Pay the legacy carer premium per qualifying claimant or partner; IS carer route through them only

[MaxGhenis]

Stacked on #1896 (remove-is-child-flags), which adds is_claimant_or_partner. Retarget to main once #1896 merges.

Summary

The legacy carer premium and the Income Support carer route counted every member of the benefit unit, so a caring child or qualifying young person qualified the family. The law names only the claimant and partner. Separately, two caring partners got one premium, because gov.dwp.carer_premium.couple has always held the single rate.

Carer premium. IS Regs 1987 Sch 2 para 14ZA(1):

the condition is that the claimant or his partner is, or both of them are, entitled to a carer's allowance under section 70 of the Contributions and Benefits Act or carer support payment.

Para 15(7):

Carer Premium. £48.15 in respect of each person who satisfied the condition specified in paragraph 14ZA.

The same condition and the same "in respect of each person" amount appear in:

• JSA Regs 1996 Sch 1 para 17 and Part IV (8);
• ESA Regs 2008 Sch 4 paras 8 and 11(3);
• HB Regs 2006 Sch 3 para 17 and Part 4 (8);
• HB (SPC) Regs 2006 Sch 3 paras 9 and 12(4);
• CTR England (SI 2012/2885) Sch 2 para 9 and Part 4 (4);
• CTR Wales (WSI 2013/3029) Sch 2 para 9 and Sch 7 para 14;
• CTR Scotland: SSI 2012/319 Sch 1 para 10. SSI 2021/249 Sch 1 para 5(3) pays "in respect of each partner if they both qualify for it, but only if they are not caring for the same severely disabled person".

Income Support carer route. SSCBA 1992 s.124(1)(e) makes a person entitled if "he falls within a prescribed category of person". IS Regs reg 4ZA(1) prescribes the people in Schedule 1B, and Sch 1B para 4 covers "A person (the carer)". A couple's claim "shall be made by whichever partner they agree should so claim" (Claims and Payments Regs 1987 reg 4(3)), so either partner's caring opens the route. A dependant's caring does not.

All text was fetched from legislation.gov.uk data.xml on 2026-10-01 (current revised versions, plus point-in-time copies of IS Sch 2 for 2015-2026).

Changes

• carer_premium: the number of members who are both is_claimant_or_partner and is_carer_for_benefits, times gov.dwp.carer_premium.single, times 52. That gives 0, one amount, or two. It used num_carers, which counts every member.
• gov.dwp.carer_premium.couple is removed. It has never been twice the single rate. Both files were created in 2021 at £37.50, and from 2023 each uprating copied the single value into couple. Two qualifying partners therefore got £48.15 a week instead of £96.30. single is relabelled as the per-person amount. Reforms should change single; a reform that sets couple will now fail loudly instead of silently doing nothing.
• gov.dwp.carer_premium.single 2025-26: £46.38 → £46.40. That is IS Sch 2 para 15(7) as in force from 7 April 2025. Every other year, 2015-2026, matches the point-in-time legislation.
• New benefit-unit variable partners_care_for_same_severely_disabled_person. SSCBA s.70(7ZA) says that where two people "would have a relevant entitlement for the same day in respect of the same severely disabled person, one of them only shall have that entitlement", and SSI 2023/302 reg 5(3) does the same for Carer Support Payment. So partners caring for the same person satisfy the premium condition once, in every scheme; SSI 2021/249 Sch 1 para 5(3)-(4) says so expressly for Scottish working-age CTR. When the variable is true, the premium is capped at one amount for the benefit unit.
  • Default: true, unless at least two of the claimant and partner report a Carer's Allowance award (carers_allowance_reported). It can be supplied.
  • Why: in law, two awards mean two different cared-for people. The model's carers_allowance and carer_support_payment, however, are paid on caring hours as well as on a reported award. So a modelled "receipt" does not show who is cared for, and a couple who both qualify on hours alone get one premium unless the variable is set to false. That is the same result as before this PR for that case.
  • The two allowances themselves still pay both partners (Carer's Allowance and Carer Support Payment pay two people caring for the same person (SSCBA s.70(7ZA), SSI 2023/302 reg 5(3)) #2028).
• income_support_eligible: the carer route now needs a claimant or partner who is a carer, not any member.
• num_carers and benunit_has_carer are unchanged. Their remaining reader on this branch is uc_carer_element (defined_for), which the UC claimant-elements branch stacked on Count only claimants' income in Universal Credit #1978 rewrites.
• In this model the premium reaches the Income Support, Housing Benefit and council tax reduction applicable amounts (benefits_premiums). Income-based JSA and income-related ESA come from reported awards.

Impact (Enhanced FRS 2024-25, real microsimulation runs)

There were four runs on a private copy of the Enhanced FRS 2024-25 file (sha256 e433e532…): this branch's base (#1896 head 143cc3c), then each step added in turn. Each step is a full run. Runs at 63b633c (same-person input defaulting to false) and at the head aa360aa (same-person default derived from reported awards) both match the step 3 run exactly in every year. The reason: all 6 dataset records with two caring partners have two reported awards, and no caring claimant or partner lacks one.

£m	2025	2026	2027	2028	2029	2030
Carer premium as computed for every benefit unit	+23.06	+22.73	+23.34	+23.90	+24.49	+25.09
1. claimant and partner only (para 14ZA(1))	+0.00	+0.00	+0.00	+0.00	+0.00	+0.00
2. one amount per qualifying person; £46.40 in 2025-26 (para 15(7))	+23.06	+22.73	+23.34	+23.90	+24.49	+25.09
3. IS carer route through claimant or partner only (Sch 1B para 4)	+0.00	+0.00	+0.00	+0.00	+0.00	+0.00
Income Support	+0.09	+0.00	+0.00	+0.00	+0.00	+0.00
Housing Benefit	+0.19	+0.20	+0.20	+0.21	+0.21	+0.22
Council tax reduction	+0.00	+0.00	+0.00	+0.00	+0.00	+0.00
Household net income	+0.29	+0.20	+0.20	+0.21	+0.21	+0.22

• Steps 1 and 3 are zero by data, not by accident. The Enhanced FRS has no care_hours (Take-up inputs for Carer's Allowance and the UC childcare element, and caring hours in is_carer_for_benefits #1856 explains why it is left unfilled); its only caring signal is reported Carer's Allowance. Every one of the 1,197 recipients is a claimant or partner. The 14 aged 16-19 are the claimant or partner of their own benefit units, and no child or qualifying young person reports CA, which is what the law predicts: CA excludes people in full-time education. The fix still matters for household calculations that enter caring hours or CA for a dependant.
• Step 2 raises the premium for the 9.1k benefit units (6 survey records) where both partners receive CA. Only 0.12k of those units are on IS, HB or CTR, so the benefit change is +£0.2m a year. Most of the +£23m premium total sits in units that receive none of those benefits. In 2025 the 2p a week rate correction also reaches 94.5k households on IS, HB or CTR.
• No household is worse off in any year, and Income Support eligibility changes for nobody. The model pays no Income Support from 2026.

Scripts and full comparisons are kept locally: dataset_impact.py, compare_impact.py, and the step JSONs. The comparisons are aggregates only, with no record-level values.

Tests

• tests/policy/baseline/finance/benefit/carer_premium.yaml (17 cases, hand-computed from the law). PE-UK reads gov parameters at 30 April of the simulation year, so 2025 uses the April 2025 rate. The cases:
  • single carer; partner only;
  • both partners with Carer's Allowance awards, at the 2026 rate (2 × £48.15 × 52) and the 2025 rate (2 × £46.40 × 52);
  • a caring QYP gives nothing; a caring child under 16 gives nothing;
  • a lone-parent carer with a caring QYP gets one premium;
  • two caring partners plus a caring QYP get two. This replaces the Specify carer_premium logic for >2 carers #420 three-carer regression test;
  • a working-age couple who both care: benefits_premiums and the CTR applicable amount, (HB Sch 3 para 1(3)(b) £150.15 + 2 × £48.15) × 52;
  • a pension-age couple who both receive CA: the HB applicable amount, (HB (SPC) Sch 3 para 1(2)(b) £383.35 + 2 × £48.15) × 52;
  • both partners supplied as caring for the same person (one premium) or for different people (two);
  • both partners caring by hours alone are treated as caring for the same person (one premium);
  • three members supplied as claimant or partner (a polygamous marriage), all with awards: three premiums, or one if they care for the same person;
  • a reported award its holder would not claim (would_claim_carers_allowance: false) does not count towards two awards;
  • an 18-year-old 15 years younger than the other adult is a partner; 16 years younger is presumed a child (the property generator's boundary);
  • a 17-year-old partner supplied through is_claimant_or_partner counts. The inferred roles treat a 17-year-old as an HBAI child; that limit comes from Replace generic child and adult flags with each programme's legal definitions #1896.
• tests/policy/baseline/finance/benefit/family/income_support/income_support_carer_route.yaml (4 cases): single carer and partner carer are eligible. A lone parent whose only carer is a QYP, and a couple whose only carer is a 12-year-old, are not.
• care.yaml: three tests that fed num_carers straight into carer_premium are removed. Their cases are now in carer_premium.yaml, built from people.
• Against the base branch, 8 of the first 13 YAML cases (premium and IS) fail, each for the intended reason. The other 5 are cases this PR leaves unchanged.

Invariants (Hypothesis, tests/test_legacy_carer_premium_properties.py)

The properties are drawn over families of a claimant aged 18-85, an optional partner, and up to three children or QYPs, each with a random reported Carer's Allowance award and random caring hours. The generator never draws an adult under 20 who is 16 or more years younger than the other adult, because the model would presume them to be the other's child. That boundary (18 with 33) is pinned with @example, and 20,000 generator draws find no violation.

1. A dependant's caring never changes carer_premium or income_support_eligible.
2. carer_premium = (caring claimants and partners, capped at one when treated as caring for the same person) × per-person amount × 52, so it is bounded by 0 and twice the amount. The default same-person rule (fewer than two reported awards) is checked in the same test.
3. Caring by the claimant or partner never removes Income Support eligibility and never lowers the premium. Below pension age it always opens the IS route.
4. Supplying "same person" never raises the premium and caps it at one amount.

The first test (1 and 2) fails on the base branch. Hypothesis's minimal counterexamples were a couple who both receive CA, paid one premium, and a lone parent whose 16-year-old QYP receives CA, which gave the family a premium. Invariant 3 is a positive invariant that also held before this PR. Invariant 4 tests the new input.

Locally, because PE-UK CI runs only on PRs into main:

• at 812878e: full YAML suite 1,487 passed; pytest policyengine_uk/tests/ 2,016 passed, 45 skipped;
• at 63b633c: the benefit and gov YAML suites, 1,284 passed; the property tests; the parameter, metadata and variable pytest subset, 1,703 passed;
• at 9c7cde8: the benefit, gov and local-authority YAML suites (441 + 885 passed); the property tests (3 passed);
• at aa360aa: carer_premium.yaml and care.yaml (21 passed); the property tests (3 passed); ruff check clean;
• ruff format --check . clean.

Review

An independent review on a Subfleet lane (GPT-6.1 Sol) returned REQUEST CHANGES:

1. Two premiums overpay Scottish working-age CTR when both partners care for the same person. Fixed, and more generally: s.70(7ZA) applies the one-entitlement rule to every scheme (the new input above).
2. Other IS gates still read every member: any member over pension age, any member's ESA, any member's reported IS. This predates the PR and is outside it. Fixed in Apply the Income Support conditions to the claimant and partner only #2013, which is stacked on this PR.
3. Missing boundary cases (young partner, pension age, applicable amounts asserted directly). Added; the extra-adult case is now stated in the variable documentation.
4. Documentation equated is_carer_for_benefits with legal entitlement and over-generalised the CTR schedules. Reworded.
5. An earlier commit message said both property tests fail on base; only the first does. Corrected above.
6. Changelog fragments lacked the bullet prefix. Fixed.

Round 2 (same reviewer) confirmed the s.70(7ZA) argument, the hand-computed expectations and fixes 2, 4, 5 and 6. It returned REQUEST CHANGES on two new problems, both fixed in 9c7cde8:

• N1. The widened generator could draw an 18-19-year-old with a partner 16 or more years older, whom the model presumes to be a child; the role check would then fail on some seeds. The partner's age is now bounded in both directions.
• N2. The default's stated reason was "two partners who both receive CA cannot be caring for the same person". That is true in law but false in the model, where receipt follows entered hours. The default is now derived from reported awards, and the wording above is corrected.

Nits fixed in the same commit:

• the same-person cap applies to the whole benefit unit;
• the CSP same-person rule is cited (SSI 2023/302 reg 5(3));
• Welsh working-age CTR is named in the formula comment.

Round 3 (read-only): APPROVE WITH NITS. N1 and N2 resolved: the reviewer traced every branch of the role inference against the generator, and checked all 15 YAML cases by hand. Three low nits were fixed in aa360aa:

• the same-person default counts reported awards only among carers;
• the wording says "Carer's Allowance or Carer Support Payment award", because carers_allowance_reported carries both;
• both sides of the generator's age boundary are pinned in YAML.

The reviewer also noted that the model's 2026 IS personal allowances (£104.43 single 25+, £164.06 couple) differ from IS Sch 2 para 1 (£95.55, £150.15). That is why no case asserts the IS applicable amount. Those parameters are rewritten in #1925.

Downstream

PolicyEngine/impact-iran-war-living-standards lists gov.dwp.carer_premium.couple in CPI_UPRATED_BENEFIT_PARAMETERS. It pins policyengine[uk]==5.3.0, so it is unaffected until its next bundle bump, which should drop that entry.

axiom: TheAxiomFoundation/rulespec-uk#403 queued (no module encodes the carer premium condition or IS Sch 1B para 4; HB Sch 3, HB (SPC) Sch 3 and ESA Sch 4 are in the pinned corpus, IS Sch 2/1B and JSA Sch 1 need ingesting)

🤖 Generated with Claude Code

[MaxGhenis]

Composition note from #2078 (stacked on #2009), which makes the CTR means test assess the applicant and partner rather than the benefit unit's claimant and partner.

#2078 adds council_tax_reduction_premiums, which the CTR applicable amount now reads in place of benefits_premiums. For ordinary families it is benefits_premiums, so this PR's premium changes flow through unchanged. Where the household head applies alone (council_tax_reduction_head_applies_alone: the head is liable but is not the benefit unit's claimant or partner), it computes the head's own premiums at the single rates, mirroring the single-claimant rules on #2009's base.

A property test in #2078 (test_head_applying_alone_matches_her_own_benefit_unit) checks that a head who applies alone gets the same applicable amount as the same head entered as her own benefit unit. Whichever of the two PRs lands second should therefore carry this PR's rules into the head-alone path of council_tax_reduction_premiums. For example: pension-age schedules with no adult disability premium; severe disability premium conditions; the carer premium per qualifying claimant. That test fails until it does.