US surface binds Fed Z.1 net worth; the UK has no analogue.)
• The spine's wealth distribution currently verifies well against the published Round 8 anchors purely by inheritance from the imputation — measured 2026-08-28 on the candidate-of-record: median £281,170 vs published £293,700; top-decile threshold £1,210,748 vs £1,200,500; top-1% share 11.1% vs 10%. Nothing in the target surface or the armed gate battery would notice if a future spine lost that shape (the household_qrf_tail grids are measured but unarmed — a declared follow-up on microcosm#796).

This issue requests the publications be ported so a wealth surface can exist downstream. Porting never activates: microcosm owns the active subset, and the intended first consumption is as fenced measurement facts, not calibration targets.

Publications to package

All from the ONS Household total wealth in Great Britain: April 2020 to March 2022 release (published 2025-01-24, WAS Round 8) and its associated datasets, plus one FOI ad hoc:

Batching per the wave-2 pattern: one publication table per package, ~5–7 packages ≈ 2–3 PRs. Exact package IDs to the ingestor lane's convention; vintage-suffixed so a future re-accredited round lands beside, not over, these.

Publisher semantics the packages must carry

• survey_instrument: Wealth and Assets Survey, Round 8 (per the README provenance rules for survey-derived aggregates).
• Accreditation suspended from Round 8 onward while ONS works on quality (declining response rates, granular-level estimate quality). This is a publisher statement about the fact, and any downstream promotion adjudication needs it on the record.
• Pension methodology changed between Round 7 and Round 8 — pension wealth figures are not comparable with earlier rounds (stated in FOI-2025-3228).
• Collection was telephone-only during COVID-19, with stated underrepresentation of renters (affects London estimates).
• Coverage is Great Britain, not UK (no Northern Ireland). Reference period April 2020 – March 2022.
• Values are net-of-liabilities four-component concepts. The spine's columns are gross and three-component (no physical wealth, no debt stock, pension conflated into corporate_wealth until microcosm#750) — that translation is a populace-side binding problem and must be recorded there, not resolved by editing the facts.

Acceptance

• Every package: manifest source_url + sha256 + R2 URI; cell-geometry guards; concept_evidence_url per measure; deterministic checks green; judge lanes as headed.
• No consumer-artifact obligation in this issue — these facts join a consumer artifact when populace declares consumption.

Out of scope

• Binding, activation, or target-profile membership (microcosm).
• The £10m+ top tail: WAS cannot see it, and its eventual fill is an administrative surface (HMRC inheritance tax / estates statistics) in the CGT-stage pattern — separate issue when a use case lands.
• Local-geography wealth splits (nothing below region exists in the publication).
• Historical rounds (standing backfill lane).

Start with the ingestor lane's wave-2 packaging pattern and the README provenance rules, then review how manifests, cell-geometry guards, concept evidence, and deterministic checks are represented. Package the listed Round 8 and FOI tables with source URLs, hashes, R2 URIs, publisher semantics, and vintage-specific IDs; done means all judge lanes are green without consumer artifacts.