UK spine: preserve FRS benefit-unit capital for UC while retaining WAS household wealth

[juaristi22]

Problem

The UK spine currently replaces the FRS relationship between reported Universal Credit receipt and benefit-unit capital with a household-level proxy assembled from WAS-imputed wealth. That proxy is useful for general wealth analysis, but it is not the same measurement as the liquid, benefit-unit capital relevant to UC eligibility.

This creates internally contradictory records: a benefit unit can report UC in the FRS while the later WAS-imputed proxy assigns it more than the £16,000 UC capital limit. The model then removes UC support that the source survey directly observed.

This surfaced while evaluating the first microcosm_uk_2024 national candidate on the PR #812 stack. It is a more specific successor to the wealth-side work in #750 and the residual UC investigation in #809.

Evidence

We joined the raw 2024-25 FRS benunit.tab to spine-k by source benefit-unit ID and evaluated capital using the model's current UC proxy.

The join covered all 18,850 source benefit units, including 2,212 raw UC-reporting benefit units. Weighted results for those reporters were:

Measure	Above £16,000
FRS TOTCAPB4 benefit-unit capital	0.017 million (0.36%)
Current WAS-derived UC capital proxy	0.478 million (10.09%)
Direct FRS value at/below £16,000 but proxy above £16,000	0.475 million

Of the 0.475 million false-high cases, approximately 0.373 million are driven by allocated savings + corporate_wealth; approximately 0.102 million require property components to cross the limit.

On the full spine-k support:

• 0.893 million weighted benefit units with reported UC have would_claim_uc = false;
• 0.939 million reported-UC support is above the £16,000 proxy capital limit; and
• the union of those two blockers is 1.495 million.

These are aggregate, disclosure-safe measurements; no source records are disclosed.

This is not evidence that the WAS wealth imputation lacks a top tail. The candidate has a substantial and survey-faithful wealth top tail. The defect is measurement grain and source authority: household WAS wealth is being used where the FRS supplies a benefit-unit capital measure that preserves the observed UC relationship.

Why this happens

The stage order currently allows two independent breaks in the observed UC relationship:

1. The FRS stage carries reported UC receipt, but the later WAS stage populates savings, corporate_wealth, and property variables. uc_assessable_capital then allocates those household assets across benefit units as a residual proxy.
2. SPI support/fill can rewrite universal_credit_reported after the take-up anchor has been established, without refreshing the corresponding would_claim_uc value.

PolicyEngine-UK already has the intended model seam. uc_reported_capital was introduced in policyengine-uk#1572 as an optional benefit-unit reported-capital override. A nonnegative value is used directly; its default -1 means that no reported value is available and the household proxy should be used.

Microcosm and policyengine-uk-data do not currently populate this variable.

Proposed fix: preserve two measurements for two purposes

Do not replace WAS wealth and do not introduce selective calibration twins.

1. Carry the FRS TOTCAPB4 benefit-unit capital value through the compact FRS source stage with source/provenance bindings.
2. Materialize it as uc_reported_capital at benefit-unit grain wherever the source value is available.
3. Continue imputing WAS wealth into the existing general wealth variables (savings, corporate_wealth, property wealth, and the other wealth channels). Those remain authoritative for wealth-distribution and other asset analyses.
4. For SPI support rows, preserve or jointly impute the FRS capital field alongside the other FRS-only UC characteristics so a synthetic UC reporter receives a coherent capital/receipt combination.
5. Recompute or re-anchor would_claim_uc after the final stage that can alter universal_credit_reported.
6. Leave uc_reported_capital = -1 only where a reported/imputed benefit-unit value is genuinely unavailable. Do not use the sentinel to switch channels between cloned alternatives.

This uses the existing model contract directly:

• FRS benefit-unit capital governs UC eligibility and tariff income when present;
• WAS-imputed household wealth remains available for wealth and distributional analysis; and
• the existing household proxy remains the fallback for genuinely unobserved benefit-unit capital.

Explicitly out of scope

• Do not clone only records where FRS and WAS values disagree. Existing UK and US clone precedents select a complete domain or a seeded sample ex ante; disagreement-conditioned twins would be a new modeling convention.
• Do not put -1 on one twin and a reported value on another so calibration chooses the data channel. The sentinel is an availability contract, not a calibration routing variable.
• Do not suppress or cap the WAS top tail.

Acceptance criteria

• TOTCAPB4 is carried from the FRS source to a benefit-unit field with source-ID provenance and no household-grain reassignment.
• Native FRS rows with an available value export nonnegative uc_reported_capital; -1 is reserved for genuine absence.
• SPI support rows have a receipted rule for uc_reported_capital that preserves its joint relationship with reported UC and family composition.
• would_claim_uc is refreshed after the last write to universal_credit_reported, with a stage-order regression test.
• WAS wealth outputs and the existing wealth-tail gates remain unchanged within their declared stochastic tolerances.
• A same-source regression test confirms that FRS UC reporters are not made ineligible solely because the later WAS household proxy exceeds £16,000 when their direct FRS capital does not.
• A before/after developer build reports the three blocker aggregates above and the six signed UC/children calibration differences.
• The rebuilt national candidate is recalibrated and the T0-T6 evaluation battery is rerun before changing or retiring any signed difference.

References

• UK WAS stage: split private pension wealth out of corporate_wealth (uk-data#452 M2) #750 — split pension wealth from corporate wealth
• UC residual after the pension-wealth split: support 4.24m vs 6.76m — re-draw undershoot, the 0.55 take-up freeze, and composition shape #809 — earlier UC-support residual investigation
• WS-E spine deferrals: known bugs and signed inconsistencies awaiting their increment #796 — WS-E deferral and certification tracker
• PR Assemble the certified UK national candidate into a publishable release (#806) #812 — national publication tooling stack on which this was observed
• Add Universal Credit capital limits policyengine-uk#1572 — introduced UC capital limits and uc_reported_capital

[juaristi22]

Follow-up filed as #832. PR #829 resolves the measured capital and take-up blockers; #832 owns the remaining SPI receipt-income-family incoherence and the UC-specific child-count binding.