Apply the income-related ESA and income-based JSA remunerative work conditions

[MaxGhenis]

Fixes #2043

Based on main, last merged at e89b8ab. The PRs this one built on have all merged:

• Apply the Income Support remunerative work and JSA conditions (s.124(1)(c), (f)) #2025 (4ed8713): the Income Support gate's readers of the screened awards;
• Count only the claimant's and partner's legacy awards in means tests and passports #2027 (bbb9576): the claimant-or-partner awards and their readers;
• Fix Class 2 and Class 4 NICs against SSCBA 1992 for 2015-16 to 2026-27 #2058 (cfee0e6): the Class 2 and Class 4 amounts corrected to SSCBA 1992.

This PR now changes no National Insurance parameter or test.

What the law says

Income-related ESA. Welfare Reform Act 2007 Sch 1 para 6(1): the claimant "(e) is not engaged in remunerative work; (f) is not a member of a couple the other member of which is engaged in remunerative work".

• The claimant (reg 41(1)). ESA Regs 2008 reg 41(1) defines the claimant's remunerative work as "any work which a claimant does for which payment is made or which is done in expectation of payment, other than work listed in paragraph (2) of regulation 40". Reg 40(2)(f) includes the exempt work in reg 45:

  • "(2) Work for which the earnings in any week do not exceed £20.00";
  • "(4) Work which is done for less than 16 hours a week, for which earnings in any week do not exceed 16 x National Minimum Wage", rounded up to the next 50p (reg 45(9A)).

  So the claimant's test is not an hours threshold. Earnings of £20 a week or less never bar the award. Above £20, work of 16 hours or more bars it, and work under 16 hours bars it only above the higher limit.

• The partner (reg 42(1)). Paid work "for not less than 24 hours a week". Under reg 43(2)(c), a partner who would satisfy IS Regs Sch 1B para 4 (a carer) "is not to be treated as engaged in remunerative work".

Income-based JSA. Jobseekers Act 1995 s.1(2)(e): the claimant "is not engaged in remunerative work". S.3(1)(e): the claimant "is not a member of a couple the other member of which is engaged in remunerative work". JSA Regs 1996 reg 51(1) sets 16 hours "in the case of a claimant" and 24 "in the case of any partner of the claimant".

• Joint claims. A couple without children claim jointly, and each member is then a claimant held to 16 hours (s.1(2B)(b), s.35(1)). But reg 3E(1) lets a member claim alone, under s.3, where the other "is engaged … in remunerative work for 16 hours or more per week but less than 24 hours per week" (reg 3E(2)(g), in force since joint claims began on 19 March 2001). So for every couple, an award ends only when the other member works 24 hours or more.
• Carers. Reg 53 has no exception for a carer doing unrelated paid work. Its work-specific exceptions (placement care and the others it lists) are not modelled.

Someone outside the couple (a non-dependent adult in the same benefit unit) who has an award of their own claims it in their own right. They are its claimant, so the same claimant conditions apply to them: para 6(1)(e) for ESA, s.1(2)(e) for JSA.

What changes

esa_income_eligible and jsa_income_eligible, the screens applied to reported awards, tested only capital. They now also apply the work conditions:

• ESA claimant. esa_income_claimant_remunerative_work: earnings over the lower limit, unless the work is under 16 hours and within the higher limit. Being a carer does not take a claimant out of remunerative work.

• ESA earnings (new esa_exempt_work_earnings, worked out under the ESA(IR) earnings rules per DMG 41189):

  • Pay (reg 96(3)): less the income tax and primary Class 1 deducted from it, and half the occupational or personal pension contributions. The regulation takes the amounts actually deducted, which the inputs don't record. The model approximates them on the pay alone:

    • Income tax: the rates on the pay less employment expenses and employee pension contributions (net pay arrangements), above the personal and blind person's allowances, less the married couple's allowance reduction up to that tax. The allowances are the person's own, assessed on whole income, so other income can still change them through the personal allowance taper, as can anything else that moves adjusted net income (such as a loss from another trade). But the tax on other income never comes off the pay, a self-employment loss is never deducted from the pay itself, and a credit settled through a tax return (such as foreign tax credit relief) is not taken off it.
    • Class 1: the main and additional primary rates on the pay.
    • Statutory sick, maternity and paternity pay are not earnings (reg 95(2)(b)). They add nothing to the pay, nor to the tax and Class 1 taken off it. The shared bases include them, so neither is read: the shared taxable employment income includes statutory sick and maternity pay, and the shared Class 1 base (ni_class_1_income) includes all three.

    A second job taxed at the basic rate through PAYE would leave lower net earnings than this, and the variable can be entered directly.

  • Self-employment (reg 98(3), reg 99): the profit less several notional deductions, and half the personal pension contributions. Personal pension contributions go to self-employment when there is a profit and to pay otherwise. The notional deductions are:

    • basic-rate tax on the profit above the personal and blind person's allowances. A Scottish taxpayer pays the Scottish basic rate: the first band before 2018-19, the second once the starter rate came in;
    • Class 4 at the main rate between the profits limits, on reg 99(4)'s chargeable income: the reg 98 profit itself, under the model's Class 4 age rule (ni_class_4_liable). Since Charge Class 4 NICs on Chapter 2 profits, after capital allowances, the trading allowance and Schedule 2 losses #1919 the shared ni_class_4_profits is after capital allowances, the trading allowance and loss relief, which reg 98(5) and (11) keep out of the ESA profit, so the rule does not read it;
    • Class 2 until 5 April 2024 (reg 99(3)(a)), with its own rule (gov.dwp.ESA.income.self_employment_class_2) rather than ni_class_2. It is due from the small profits threshold. From 6 April 2022 it is due only above the lower profits threshold (gov.hmrc.national_insurance.class_2.lower_profits_threshold, from Fix Class 2 and Class 4 NICs against SSCBA 1992 for 2015-16 to 2026-27 #2058). SI 2022/1329 wrote that threshold into reg 99(3)(a) with effect from 6 April 2022 (reg 1).

    The notional amounts read the Class 2 and Class 4 rates and limits that Fix Class 2 and Class 4 NICs against SSCBA 1992 for 2015-16 to 2026-27 #2058 corrected to the statute.

  • Losses: a self-employment loss is not set against pay (reg 98(11)).

• ESA partner. 24 hours or more of paid work, from the new esa_income_remunerative_work_hours, unless is_carer_for_benefits (the IS gate's proxy for Sch 1B para 4).

• JSA. The claimant at 16 hours or more, and the other member of the couple at 24 or more, from the new jsa_remunerative_work_hours. There is no exception for carers doing unrelated paid work.

• Who is tested. The claimant is a member who reports the award, tested against the other member of the couple. When the claimant or partner reports one, only they are candidates. A member outside the couple who reports an award is tested in the benefit unit's screen only when neither does, so their award can never change the couple's. Without this rule a non-dependant's award could flip the benefit-unit screen, and with it the Income Support gate.

• Members outside the couple on an award (is_on_income_related_esa, is_on_income_based_jsa, from Count only the claimant's and partner's legacy awards in means tests and passports #2027; helper own_report_is_paid). These statuses feed the council tax reduction non-dependant exemptions and the maintenance-loan benefits signal. Such a member is on the award only when their own claim passes:

  • they are not in remunerative work as its claimant (esa_income_claimant_remunerative_work, or 16 hours for JSA);
  • their report exceeds the tariff income from the benefit unit's capital, within the capital limit.

  The benefit unit's award must also be in payment: positive, or nil while the benefit unit's screen fails and the award is not neutralised. On the formula path, a nil award alongside a passing own claim can only mean the claimant's or partner's own claim failed the screen. A zero cannot say why it is zero, so:

  • neutralising the award always removes the status, and so does JSA closing (gov.dwp.JSA.income.active);
  • a nil award entered directly, or a reform that replaces the award with nil, removes it only while the screen passes. While the claimant's claim fails, the status follows the member's own claim.

  Before this, the status read the benefit unit's screen, which tests only the claimant and partner when either reports. A working outside member was then on the award whenever the claimant's claim passed, and a non-working one lost it whenever the claimant's failed. In a synthetic Merton 2027 household, that took a working non-dependant out of a £551.20-a-year deduction.

New parameters, each dated from the point-in-time legislation:

Parameter	Values
gov.dwp.ESA.exempt_work.lower_earnings_limit	£20
gov.dwp.ESA.exempt_work.hours_limit	16
gov.dwp.ESA.exempt_work.higher_earnings_limit	£92 from 27 Oct 2008; £93 from 1 Oct 2009; 16 x NMW rounded up to 50p from 11 Apr 2011, at each NMW change. That gives £183.50, £195.50 and £203.50 for 2024/25-2026/27, matching DWP's published limits. Later years follow average earnings, rounded up to 50p
gov.dwp.ESA.income.remunerative_work.partner_hours	24 from 27 Oct 2008
gov.dwp.ESA.income.self_employment_class_2.deducted	true from 27 Oct 2008; false from 6 Apr 2024 (SI 2024/377)
gov.dwp.ESA.income.self_employment_class_2.above_lower_profits_threshold	false from 27 Oct 2008; true from 6 Apr 2022 (SI 2022/1329)
gov.dwp.JSA.remunerative_work.claimant_hours	16 from 7 Oct 1996
gov.dwp.JSA.remunerative_work.partner_hours	24 from 7 Oct 1996

The existing gov.dwp.JSA.hours.single and couple are kept, and a test pins them to the new ones. Their descriptions called 24 hours the "joint claimants'" requirement; it is the partner's.

National Insurance. Earlier heads of this PR corrected the 2015-16, 2022-23 and 2023-24 Class 2 and Class 4 amounts, because reg 99(3) deducts notional Class 2 and Class 4. #2058 landed the same values with references, and the four 2023 NI tests at the same hand-derived results (£179.40, £1,568.70, £3,393 and £1,748.60). The merge of main took main's version of all nine files.

Not modelled (each documented on the variable; enter the variable directly to apply it):

• ESA: supervised work, test trading, voluntary work, work placements and the other reg 40(2) work.
• The pre-April 2017 52-week limit on ESA permitted work.
• The transferable (marriage) allowance and other reliefs in reg 99(1) beyond the personal and blind person's allowances.
• The reg 43(1) and JSA reg 53 employments, leave, illness and trade disputes.
• The other reg 3E(2) grounds, which matter for the joint-claim route but not for remunerative work.

Design points

In the screen, beside the capital test. The work conditions are further conditions of entitlement alongside the capital test the screens already apply. For ESA they are para 6(1)(e)-(f) beside (b); for JSA they are ss.1(2)(e) and 3(1)(e) beside s.13. Work ends an award. Within the model's bounded reported-award approach (no new claims can be made, so a reported award is screened rather than computed), applying them in the screen keeps every reader consistent:

• household income (esa_income, jsa_income);
• the Income Support gate (s.124(1)(f), (h));
• Count only the claimant's and partner's legacy awards in means tests and passports #2027's claimant-or-partner awards, and the readers and passports that use them.

The alternative would be household income keeping the reported award while the gate and passports see no entitlement. The household would then both have and not have the benefit. The cost is that household income falls where a survey report conflicts with the rules. The impact section sizes that.

Helpers and award formulas unchanged. income_related_esa_award, income_related_jsa_award and the esa_income/jsa_income formulas are untouched; only the variables' documentation changed. They read the screens. So #2025's and #2027's value rule for direct entries is unaffected: a calculated esa_income is still exactly the award on everyone's reports. #2027's family readers call the same helpers, so they inherit the screens.

One screen for the benefit unit's award. Because esa_income and jsa_income stay that award, the benefit unit's award applies one screen to every report in it. When the claimant or partner reports, a report by someone outside the couple is paid or not with the claimant's claim. That member's own status is tested on their own claim (above), so passports follow the law, but the benefit unit's total can include or omit their award. No Enhanced FRS record has anyone outside the couple reporting either award in 2025-2030 (0 records), so this does not reach the impact below.

Income Support property tests (test_income_support_eligibility_properties.py). The reference's award on the claimant's and partner's reports, and explained_by_reports' award on everyone's reports, now pass the same work screens (new shared tests/legacy_award_work_reference.py). The monotonicity property needed one law-based allowance. A carer is not in remunerative work for Income Support (reg 6(4)(c)) but is for JSA, so more hours can end a carer's own JSA award that barred the claim. The property therefore allows a gain only where a calculated esa_income or jsa_income award actually ended.

Not every receipt signal is screened. bursary_fund_16_to_19_receives_uc_or_esa_in_own_right reads esa_income_reported and esa_contrib_reported directly, as its documented reported-benefit proxy. This PR does not change it, so a student whose income-related ESA report fails these screens keeps that signal (#2263).

Tests

• YAML. esa_income_remunerative_work.yaml has 45 cases and jsa_income_remunerative_work.yaml has 18, each hand-computed from the law. They cover:

  • £20 a week at 30 hours, which is exempt;
  • £21 at 16 hours, which is remunerative;
  • £195.50 and £196 at 15 hours in 2025-26, and £196 and £204 in 2026-27;
  • half pension contributions;
  • tax on pay as the lowest slice;
  • reg 99's notional tax on a £13,200 profit with a £5,900 pension premium (£193.97);
  • reg 99's Class 4 on the profit before capital allowances, and without another trade's loss (£195.12 within £195.50 in both, where the shared Class 4 profits would give £195.85);
  • a self-employment loss of £1,000 or £20,000 beside £11,000 of pay (£211.54 either way);
  • the blind person's allowance in reg 99's notional tax (£196.39);
  • a 2017-18 Scottish profit with the Scottish basic rate and Class 2 (£122.28 against £120);
  • the married couple's allowance with a loss, and a foreign tax credit on other income, neither changing the tax on pay;
  • statutory sick, maternity or paternity pay beside £10,300 of pay (£198.08 each time), and a control that sets the shared taxable pay and Class 1 base to £13,300 without moving the result;
  • 2023-24 reg 99 cases: £12,500 of profit (£170.19 against £167), a profit equal to the £12,570 lower profits threshold (no Class 2: £168.65), £1 over it (Class 2 brings it to £165.22, the intended cliff), and £13,000 (Class 2 at £3.45: £167.23);
  • 2022-23 reg 99 cases: a profit equal to the £11,908 lower profits threshold (no Class 2: £154.00 against £152) and £1 over it (Class 2 at £3.15 and 9.73% Class 4: £150.87, the same cliff);
  • 2015-16 cases: £5,000 of profit (no Class 2: £96.15 against £104), and a profit equal to the £5,965 small profits threshold (Class 2 due: £102.30);
  • a carer claimant and a carer partner;
  • a partner at 23 and 24 hours;
  • both members reporting;
  • non-dependants: their own award tested on their own work, their award not rescuing the claimant's, and their own status both ways for each benefit. Working, they are not on the award whatever the claimant reports; not working, they stay on it when the claimant's claim fails;
  • couples without children at 15, 16, 20, 23 and 24 hours (reg 3E), and a couple caring for a placed child;
  • the Income Support gate.

  With main's two capital-only screens and everything else at 6880c0e, 32 of the 63 cases fail. They are every case where the work condition bars an award, plus the two where a non-working non-dependant keeps their award while the claimant's claim fails (with the capital-only screen the claimant's claim never fails).

• Hypothesis, work screens (test_legacy_award_work_properties.py, 4 properties, 20 examples each):

  • the screens, and each adult's weekly earnings, equal a family-by-family reading of the law in 2025 and 2026. Families mix employees, self-employed people, pay with profit, losses beside pay (including losses larger than the pay), pension contributions, statutory sick, maternity and paternity pay, ages either side of state pension age, placed children and non-dependants;
  • more hours or pay never makes a family eligible (in the tested 2025 and 2026, within the generated scope);
  • caring never ends an ESA award and never matters for JSA (no exception for carers doing unrelated paid work);
  • a member outside the family never changes the couple's screen.

• Hypothesis, Count only the claimant's and partner's legacy awards in means tests and passports #2027's readers (test_legacy_award_readers_properties.py, updated). The claimant-or-partner reference now applies esa_screen/jsa_screen, the same independent reading of the law. The status reference tests a member outside the couple on their own work. The claimant, partner and outside adults now draw hours (0, 15, 16, 24 and 40 a week) and pay (£0, £8,000 or £20,000), so the screens bite for both awards. In the examples drawn:

  • a reported claimant-or-partner award is screened out in 87% (ESA) and 80% (JSA);
  • an outside member's own work bars them in 60% of examples for each award.

  Three deterministic tests keep this coverage independent of the examples drawn:

  • a working claimant's reported ESA and JSA give nil claimant-or-partner awards (the review's £20,000 case, at 40 hours so that JSA's test bites too);
  • an outside adult's status for ESA, JSA and the maintenance-loan signal, both ways;
  • the Merton 2027 non-dependant deduction, which a working outside adult's own report no longer removes.

  With ffd7525's status code (_legacy_award_payee.py and the two status variables) and everything else at 6880c0e, the four new YAML status cases fail (59 pass). Four of the 14 reader tests fail: the invariance and person-status properties, the outside-adult test and the Merton test.

• Parameters.

  • The higher limit equals 16 x NMW rounded up to 50p at every rate in DMG ch. 41 Appendix 5, and matches DWP's published limits.
  • Projected years are multiples of 50p and never fall.
  • The JSA hours parameters match the older ones.

• Suites. The head 672d0a6 changes only documentation strings from 6880c0e (the reviewer checked the ASTs).

  • At 6880c0e, run locally: ruff format --check clean (1,315 files); YAML under tests/policy/baseline/finance/benefit and tests/policy/baseline/gov, 2,648 passed.
  • Also at 6880c0e: the targeted pytest files, 121 passed. They are the work, Income Support and reader properties, the direct-input, entered-directly and lifecycle tests, the parameter tests and the NI Class 4 tests.
  • At 672d0a6: the two remunerative-work YAML files, 63 passed.

Invariants

Each holds for every input in the stated scope:

1. Subset of the old screen. An award that passes the new screen passed the capital-only one: the screen is the old capital test and the work conditions together.
2. Monotonicity. More hours or pay for anyone never makes an award pass, from 6 April 2024. The property tests 2025 and 2026, for adults with no income other than their pay, profit and statutory payments, within the basic rate band and the Class 4 upper limit. Two exceptions are not claimed:
   • Before 6 April 2024, reg 99(3)(a)'s fixed Class 2 deduction begins once profit reaches its threshold (from 6 April 2022, once it exceeds it), so £1 more profit can lower net earnings. YAML cases pin this intended cliff in 2022-23 and 2023-24.
   • Other income can move the personal allowance through the taper, so the invariant is not claimed outside that scope.
3. Caring. Caring never ends an ESA award and never changes a JSA award.
4. Non-dependants and the couple. When the claimant or partner reports the award, adding a non-dependent adult, with any award and any work, never changes either screen.
5. Non-dependants' own awards. On the formula path of an open scheme, with no award entered or replaced, a member outside the couple is on income-related ESA or income-based JSA exactly when their own claim passes: their own work, their own report above tariff income, and the capital limit. Nobody else's report or work changes this.
6. Readers. The claimant-or-partner awards equal the award on the claimant's and partner's reports after tariff income and within the capital limit, while the benefit unit passes the work screen.
7. Placed children. A child placed with a couple never changes the JSA work test.
8. Parameters. The higher limit equals 16 x NMW rounded up to 50p whenever it is NMW-linked, and projected values are 50p multiples.

Invariants 2-6 are property tests, with deterministic cases for 5 and 6. Invariant 7 is a YAML case and follows from reg 3E. Invariant 1 follows from the screen's form and is checked through the reference test. Invariant 8 is a table and range test.

Enhanced FRS impact

Real Microsimulations on fresh copies of enhanced_frs_2024_25.h5 (sha256 e433e532…), base main e89b8ab against this branch at 6880c0e (the head 672d0a6 changes only documentation strings), each from its own worktree venv with a clean tree. Benefit units and households are weighted thousands; "records" are dataset records. Figures resting on fewer than 10 records are not shown.

Year	Income-related ESA	Benefit units losing it	Household net income	Households losing
2025	−£88.3m	7.7k (40 records)	−£89.2m	7.9k (48 records)
2026	−£90.7m	7.7k (40)	−£90.8m	7.7k (40)
2027	−£92.7m	7.8k (40)	−£92.7m	7.8k (40)
2028	−£94.9m	7.8k (40)	−£94.9m	7.8k (40)
2029	−£304.2m, of which −£97.1m outside the salary-sacrifice artefact	24.7k (92), of which 7.8k (40) outside it	−£305.1m	24.7k (92)
2030	−£311.6m, of which −£99.4m outside it	24.8k (92), of which 7.9k (40) outside it	−£312.4m	24.8k (92)

• By channel in 2025 (first match: the reporting claimant's own work, then the other member's 24 hours):
  • ESA claimant in remunerative work: 3.0k units, 22 records, −£32.4m.
  • ESA partner working 24+ hours: 4.7k units, 18 records, −£55.9m.
• Other benefits.
  • Income-based JSA changes only in 2025 (the scheme is closed from 2026), and council tax reduction in 2025-2027 and 2029-2030. Each changes for fewer than 10 records a year, so they are in the household totals but not shown.
  • Housing Benefit changes only in 2029 and 2030: −£0.9m in each year, from 10 records, all in benefit units with the salary-sacrifice artefact.
  • Universal Credit, Pension Credit and Income Support do not change in any year (0 records).
• Poverty. Changes in 2025-2028 rest on fewer than 10 households and are not shown. In 2029-2030 the all-person poverty rates (before and after housing costs) rise by at most 0.012 percentage points, from 18 to 32 households.
• Against the previous head. Every saved array (101 a year) and every aggregate equals the run at ffd7525. No record has anyone outside the couple reporting either award, so the status fix does not reach these figures. The final review of ffd7525 found that run equal to the earlier runs at b9816e4 against main 88ecf58.

The 2029-30 jump is a data defect, not this rule. From 2029 the £2,000 salary-sacrifice cap moves sacrifice above the cap into employment_income (salary_sacrifice_returned_to_income).

• In 2029, 1,283 records (874k weighted) with no pay of their own gain employment income from it.
• Among income-related ESA reporters that is 106 records (27k), 97 of them with zero hours.
• The claimant earnings test reads that as work above the higher limit. Splitting the same runs, the units whose ESA claimant has no pay or profit of their own account for −£207.1m of the 2029 change and −£212.2m of 2030's (52 records). The rest continues the 2025-2028 trend.
• Chip task_d73fcf28 restricts the imputation to people with pay and guards the model.

What the 40 records are. They are survey reports of income-related ESA alongside work the regulations say ends it: 22 records with a claimant over the exempt-work limits and 18 with a partner working 24+ hours. The contributory/income-related split is the respondent's own report (FRS BENEFITS VAR2). Some may be contributory ESA, which a partner's work does not affect (DMG 41302) and which this screen does not touch. The rule is the law; whether the survey labels are right is a data question.

Review

• r1-r4 (Subfleet, GPT-6.1 Sol): each requested changes, all addressed:

  • r1 (fixed in d146b96): reg 3E single claims, reg 99's notional tax, reg 98(11) losses; also rounding, the monotonicity waiver and documentation.
  • r2 (fixed in ebed8bd): a loss reaching the pay's tax through the shared pension relief, the Scottish basic rate before 2018-19, reg 99(3)(a) Class 2, and the blind person's allowance.
  • r3 and r4 (fixed in 17ebd16 and e38ea0c): the shared tax reductions on the pay side, the NI amounts (since landed by Fix Class 2 and Class 4 NICs against SSCBA 1992 for 2015-16 to 2026-27 #2058), and statutory pay (reg 95(2)(b)).

• r5: APPROVE WITH NITS at e38ea0c; the nits were fixed in 232358d.

• Final review of b9816e4 (after merging main with Apply the Income Support remunerative work and JSA conditions (s.124(1)(c), (f)) #2025): APPROVE WITH NITS. The source nits were fixed in a359287; the rest are in this description.

• Final review of ffd7525 (after merging Count only the claimant's and partner's legacy awards in means tests and passports #2027 and Fix Class 2 and Class 4 NICs against SSCBA 1992 for 2015-16 to 2026-27 #2058): REQUEST CHANGES, with two should-fix findings and a nit, all fixed in 6880c0e and this description:

  • a working member outside the couple could be on the award through the couple's screen, and a non-working one lose it;
  • Count only the claimant's and partner's legacy awards in means tests and passports #2027's reader properties compared against capital-only references;
  • this description was stale, and reg 99's comment dated SI 2022/1329 to 2023.

• Final review of 6880c0e (Subfleet, GPT-6.1 Sol): APPROVE WITH NITS. It re-ran:

  • the 63 remunerative-work YAML cases and the reader properties (14 passed, with the same coverage statistics);
  • the legacy-reader YAML (72) and the override, removal and float32 tests (62);
  • 180 regional and 32 removal and precision probes.

  It also recomputed the impact aggregates. Its two nits, a record count here and the wording on when a nil award removes an outside member's status, are fixed in 672d0a6 and this description.

• Final review of 672d0a6 (Subfleet, GPT-6.1 Sol): APPROVE, no findings. Both nits resolved, and no executable line changed.

Axiom

axiom: TheAxiomFoundation/rulespec-uk#417 queued. rulespec-uk has no module for:

• WRA 2007 Sch 1 para 6;
• ESA Regs 2008 regs 40-45, 96, 98 and 99;
• Jobseekers Act 1995 ss.1, 3 and 35;
• JSA Regs 1996 regs 3E and 51-53.

The statute provisions are not in the corpus. The regulations are only in the non-release scope 2026-06-01-uk-frs-microsim. The issue carries the verbatim law, the required outputs, a review_finding and 12 groups of companion tests from the same ground truth as the YAML.

Follow-ups

• task_d73fcf28: stop imputing salary sacrifice to people with no pay (uk-data), and guard salary_sacrifice_returned_to_income.
• task_3ccdecda: pension_contributions_relief goes negative when losses exceed income.
• Legacy award follow-ups from #2031: the bursary own-right ESA proxy, and one screen for reports outside the couple #2263: bursary_fund_16_to_19_receives_uc_or_esa_in_own_right reads reported ESA, not the screened status.
• Legacy award follow-ups from #2031: the bursary own-right ESA proxy, and one screen for reports outside the couple #2263: the benefit unit's award applies one screen to reports by members outside the couple (design point above; no survey records).
• National Insurance: Fix Class 2 and Class 4 NICs against SSCBA 1992 for 2015-16 to 2026-27 #2058 does not change Class 1, so Class 1 rates and thresholds that change part-way through 2022-23 and 2023-24 are still sampled at 30 April. They reach the pay side's Class 1 here as they reach every Class 1 figure.

🤖 Generated with Claude Code

[MaxGhenis]

The NI parameter corrections this PR carries (review r4 finding 1) now have their own PR off main: #2058. Note for whoever next works on this branch.

The hunks are not byte-identical. #1888 already held the Class 2 fix in a different form (every value cited, new class_2/lower_profits_threshold.yaml and lower_profits_threshold_applies.yaml, and ni_class_2 charging only above the lower profits threshold in 2022-23 and 2023-24). #2058 takes that commit and adds the Class 4 fix; #1888 is closed as superseded.

Every value in this branch's NI hunks is in #2058 with the same number:

• small profits threshold £5,965 from 2015;
• Class 2 £3.45 from 6 April 2023;
• lower profits limit £12,570 from 6 April 2023;
• Class 4 main rate 9.73% for 2022-23 and 9% from 6 April 2023; additional rate 2.73% and 2%.

The four 2023 test expectations match too (179.40; 1,568.70; 3,393; 1,748.60).

#2058 also has:

• the small profits threshold at £6,725 for 2023-24 and 2024-25, with the uprating anchor removed;
• the 2015-16 upper profits limit of £42,385;
• corrected dates for the 2016-17 upper limit, the 6% main rate and the end of compulsory Class 2.

When #2058 lands, merge main and take main's version of these paths whole, without hand-merging hunks:

git checkout origin/main -- policyengine_uk/parameters/gov/hmrc/national_insurance/class_2 policyengine_uk/parameters/gov/hmrc/national_insurance/class_4 policyengine_uk/tests/policy/baseline/gov/hmrc/national_insurance

Checked against this branch. I overlaid #2058's NI parameter, variable and test files onto head e38ea0c in a probe worktree. esa_income_remunerative_work.yaml plus the NI YAML: 108 passed. The 2023-24 case (£12,500 profit, £7,300 personal pension) gives esa_exempt_work_earnings 170.19 and esa_income_eligible false.

esa_exempt_work_earnings reads the Class 4 lower profits limit and this PR's own DWP switch for the Class 2 threshold. That still works. The HMRC-side equivalents after #2058 are gov.hmrc.national_insurance.class_2.lower_profits_threshold and .lower_profits_threshold_applies.

Not covered by #2058: the Class 1 rates and thresholds that changed mid-year in 2022-23 and 2023-24 (#1807).

Axiom: rulespec-uk#353 (s.11) and rulespec-uk#423 (s.15 and regulation 100) are the queued parity issues for the NI history.

[MaxGhenis]

Cross-PR note from #2075 (stacked on #2027 + #2025): #2075 switches income_support_eligible's ESA and JSA limbs to read claimant_or_partner_esa_income > 0 / claimant_or_partner_jsa_income > 0. Those variables still go through income_related_esa_award / income_related_jsa_award, so this PR's work conditions in the screens reach the gate unchanged.

#2075 does not edit test_income_support_eligibility_properties.py, esa_income_eligible or jsa_income_eligible. Its new differential test imports FAMILIES, input_settings, situation and DIFFERENTIAL_SETTINGS from that property file, so renaming any of them would need a matching change in #2075.

Because the gate now reads a calculated variable, a test that changes esa_income/jsa_income after calculating the gate must also delete the claimant_or_partner_* arrays.

[vahid-ahmadi]

Thanks Max. The law is read correctly:

• ESA claimant: remunerative work is any paid work other than reg 40(2) work (reg 41(1)). Work is exempt where earnings "do not exceed £20.00" (reg 45(2)), or where it is done for "less than 16 hours a week" with earnings within 16 × NMW, rounded under reg 45(9A).
• ESA partner: the test is "not less than 24 hours" (reg 42(1)), with the carer exception in reg 43(2)(c).
• JSA: 16 hours for the claimant and 24 for the partner (reg 51(1)(a)-(b)).
  • Reg 3E(2)(g) ("16 hours or more per week but less than 24 hours per week") makes 24 hours the limit for the other member of any couple.
  • Reg 51(3)(c) only discounts unpaid caring hours, so having no carer exception for paid work is right.

Verified at 232358d:

• Higher limit: £183.50, £195.50 and £203.50 for 2024-25 to 2026-27 (16 × £11.44, £12.21 and £12.71, rounded up). The projections run from £208.50 to £223.00 in 2030.
• The NI corrections match the statutory values:
  • 2023-24: Class 2 at £3.45, the lower profits limit at £12,570, and the Class 4 additional rate at 2%;
  • 2022-23: Class 4 at 9.73% and 2.73%;
  • 2015-17: the small profits threshold at £5,965.
• Tests:
  • YAML: esa_income 52/52, jsa_income 28/28, income_support 86/86, NI 49/49.
  • The new properties pass 4/4, the IS properties 4/4 and parameters 6/6.
• Merges cleanly with #2027 and #2075.

1. Blocking: the 2029-30 salary-sacrifice artefact ships with this PR.
   • The probe: a single 45-year-old with no pay and 0 hours, who reports £5,000 of income-related ESA and has £30,000 of imputed salary sacrifice.
   • In 2028 ESA is £5,000.
   • In 2029 salary_sacrifice_returned_to_income puts £28,000 into employment_income. esa_exempt_work_earnings becomes £246.27 a week, above the £217.50 limit, and the award drops to £0.
   • The body puts about £207-212m a year of the 2029-30 change down to this, against about £97-99m that is real. So in the years our fiscal projections cover, this PR roughly triples the cut.
   • Fix: land the guard from task_d73fcf28 first, or add it here. In esa_exempt_work_earnings.py:76, don't count salary_sacrifice_returned_to_income for a person with no employment income before the cap. Add the 2029 zero-hours case.
2. Should fix: the NI parameter hunks duplicate #2058, and your comment says the values are the same. A trial merge conflicts in all five NI parameter files (class_2/flat_rate.yaml, class_2/small_profits_threshold.yaml, class_4/rates/main.yaml, class_4/rates/additional.yaml, class_4/thresholds/lower_profits_limit.yaml) and all four NI test files. Drop these hunks once #2058 lands, or rebase onto it; the reg 99 cases should then pass unchanged.
3. Should fix: these parts of the body give unweighted record counts, several below 10: the impact table, "By channel in 2025", the statutory-pay paragraph, the 2029-30 section and "What the 40 records are". Keep the weighted figures and remove or round the record counts.
4. Should fix (inherited): this needs the same #2013 rebase as #2025 (is_SP_age against main's qualifying-age variable, and the s.124(1)(g) limb). The IS gate reads these screens, so re-run the IS properties after it.

[MaxGhenis]

Status at head 672d0a62e, from the scheduled lane on 10 October:

• Review: the final review of ffd752576 requested changes (GPT-6.1 Sol, Subfleet). 6880c0ebd fixes them:

  • a member outside the couple is now on income-related ESA or income-based JSA only when their own claim passes the work, capital and tariff tests;
  • Count only the claimant's and partner's legacy awards in means tests and passports #2027's reader property references now apply the work screens.

  The review of 6880c0ebd gave APPROVE WITH NITS. 672d0a62e fixes its nits (documentation only), and the follow-up review gave APPROVE at exact head 672d0a62e.

• Local runs at 6880c0ebd: ruff is clean; the YAML under finance/benefit and gov passes (2,648); the targeted pytest files pass (121).

• Enhanced FRS impact: re-run against main e89b8abcf. It is unchanged from the previous head, and the description is updated.

• CI: Test was cancelled at the 60-minute limit twice, in run 38067320074 attempts 1 and 2. Each time 2,950 YAML had passed and pytest was at 98%, with no failures. This is the repo-wide timeout in PR Test job times out at 60 minutes: the suite now takes about an hour #2257, which Raise the Test job timeout to 120 minutes #2258 fixes. Once Raise the Test job timeout to 120 minutes #2258 merges, this branch needs main merged in, a green run, and a review of that merge before it lands.

• Follow-ups: Legacy award follow-ups from #2031: the bursary own-right ESA proxy, and one screen for reports outside the couple #2263.

[MaxGhenis]

UK hub, for this PR's owner: the 2029-30 impact here should be re-run on current main.

#2063 merged on 10 October (88d4ff48d). With its guard in main, salary sacrifice attributed to people with no pay is no longer counted as pay, so it no longer reaches the ESA permitted-work test on the published dataset. The session that built #2063 reports 106 ESA-reporter records (27k weighted) had such pay in 2029. I have not re-measured that figure.

To pick the change up, merge main into this branch and re-run the impact for 2029-30. Merging main also gives the Test job its new 120-minute limit (#2258).