UK wave 2: housing base facts — PIPR bulletin rents and EHS weekly housing costs (#133)

[juaristi22]

Closes the last unported wave-2 national family on #133: the publisher base facts behind uk-data's three housing/* targets. All three uk-data targets are rate×count products and stay excluded (computed) per the facts-only ADR — what lands here are the publisher rates they multiply, plus the honest paper trail for the factors that have no source.

Packages (2 packages, 309 facts, registry-only)

ons-pipr-private-rent-march-2026 (5 facts, html_tables_and_text)
The archived March-2026 "Private rent and house prices, UK" bulletin — the exact citation in uk-data's housing.py. Average monthly private rent: UK £1,374, England £1,430, Wales £828, Scotland £1,022 at month:2026-02, and Northern Ireland £875 as its own record set at month:2025-12 — NI's series lags the GB collection and the bulletin itself keys NI to December, so it is period-keyed truthfully rather than dropped or mis-keyed. February 2026 is a provisional PIPR estimate (bulletin wording) subject to routine revision; it is an observation of an elapsed period, so assertion stays at the default.

Why the bulletin and not the dataset: the monthly price-statistics workbook (25-March-2026 edition) is a 47,800-row × 40-column table whose xlsx_used_range parse materializes ~1.9M source cells on every bundle build (>2 minutes, gigabytes of RSS — measured before authoring). Regional and property-size averages are therefore a named exclusion pending a row-filtered workbook lane; LA grain is wave-3 scope (#134). All five quoted values were cross-checked cell-by-cell against that workbook's Table 1 before authoring.

mhclg-ehs-weekly-housing-costs-2023-24 (304 facts, ods_used_range)
Annex Table 1.7 (mean and median weekly housing costs, £/week) of the EHS 2023-24 rented-sectors Chapter 1 annex — uk-data's citation for social rent. Whole-table port: six characteristic blocks (age of HRP, economic status of HRP, region, weekly income quintile, receipt of housing support, dependent children) plus the all-households row, × five tenure columns × {mean, median}. New mhclg source. The all-households row carries the three headline anchors:

tenure	mean £/week	uk-data's figure
all social renters	118.12	£118/week (rent_social) — exact source
private renters	236.59	— (England PRS complement to the UK-level PIPR rate)
all mortgagors	222.42 (~£964/month)	see mortgage note below

Publisher markers are carried as record-set structure, not skipped cells: the full-time-education row's mortgagor cells are u (disclosure control, note 1) and the housing-support rows have no mortgagor columns (mortgagors are ineligible, note 3), so those two blocks declare the eight non-mortgagor measures only. Known quirks kept as-is and guarded: mortgagor costs include nominal £0.02/week values (note 5) — that is why the 75-or-over mortgagor median is 0.023; sample-size columns are survey metadata (named exclusion).

What deliberately does not land, and why

• The tenure counts (7.5m mortgagors / 5.4m private renters / 5.0m social renters). uk-data marks all three ~ UK-wide with no table citation; EHS is England-only, so no single publisher table carries them. Constructing UK-wide counts means composing EHS headline AT1.1 with three devolved housing surveys plus an adjudicated composition rule — its own curation with visible adjudications, tracked on the checklist, not smuggled in here.
• The £1,100/month mortgage payment has no source. uk-data cites the ONS rent bulletin for it, which carries no mortgage figure (mismatched citation). Nothing to port; the nearest sourced anchor is AT1.7's all-mortgagors mean (£222.42/week ≈ £964/month), which lands here. The populace-side total_mortgage declaration will need either a new adjudicated source (e.g., UK Finance data, if accessible) or to be rebuilt on the EHS mean.

Verification

• Both packages' 309 built facts re-derived independently against the lane parsers' canonical grids (the EHS ODS grid and the bulletin's document_numbers rows); every value matches, including full-precision floats behind the published roundings (118.12004… → 118).
• Guard cells pin block headers, unit annotations (£ per week (mean)/(median)), the table title, the source footer, publisher suppression markers, and — for the bulletin — element index, element type, and full paragraph text per fact.
• docs/pe-uk-source-checklist.md consolidated into one full-inventory table (the per-wave "Ported" sections were porting-sequence legacy): every national family in a single table with status and PR columns, and source-not-obtainable families as first-class blocked rows (SPP Review, the mortgage payment, the UK-wide tenure counts) instead of notes bundled into other rows. The previously garbled Stat-Xplore heading is gone with the sections.
• Credentialed chronicle publish-raw run for both artifacts; manifests commit real storage: blocks (R2 round-trip verified).
• Full bundle-expectation surface re-measured from one build; full local suite green before push.

🤖 Generated with Claude Code

[juaristi22]

Restacked on the post-#155 main and re-measured: 44,867 facts / 96 packages / 37 sources / 116 periods (= #155's 44,558 + this branch's 309), full local suite green before push.

This head also replaces the wave-era checklist layout: docs/pe-uk-source-checklist.md now carries one full-inventory table for every national target family, with explicit status and PR columns instead of per-wave "Ported" sections. Families whose publisher source cannot be obtained are first-class rows marked blocked — SPP Review NI-relief bases (source lost), the £1,100/month mortgage payment (mismatched citation, no source), and the UK-wide tenure counts (no pinned publisher table) — rather than being bundled into other rows' notes.

[vahid-ahmadi]

The two things I'd most want from a wave-closing PR are both here: the counts reconcile, and the things that couldn't be sourced are written down as first-class blocked rows rather than quietly omitted.

Arithmetic checks out — 44,558 → 44,867 is +309, splitting exactly as mhclg 304 + ons +5 (5,112 → 5,117), with source_count 36 → 37 for the new mhclg source and source_package_count 94 → 96. Geographies unchanged at 1,068 is consistent with PIPR landing on countries that already exist.

The mortgage-citation finding deserves to outlive this PR. uk-data cites the ONS rent bulletin for a £1,100/month mortgage payment, and the bulletin carries no mortgage figure at all. Refusing to port a number that has no source, and naming the nearest sourced anchor (AT1.7 all-mortgagors mean, £222.42/week ≈ £964/month) instead of quietly substituting it, is exactly right. But right now that discovery lives in a PR body, and PR bodies stop being read the moment they merge. It's a live mis-citation in a downstream repo — worth an issue on uk-data so total_mortgage gets rebuilt on a real source rather than inheriting a citation that doesn't support it.

Two things I'd change or think about.

1. The £0.02/week artifacts are carried as ordinary facts, with the explanation only in a YAML comment.

The header comment says it plainly:

mortgagor costs include nominal GBP 0.02/week values (note 5), which is why the 75-or-over mortgagor median is 0.023

But that comment lives in the source package, which the consumer of the bundle never sees. Downstream, median weekly housing cost, mortgagors, HRP 75+ is 0.023 with nothing on the fact marking it a publisher artifact — and 2p/week is the kind of number that looks like a unit error, so the charitable reader treats it as a bug and the incurious one divides by it.

Contrast the handling of the u disclosure cells and the missing housing-support mortgagor columns: those become record-set splits, so the measures are simply absent. Absence is self-describing. A present-but-artifactual value is the case that needs the marker, and it's the one that doesn't have one. If there's a per-fact slot for a publisher caveat, note 5 belongs in it; if there isn't, that's worth knowing, because this won't be the last table with one.

Related: note 2's italic small-sample flags are dropped as "formatting-only". Italics are formatting, but small sample is a data property — worth confirming the flag isn't recoverable from a sample-size column you're already excluding as survey metadata.

2. Provisionality has no representation on the assertion axis.

Your reasoning on the February 2026 PIPR value is right as far as the axis goes: a provisional estimate of an elapsed period is an observation, not a projection, so assertion correctly stays at the default. But that leaves the axis unable to say the one thing that matters about it — ONS will revise this number, and after they do, the bundle carries a stale value indistinguishable from a final one.

#154 gave us observation vs source_projection. This is the first package to want a third state (or an orthogonal provisional flag), and the gap is structural rather than something to solve here. Given how well #157 worked out as a filed issue from a review, this feels like the same shape — especially since a consumer resolving "latest" against a revisable figure is exactly the silent-wrong-answer case the assertion axis was built to prevent.

Smaller

• Choosing the archived bulletin over the 47,800-row workbook is well justified with measured numbers (>2 minutes and gigabytes of RSS on every bundle build is not a tax worth paying for 5 facts), and cross-checking all five cell-by-cell against the workbook before authoring is the right diligence. Worth confirming the bulletin URL is a permanent archive link rather than a live path — the expected_value guards carry full prose ("This is £47 (3.5%) higher than 12 months ago"), which is a strong guard but fails on any re-publication that rewords.
• Keying NI to month:2025-12 because the bulletin itself does, rather than forcing it onto the GB month or dropping it, is the honest call and worth the comment it has.
• The checklist consolidation into one full-inventory table with blocked rows as first-class entries is a real improvement over per-wave "Ported" sections — those were only ever meaningful during the porting.
• This is the sixth wave PR to hand-edit the whole-bundle constants. #157 is still open; no action for this PR, just noting the cost is still being paid.