You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
UK national release: main_residence_value keeps the WAS-imputed tail, so the surcharge's top bands stay empty #1112
Narrowed to the national line on 16 October 2026 (María's ruling). #1270 fixed the top tail on the local line, and #1240 closed into #1278 with the rest of that work. This issue now covers only the national release. The earlier text described an older release and is replaced.
#1270 (merged 15 October) delivered #1240's fixes 1 to 5 and 8:
uk.full.property.value, a post-geography node of the dense graph (uk_runtime/property_localisation.py), gives each England-and-Wales owner clone its home value from its assigned constituency's uprated latest-sale distribution (HM Land Registry Price Paid), at the household's weighted rank among its region's owners. uk.full.property.band draws the council-tax band from the LSOA's VOA band mix at the same uniform.
high_value_home_support copies the top 5% of owner households in each region six ways, so more distinct households carry the top of the distribution.
microcosm.high_value_homes.over_1m, over_1_5m and over_2m bind Land Registry-derived stock counts at constituency and local-authority grain. Two report-only gates compare the surcharge bands with OBR Table 1.1, and the regional £2m-and-over shares with the derived stock.
What is left: the national release
The national line solves the bound spine (uk_runtime/graph_national.py). It has no geography assignment, so the two property nodes never run, and main_residence_value stays the WAS-imputed value after regional uprating (uk_runtime/was_wealth.py, uk_runtime/regional_uprating.py). The support copies are in the national file, but each keeps its source household's value.
Measured on Port the uk-data wave to policyengine-uk 2.132.6 (fifth #1095 PR) #1299's head (the national file read by policyengine-uk at 2026 values; aggregates only): no owner-occupied home in England falls in the surcharge's two top bands, £3.5m to £5m and over £5m. A national HVCTS estimate therefore rests on the lower bands alone, and a reform to the upper bands, or a new band, reads nothing there.
Options
Run a value step on the national line. The spine knows each household's region but not its constituency, so read the value off the region's uprated sale distribution at the same weighted rank, or draw a constituency within the region as the local line does.
Extend the WAS tail inside was_wealth for both lines.
State that HVCTS estimates come from the local release, and label the national release as missing the top surcharge bands.
Done when
The national release places owner-occupied homes in every surcharge band, checked as a diagnostic against OBR Table 1.1 (2026 valuation bands, no back-casting) and against the Land Registry regional £2m-and-over shares.
Receipts carry aggregates only: no maxima or minima, and no cell under 10 source households.
Confirmed the observed upper-tail limitation on the drill H5 (aa31bdf67c977927ea2b325567d1cf7a79d94381239bc79918a0a0fc9c9588af): the maximum stored main_residence_value is £2,821,389, there are 32 distinct values at £2m+, and 334 records share the maximum. Under PE-UK 2.122.2/core 3.32.17, the maximum reaches £3,030,959 in 2026. The England counts reproduce at approximately 40k in £2m–£2.5m, 106k in £2.5m–£3.5m and zero in both £3.5m+ bands.
The issue identifies real data work. Please qualify the proposed benchmarks and causal explanation before using them to choose a repair:
The OBR costing gives an overall non-owner-occupied share of about 40%, not a tenure-by-value-band table. Multiplying every band by 0.6 is an explicit approximation. Keep it as a sensitivity until owner counts by band are available; do not present those products as official calibration targets. Applying 0.6 to revenue additionally assumes the same band mix by tenure, so the approximate £0.36bn owner-only yield is not an independently observed benchmark.
Table 1.1 gives 2028–29 property counts in 2026 valuation bands already. Do not back-cast those bands again with a house-price index. Distinguish the valuation date from changes in the property stock, including new construction, between 2026 and 2028–29.
The claim that missing upper-tail mass necessarily piled into the next band remains a hypothesis. Inspect the donor distribution and draws before clipping and regional scaling, as step 1 proposes, before selecting a Pareto extension or another tail method. The observed £2.8m maximum is below the declared £8m support bound, so it does not identify the responsible stage by itself.
Add an explicit valuation-date contract for the stored variable. Regional HPI scaling and subsequent GDP-per-capita projection must not silently treat a value anchored to another date as 2024 prices.
The owner-liability correction in PE-UK #2164 is supported, but it does not add landlords' let properties, second homes or company-owned property. Keep the owner-occupier tail repair here, with separate data/model ownership for full taxable-property coverage. This audit did not validate an official £1.5m–£2m benchmark supporting the mock drill's 140k count.
The saved H5 was certified with UK 2.100.0/core 3.32.5. These newer-runtime aggregate checks diagnose the limitation; they do not certify a repaired release.
changed the title [-]UK: `main_residence_value` stops at about £2.8m, so there are no homes in the £3.5m+ surcharge bands and too many in £2.5m–£3.5m[/-][+]UK: main_residence_value stops at about £2.8m (about £3.0m in 2026), so there are no homes in the £3.5m+ surcharge bands[/+]on Oct 6, 2026
changed the title [-]UK: main_residence_value stops at about £2.8m (about £3.0m in 2026), so there are no homes in the £3.5m+ surcharge bands[/-][+]UK national release: main_residence_value keeps the WAS-imputed tail, so the surcharge's top bands stay empty[/+]on Oct 16, 2026
Update after #1308 (merged 19 October): option 1 is in, at region grain.
The spine stage regional_home_values gives every England-and-Wales owner household with a home value the value that its region's uprated Land Registry distribution holds at the household's weighted mid-rank (the region level of uk/high_value_home_value_knots.json). It moves property_wealth by the change. Scotland, Northern Ireland, renters and owners without a value keep theirs, the council-tax band is unchanged, and the dense line does not change.
Still open, so this issue stays open:
The knots' tails run up to each region's largest sale, on both lines (declared on María's ruling of 17 October). Fix both lines together.
The "Done when" check has not run yet: a report-only national diagnostic of the surcharge bands against OBR Table 1.1 and the regional £2m-and-over shares. It is planned for a follow-up PR, with the tails fix.
Activity
juaristi22 commentedon Oct 6, 2026
Confirmed the observed upper-tail limitation on the drill H5 (
aa31bdf67c977927ea2b325567d1cf7a79d94381239bc79918a0a0fc9c9588af): the maximum storedmain_residence_valueis £2,821,389, there are 32 distinct values at £2m+, and 334 records share the maximum. Under PE-UK 2.122.2/core 3.32.17, the maximum reaches £3,030,959 in 2026. The England counts reproduce at approximately 40k in £2m–£2.5m, 106k in £2.5m–£3.5m and zero in both £3.5m+ bands.The issue identifies real data work. Please qualify the proposed benchmarks and causal explanation before using them to choose a repair:
The owner-liability correction in PE-UK #2164 is supported, but it does not add landlords' let properties, second homes or company-owned property. Keep the owner-occupier tail repair here, with separate data/model ownership for full taxable-property coverage. This audit did not validate an official £1.5m–£2m benchmark supporting the mock drill's 140k count.
The saved H5 was certified with UK 2.100.0/core 3.32.5. These newer-runtime aggregate checks diagnose the limitation; they do not certify a repaired release.
[-]UK: `main_residence_value` stops at about £2.8m, so there are no homes in the £3.5m+ surcharge bands and too many in £2.5m–£3.5m[/-][+]UK: main_residence_value stops at about £2.8m (about £3.0m in 2026), so there are no homes in the £3.5m+ surcharge bands[/+][-]UK: main_residence_value stops at about £2.8m (about £3.0m in 2026), so there are no homes in the £3.5m+ surcharge bands[/-][+]UK national release: main_residence_value keeps the WAS-imputed tail, so the surcharge's top bands stay empty[/+]Give the national line the regions' Land Registry home values (#1112, #…
Read the vendored OBR rows in the high-value home benchmarks' OBR leg (…
Give the national line the regions' Land Registry home values (#1112, #…
Band the home-value receipt on source households (#1112, #1095)
Give the national line the regions' Land Registry home values (#1112, #…
Band the home-value receipt on source households (#1112, #1095)
Give the national line the regions' Land Registry home values (#1112, #…
Band the home-value receipt on source households (#1112, #1095)
juaristi22 commentedon Oct 20, 2026
Update after #1308 (merged 19 October): option 1 is in, at region grain.
regional_home_valuesgives every England-and-Wales owner household with a home value the value that its region's uprated Land Registry distribution holds at the household's weighted mid-rank (the region level ofuk/high_value_home_value_knots.json). It movesproperty_wealthby the change. Scotland, Northern Ireland, renters and owners without a value keep theirs, the council-tax band is unchanged, and the dense line does not change.