UK SPI channel after #840: residuals, post-merge measurements and stale pointers #1288

Summary

#1256 merged on 14 October and closed #840. It delivered two things:

• the bridge predictors: main source, has pay, has trade, and at or over State Pension age;
• the twin-conditional draw.

#840 closed with its acceptance list unticked. Several residuals recorded in #1256's receipts (experiments/840-spi-draw-receipts.md) had #840 as their only home. This issue collects them.

The full local build on the merged head (R2 in the receipts note) is running. Its results will go on #840.

Weighted aggregates and shares only. Figures marked "design weights" come from the spine twins in the receipts note. Figures marked "calibrated" come from the first national built on 121fd4dd6. That build is not a matched twin: the build it is compared with, 21416b03e, also lacks #1242, #1260 and #1279.

#840's acceptance list

What the first calibrated national after #1256 shows

Final errors against target, 121fd4dd6, with 21416b03e in brackets:

• UC households by family: single with children −15.0% (−15.2%); couple with children +11.7% (+9.5%); one child −7.6% (−9.8%); two children −12.9% (−12.4%).
• UC spend (OBR) at initialization: −8.0% (−31.8%). Draw SPI-channel incomes at the FRS twin's conditional rank (#840) #1256 changed who carries the UC mass. The calibrated with-children cells barely moved.
• Childless cells at initialization: couples without children +99% (+53%); singles without children +14.5% (−11.6%). Both fit after calibration, which gets there by down-weighting.
• Bands:
  • State Pension £20–30k: +12.8% (+11.0%).
  • State Pension £30–40k: +13.3% (+7.5%).
  • Self-employment £20–30k: +10.1% (+0.3%).
  • All are inside the 25% fence.
• Top-1% weight share: 0.283 (0.280).
• Certification: the release cut is blocked by uk_qrf_tail_concentration on hmrc_spi_incapacity_benefit_income, which passed before, and the hmrc_spi_other_income exclusion is now stale. Both belong to UK SPI stage 1 keeps 4–17% of the tape's donors; replace the removed QRF tail gate with a donor-fidelity check #1276.

To measure on R2 (post on #840)

• The childless UC cells at constituency and local-authority level (the children_0 buckets), and the award bands.
• Lone-parent UC carriers per seat: the number of seats under the floor, never a seat's value. Also the with-children ESS against the pre-Draw SPI-channel incomes at the FRS twin's conditional rank (#840) #1256 local candidate. UK local: rerun the 2025 Autumn Budget constituency analysis on the local candidate to find data gaps and per-seat support before Budget day #1241 covers the generic seat pass.
• The UC taper reform estimate. Align UK UC claim, child and period definitions #882 had −£35bn against −£17bn to −£23bn.
• The no-award SPI lone-parent mass, before and after (criterion 3).
• Housing Benefit reporters by channel, and the JSA and ESA rows. These stay on UK spine: housing benefit £2.7bn against OBR £12.1bn — pensioner and supported-accommodation HB missing after the UC migration #867 and UK spine: JSA claimants 2.8× the DWP caseload — legacy-benefit reporting after the UC migration #869.

Residuals with no other home

1. Lone-parent structural shortfall.
   • UK spine: composition-faithful SPI income draw — bridge predictors and rank-preserving replacement (with-children UC residual) #840 declared it out of scope: lone-parent families concealed inside larger FRS households.
   • With Draw SPI-channel incomes at the FRS twin's conditional rank (#840) #1256 in, it is the main candidate explanation for the lone-parent UC cell sitting at −15% after calibration.
   • First step: split the cell's shortfall into benefit-unit composition and receipt.
2. The tape's low-pay tail on earning copies.
   • The tape's pay is an annual tax-record amount. 22.5% of earning copies sit under £12,570, against 10.4% of their twins (design weights).
   • Childless units and couples with children pass the UC screen 17–37% more often than their twins.
   • The SPI channel's UC recipients per unit are 1.58× the FRS channel's for singles, and 1.35× for childless couples.
   • Calibration absorbs the national counts. The local cells are R2's.
   • Candidate fix: extend the no-record rule so twins at or below the personal allowance keep their own values. Measure it on twins.
   • Another candidate, which also bears on residual 8: a minimum-wage floor on drawn pay (branch uk-cc-spipay): comment.
3. Stage 2 on drawn copies.
   • Stage 2 sees no children and no disability.
   • Drawn copies report PIP at 0.035 against their twins' 0.025, and carers at 0.012 against 0.006.
   • Lone parents' copies report Child Benefit at 0.40 against 0.64. 5.7% of childless singles' copies get Child Benefit.
   • Candidate fix: restore these reports from the twin in spi_benefit_coherence. Watch the stage's disability-category assert and the tail-gate surface that the stage defines.
4. Pension Credit and council-tax-benefit undershares on copies. The Housing Benefit share is on UK spine: housing benefit £2.7bn against OBR £12.1bn — pensioner and supported-accommodation HB missing after the UC migration #867; these two have no home.
5. Undrawn copies rewritten downstream.
   • Undrawn copies are 18% of the domain's weight.
   • spi_benefit_coherence, the take-up stage, the UC reporter redraw and the housing shell rewrite them. Against their twins: UC 0.322 vs 0.351, benefits in own right 0.348 vs 0.459, council tax benefit 0.082 vs 0.123.
   • The UK spine: composition-faithful SPI income draw — bridge predictors and rank-preserving replacement (with-children UC residual) #840 plan held this exemption for review (step 6).
   • Candidate fix: skip benefit units and households with no drawn adult.
6. State-Pension-only twins are drawn against a tape that holds mostly taxpayers (sources.yaml, stage hmrc_spi_income_spine). There is no receipt of the lift yet. It may be behind the State Pension band movement above.
7. Investment and wealth on SPI rows.
   • At design weights: interest +32%, dividends +9%, property income +18%. The WAS wealth draws rise with them. SPI income tax is +8.5%.
   • The national band targets still fit, so this is distributional. It matters for savings, dividend, property and CGT reforms.
8. Hours.
   • The bridge conditions on main source, pay and trade, but not hours, and SPI copies keep their twin's hours.
   • policyengine-uk reads hours: for Income Support from 2.123.10, and for ESA and JSA from 2.129.1. So drawn pay and kept hours can disagree.
   • Moved from WS-E spine deferrals: known bugs and signed inconsistencies awaiting their increment #796.
9. Chain order. Put the analogue targets ahead of the identity-uniform leaves; pay has the weakest within-cell rank fidelity. The manifest's note still points at UK spine: composition-faithful SPI income draw — bridge predictors and rank-preserving replacement (with-children UC residual) #840.
10. High income × has children (HICBC). The vendored Child Benefit statistics hold opt-outs but no HICBC liability. Pick the target's definition before any Chronicle request.
11. Lower priority:
    • the INDUSTRY07 ↔ SIC bridge, which needs the Annex A coding check;
    • the stage-2 predictors (plan D6);
    • an engine test that has_pay copies draw positive pay on the synthetic tape.
12. SPI copy prior. Copies take equal prior weights within a stratum, so the channel carries the FRS respondent mix. Candidate on branch uk-cc-spiprior: comment.

Registers and pointers

• Signed differences for the columns Draw SPI-channel incomes at the FRS twin's conditional rank (#840) #1256 moved (criterion 7). spine_swap_signed_differences.json has no UK spine: composition-faithful SPI income draw — bridge predictors and rank-preserving replacement (with-children UC residual) #840 entry; the receipts note lists the inventory. It is not a certification input.
• Re-measure the spi-channel-qrf-incidence entry's savings-interest delta on a post-Draw SPI-channel incomes at the FRS twin's conditional rank (#840) #1256 spine (tools/verify_uk_spine_parity.py).
• Re-point text that names UK spine: composition-faithful SPI income draw — bridge predictors and rank-preserving replacement (with-children UC residual) #840 when it is next touched:
  • the chain-order note in uk/spec/sources.yaml; editing it moves source_stages.json and the coverage digests;
  • the obr.housing_benefit and dwp.jsa_claimants reasons in calibration_measure_exclusions.json; both entries expire 2026-12-08.

Not here

• The stage-1 donor pool, the QRF tail gate and the incapacity failure: UK SPI stage 1 keeps 4–17% of the tape's donors; replace the removed QRF tail gate with a donor-fidelity check #1276 and Re-arm the UK QRF tail-concentration gate on independent draws #1277.
• uk-data#529's LFS employee and self-employed levels, and uk-data#551: UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095.
• The employment-indicator diagnostic and the childless-couple bands: Align UK UC claim, child and period definitions #882.

Relates #840, #1256, #882, #867, #869, #1095, #796, #1276, #1241, #1131.