Count Scottish Child Payment, Healthy Start and the Two Child Limit Payment in government spending

[MaxGhenis]

Fixes #2338

What and why

scottish_child_payment, healthy_start_vouchers and two_child_limit_payment were in HOUSEHOLD_BENEFIT_VARIABLES but not in GOV_SPENDING_VARIABLES. Households received them, but government spending didn't count them, so gov_balance overstated the public finances by the amount paid.

The issue named the first two. two_child_limit_payment had the same gap. It is zero at baseline, because the regulations were not laid, but it is paid when a reform turns it on (gov.social_security_scotland.two_child_limit_payment.in_effect).

This PR adds all three to GOV_SPENDING_VARIABLES. It also adds a code-health test so the gap can't reopen: every household benefit must be in government spending, or on an explicit, documented NOT_PAID_BY_GOVERNMENT list, which is empty today.

Invariants

1. Subset with exclusions. Every entry of HOUSEHOLD_BENEFIT_VARIABLES is in GOV_SPENDING_VARIABLES or in NOT_PAID_BY_GOVERNMENT with a reason (tests/code_health/test_household_benefits_in_government_spending.py). A third test there runs the checker on the old list and confirms it reports exactly the three.
2. Sum identity. gov_spending equals the sum of the distinct GOV_SPENDING_VARIABLES.
3. Balance identity. gov_balance = gov_tax − gov_spending.
4. Conservation, at baseline. household_benefits − NOT_PAID_BY_GOVERNMENT = gov_spending − (spending that is not a household benefit: pawhp, other_public_spending_budget_change). "At baseline" means gov.contrib.benefit_uprating is at zero: household_benefits applies that lever and gov_spending doesn't, which predates this PR (see follow-ups).
5. Metamorphic. Paying a household d more of any of the three payments raises household_benefits and gov_spending by d, lowers gov_balance by d, and leaves gov_tax unchanged.

Invariants 2-5 are Hypothesis properties in tests/test_gov_spending_properties.py. Power: on the old list, the code-health test, both YAML cases (tests/policy/baseline/gov/gov_spending.yaml) and both properties fail. On this branch they pass.

Impact

Real run, 2026, one Microsimulation each. Both runs use the certified enhanced FRS: uk-data 1.58.0, enhanced_frs_2024_25.h5, HF revision 47bbb58d, sha256 5ba399ea…d960b6, verified at load. Main is 3f174d0; this branch is b0e8fb1, a clean tree. The runs went through the heavy lock, one at a time. Script and outputs: ~/reviews/uk-statutory-pay-gov-spending-2026-10-12/ (impact.py, compare.py, runs/).

2026, £bn	Main	This branch	Change
gov_spending	552.208	552.782	+0.574
gov_balance	469.236	468.662	−0.574
universal_credit	83.589	83.589	0
household_benefits	552.579	552.579	0
household_net_income	1,710.908	1,710.908	0
gov_tax	1,021.444	1,021.444	0

• What the +£0.574bn is. The Scottish Child Payment is £0.520bn and Healthy Start £0.054bn. The Two Child Limit Payment is nil at baseline. In every household, the change in gov_spending equals the sum of the three payments, to within half a penny. This re-measures the issue's ~£0.57bn.
• Who it reaches. 394 sample households (386k weighted) see gov_balance fall. No household's income changes.

Tests run

• New: code-health 3 passed; gov_spending.yaml 2 passed; properties 2 passed (45 s). test_aggregation_lists.py passed.
• CI at this head: all six checks pass. That includes Test, which runs the microsimulation reform-impact fixtures with the HF token.

Review

GPT-6.1 Sol (subfleet run --task review --tier hard) approved head b0e8fb1 with no blocking findings. The review is at ~/reviews/uk-statutory-pay-gov-spending-2026-10-12/review/out_2366.md. It checked the following:

• All three payments are government-funded.
• None is double counted inside another spending line.
• The reform-impact fixtures move by far less than their tolerances. Reforms measure reform − baseline, so the higher baseline spending cancels.
• The interactions with Count the Pension Age Winter Heating Payment in household and HBAI income #2047 and Report each pull request's effect on national aggregates on the certified dataset #2335 are as described below.

What I did with its non-blocking notes:

• Invariant 4 only at baseline: qualified above, and a follow-up session is proposed to make gov_spending apply the uprating lever.
• Cross-test import: left, to keep the approved head. There is precedent in test_state_pension_credit_qualifying_age.py.
• Healthy Start in Scotland is Best Start Foods: noted. It is government-funded either way.

Interactions

• Count the Pension Age Winter Heating Payment in household and HBAI income #2047 (pawhp into household income) is compatible in either order. Its exact-set test still holds, because the three variables here were already household benefits.
• Report each pull request's effect on national aggregates on the certified dataset #2335 (W3's fiscal partition) must move the three heads to totals: [gov_spending, household_benefits] and reword the household_benefits description; see the comment on #2335.
• Count statutory sick, maternity and paternity pay as Universal Credit earned income #2365 also adds statutory_paternity_pay to both lists. The hunks differ, and this PR's invariant passes with it.

Ownership: no W1 or W3 file is touched.

axiom: n/a: government-spending accounting lists for microsimulation aggregates, not a policy rule.

🤖 Generated with Claude Code

Reproducer: policyengine_uk/tests/code_health/test_household_benefits_in_government_spending.py::test_every_household_benefit_is_government_spending_or_excluded, policyengine_uk/tests/code_health/test_household_benefits_in_government_spending.py::test_exclusions_are_household_benefits_outside_government_spending, policyengine_uk/tests/code_health/test_household_benefits_in_government_spending.py::test_the_check_reports_a_benefit_left_out_of_government_spending, policyengine_uk/tests/policy/baseline/gov/gov_spending.yaml::Scottish Child Payment and Healthy Start vouchers are government spending, policyengine_uk/tests/policy/baseline/gov/gov_spending.yaml::The Two Child Limit Payment is government spending when a reform turns it on, policyengine_uk/tests/test_gov_spending_properties.py::test_government_spending_counts_every_benefit_government_pays

[MaxGhenis]

Hand-off to the UK hub from session local_615fdd79-8fa2-4544-b8ce-67188d179ad6 (started by the UK hub).

• Head: b0e8fb1.
• Review: GPT-6.1 Sol (subfleet run --task review --tier hard, run 20261011-210626-review-pr2366) approved this exact head with no blocking findings. The review is at ~/reviews/uk-statutory-pay-gov-spending-2026-10-12/review/out_2366.md.
• Checks: all six pass at this head (Lint, Test 1h08m, 4× install and smoke-import). GitHub reports it MERGEABLE. It's behind main, but nothing that landed overnight breaks its code-health invariant: I checked main's lists with this PR's three additions applied.
• Impact: real 2026 runs on certified uk-data 1.58.0. gov_spending +£0.574bn, gov_balance −£0.574bn, with no change to any household's income. Details are in the PR body.
• Depends on: nothing.
• Order notes:
  • It is compatible with Count the Pension Age Winter Heating Payment in household and HBAI income #2047 and Count statutory sick, maternity and paternity pay as Universal Credit earned income #2365 in either order.
  • Report each pull request's effect on national aggregates on the certified dataset #2335 needs partition edits once this lands (comment).
• Decisions: none for this PR.
• Follow-ups owed: a proposal card to make gov_spending apply the gov.contrib.benefit_uprating lever, which is pre-existing and changes no baseline number. It's offered to Max as a session card, not started.

[MaxGhenis]

UK hub (2026-10-12): this PR's head b0e8fb179 is merged unchanged into rollup B, #2380, with #2365, #2366 and #2375. Its one CI run tests all three together on current main. Merging #2380 with a merge commit marks this PR merged. The hub cancelled this PR's own queued runs, which the rollup replaces, to free slots in the starved Actions queue. Please don't push to this branch now: a new commit would not be in the rollup, and this PR would stay open.