Apply the 6 April 2017 child cutoffs by date of birth

[MaxGhenis]

Fixes #1903

Builds on #1899 (merged 2026-10-03), which added months_since_last_birthday. This now targets main and has main merged in, including #1907's State Pension Credit qualifying age. Its variables state_pension_credit_qualifying_age and has_attained_state_pension_credit_qualifying_age, its helper months_since_attaining, and meets_savings_credit_age_requirement's "attained 65" test now take the same shared birth instant as State Pension age, so a date_of_birth input sets them consistently. Without one, every pension-age result is main's, which the impact run confirms record by record. #1907's age-only helper utils.state_pension_age.date_of_birth, left without callers, is removed.

Summary

Four rules turn on whether a child was born before 6 April 2017, and a fifth, the Pension Credit mixed-age couple saving, on whether the older partner was born on or before 5 February 1954:

Rule	Law	Variable
Universal Credit child limit: transitional protection	UC Regs 2013 reg 24A(3), in force 6 April 2017 to 5 April 2026	uc_is_child_born_before_child_limit
Higher UC child element for the first child	UC (Transitional Provisions) Regs 2014 reg 43	the same variable, via uc_individual_child_element
Child Tax Credit child limit	Tax Credits Act 2002 s.9(3A) and (3B); CTC Regs 2002 reg 9	is_CTC_child_limit_exempt
Pension Credit first child addition for the eldest child	SPC Regs 2002 Sch IIA para 10	child_minimum_guarantee_addition
Pension Credit mixed-age couple saving: the older member reached the qualifying age by 14 May 2019, so was born on or before 5 February 1954	SI 2019/37 art. 4(1); Pensions Act 1995 Sch 4 para 1	has_mixed_age_couple_pension_credit_saving

The first four compared birth_year (period less whole age) with 2017. So everyone aged 9 in 2026-27 counted as born in 2017, whether born in January (before the cutoff) or May (after it). In survey data, the whole cohort born from October 2016 to October 2017 was treated as born after the cutoff, when about half of it was born before.

This PR compares an exact date of birth with 6 April 2017:

• date_of_birth (new, YYYYMMDD). Each person was born age years and months_since_last_birthday months before 6 October of the fiscal year. The day of birth is the one starting at or after that instant, so their legal age on 6 October is age. It uses Set State Pension age from date of birth, including the rise to 67 #1899's date grid. It can also be set directly, with age as the person's age on 6 October. Every date-of-birth rule then reads one instant of birth through a shared helper, utils.dates.birth_instant, so a date_of_birth input sets state_pension_age, months_since_state_pension_age, is_SP_age and each cutoff consistently. Age-based rules still read age. A situation that sets date_of_birth for some people gives the others 0, which the helper treats as not given; see the review section. Without an input, state_pension_age is exactly Set State Pension age from date of birth, including the rise to 67 #1899's.

• born_before date parameters replace the year parameters:

  • gov.dwp.universal_credit.elements.child.limit.born_before
  • gov.dwp.tax_credits.child_tax_credit.limit.born_before
  • gov.dwp.pension_credit.guarantee_credit.child.first.born_before, which was a hard-coded 2017.

  Each is 20170406 and cites its provision.

• Children are ordered by date of birth. UC Regs reg 24B(1) orders children "by reference to the date of birth ... taking the earliest date first", and Sch IIA para 10 names the "eldest". uc_child_index and the Pension Credit child rank now sort by date_of_birth rather than whole age. This only breaks ties between children of the same whole age, but once a cutoff can fall between two such siblings the order matters. With the old ranking, a family with children born 5 and 6 April 2017 could list the later-born child second, and it would get the child element under the limit.

• birth_year is now the calendar year of date_of_birth. Its model readers on main were the three child cutoffs and the mixed-age saving; all four now compare the exact date instead.

For the child rules, household situations with whole ages are unchanged. A whole age in a situation sits in the middle of the year of age, which is 6 April of the year the person turned that age. So "born before 6 April 2017" gives exactly what period - age < 2017 gave, and birth_year is still period - age. This is tested for every year from 2015 to 2035 and every age up to 19. Only survey data, which spreads birthdays over the year (#1899), moves, along with fractional ages. A fractional age is now read as the exact age on 6 October, as #1899 already does for State Pension age: in 2025, an age of 8.25 places the birth on 6 July 2017, after the cutoff, where the old truncated birth year said 2016. A date of birth can now also be set directly (date_of_birth: 20170405) for a year, with age as the age on 6 October. A value that isn't a calendar date, or that gives a different age, raises an error rather than being silently normalised.

Invariants (stated and tested)

policyengine_uk/tests/test_date_of_birth.py:

1. Households with whole ages match the birth-year rule. Exhaustively, for 2015-2035 and ages 0-19:

   • born before the cutoff ⇔ period - age < 2017;
   • is_CTC_child_limit_exempt agrees, and so does uc_is_child_born_before_child_limit for under-16s;
   • birth_year == period - age.

   This differential test pins "whole-age situations unchanged".

2. The date of birth follows the legal age (Hypothesis). For any whole age 0-15, any position in the year (including instants within a day) and any year 2015-2040:

   • date_of_birth equals a datetime reference that shares no code with utils/dates.py, though it re-implements the same grid rule;
   • the legal-age check below is the fully independent one;
   • the person has legal age age on 6 October;
   • birth_year is its year;
   • the UC and CTC flags hold exactly when the date is before 6 April 2017.

3. Ordering refines age order (Hypothesis). uc_child_index is a permutation of 1..n. A child born earlier is always numbered first, and so is an older whole age.

4. Microdata shares match the statute. In a data-built simulation with weights, for 2022, 2024, 2025, 2026 and 2028 and ages 5-12:

   • the weighted share of each age born before 6 April 2017 equals the share of that year of age's days before it (about 181/365 for the straddling cohort, 0 or 1 for the others), within the largest weight plus one day;
   • the share with birth_year the year before period - age equals the share of birthdays from 7 October to 31 December;
   • every record's UC and CTC flags equal date_of_birth < 20170406.

5. A date input matches the same birthday placed by age (Hypothesis, births_and_year). For any year from 2015 to 2040, and people of either sex born from 1935 to the year before, setting date_of_birth (with the legal age) gives the same birth_year, is_SP_age, has_attained_state_pension_credit_qualifying_age, meets_savings_credit_age_requirement, UC flag and CTC flag as setting the matching months_since_last_birthday. state_pension_age, state_pension_credit_qualifying_age and months_since_state_pension_age agree exactly (to 1e-9). A woman's SPC qualifying-age attainment equals her is_SP_age. A person left without the input in the same situation reads date_of_birth 0 but gets the results they would with no input at all.

   A uniform draw rarely reaches the transitional cohorts in their attainment year, where a wrong birth instant would change SPC attainment, so two are pinned as @examples: a woman born 10 December 1952 in 2015, and a man born 10 November 1960 in 2027. A mutant whose attainment ignores the date input fails on both. The Savings Credit "attained 65" half can't be told apart on this path, because the input check already forces the legal age to equal age.

Boundary YAML tests (born 5 against 6 April 2017):

File	Cases
demographic/date_of_birth.yaml	9: the derivation, a December birthday, a fractional age, the UC and CTC cutoffs reached through age and months_since_last_birthday, a date set for one person only (the parent keeps State Pension age 68, not over it), and a date setting State Pension age exactly (5 January 1960: 66, attained 5 January 2026, 9.032 months before 6 October)
uc_is_child_born_before_child_limit.yaml	6 new: a third child either side of the limit (2025-26); same-age siblings ordered by date of birth; the higher first child element either side (2026-27); a sibling without a date input placed by age
tax_credits.yaml	4 new: the exemption flag and CTC_child_element (2024-25)
child_minimum_guarantee_addition.yaml	3 new: an only child either side, and the eldest of two same-age children (2026-27)

As a mutation check, reverting either ranking to -age fails exactly the two same-age ordering tests.

Microsimulation impact

Enhanced FRS 2024-25 (policyengine-uk-data-private, snapshot ace89433). Each column is a full microsimulation of current main (8090f1c19, which includes #1899 and #1907) and of this branch (297a210f8). The later commits 20e67683d and dc2ae4e22 change no model code: they remove an uncalled helper and touch only tests, docs and changelog. These figures therefore stand at the head. Children in poverty use HBAI's definition (is_hbai_dependent_child).

This branch minus main

		2024-25	2025-26	2026-27	2027-28	2028-29	2029-30	2030-31
Children in the straddling cohort (period - age = 2017), level	k	751	817	921	753	825	959	774
of whom now counted as born before 6 April 2017 (main: 0)	k	+373	+404	+472	+382	+417	+478	+384
Children with a UC child element in benefit units receiving UC (main 5,255k, 5,202k)	k	+92	+40	+0	+0	+0	+0	+0
UC children not receiving one because of the child limit	k	-91	-38	+0	+0	+0	+0	+0
UC benefit units with a child held back by the limit (main 656k, 698k)	k	-54	-20	+0	+0	+0	+0	+0
Mixed-age couples losing the Pension Credit saving (survey benefit units)	k	-1.0 (3)	-6.6 (5)	-0.0 (1)	-2.0 (6)	-0.5 (3)	-1.4 (6)	0
Universal Credit	£bn	+0.36	+0.22	+0.04	+0.05	+0.06	+0.07	+0.05
of which the child cutoffs	£bn	+0.35	+0.15	+0.04	+0.05	+0.06	+0.06	+0.05
of which the mixed-age saving	£bn	+0.00	+0.06	+0.00	+0.00	+0.00	+0.00	+0.00
Pension Credit (all from the mixed-age saving)	£bn	+0.00	-0.05	-0.00	-0.02	-0.00	+0.00	+0.00
Child Tax Credit	£bn	+0.00	+0.00	+0.00	+0.00	+0.00	+0.00	+0.00
Household net income	£bn	+0.34	+0.15	+0.04	+0.03	+0.06	+0.06	+0.05
Government balance	£bn	-0.34	-0.15	-0.04	-0.03	-0.06	-0.06	-0.05
People in absolute poverty, BHC	k	-94	-19	+0	-10	+0	+0	+0
People in absolute poverty, AHC	k	-59	-0	-48	+2	+0	-9	-1
Children in absolute poverty, BHC	k	-69	-11	+0	-6	+0	+0	+0
Children in absolute poverty, AHC	k	-47	-0	-24	+0	+0	-6	-1
Survey households leaving / entering absolute poverty, BHC; AHC	households	6/2; 2/0	2/0; 2/0	0/0; 1/0	1/2; 0/2	0/0; 0/0	0/0; 2/0	0/0; 2/0

The split between the two causes is measured, not assumed. Every change in Pension Credit, and the UC "mixed-age saving" row, comes from benefit units whose has_mixed_age_couple_pension_credit_saving changed. Every one of those lost the saving; none gained it.

• The child cutoffs: Universal Credit.
  • In 2024-25 and 2025-26 the effect is the child limit: about half the straddling cohort is now transitionally protected.
  • From 2026-27 the limit is gone (child_count is infinite from 6 April 2026). The remaining effect is the higher first child element (reg 43) for families whose eldest child is in that half.
  • Child Tax Credit doesn't change: tax credits end on 6 April 2025 in the model, and 2024-25 CTC in the data is £0.07bn.
  • The Pension Credit eldest-child amount moves no award: every Pension Credit change is in a benefit unit whose mixed-age saving changed. About 2.9-3.0k Pension Credit benefit units have a child or qualifying young person in 2024-2030, and 0.54-0.56k have someone under 18.
• The mixed-age saving.

  • main compared birth_year (period less age) with 1954. In data that admits an older member aged period - 1954, born from 7 October 1953 to 6 October 1954.

  • Only births on or before 5 February 1954 reached the qualifying age by 14 May 2019, so the fix removes births from 6 February to 6 October 1954.

  • In 2025-26, 6.56k benefit units (5 survey records) lose the saving: 3.67k lose a Pension Credit payment, 5.77k gain a UC payment (3.25k do both), and one record's payments don't change. The net effect is -£0.05bn Pension Credit and +£0.06bn UC.

  • These couples also lose Housing Benefit, so their households lose net income overall. Measured over the benefit units whose saving changed (Pension Credit, UC and Housing Benefit), with Council Tax Reduction and household net income taken at household level for their households:

    	2024-25	2025-26	2026-27	2027-28	2028-29	2029-30
    Pension Credit, £m	+0.0	-51.6	-0.1	-17.6	-0.0	+0.0
    UC, £m	+3.7	+62.0	+0.0	+4.0	+0.0	+2.3
    Housing Benefit, £m	-11.6	-17.8	+0.0	-6.4	+0.0	-2.8
    Council Tax Reduction, £m	+0.0	+0.0	+0.0	-1.3	+0.0	+0.0
    Household net income, £m	-7.9	-7.4	-0.1	-21.3	-0.0	-0.5

    This is why household net income rises by less than UC in 2024-25 and 2027-28. These are also the households entering absolute poverty in the table above. Every household that enters is one of them: 2 before housing costs in 2024-25, and 2 before and 2 after housing costs in 2027-28. This was matched by household id in saving_hh_aggregate.py.

  • Without the date fix, deriving birth_year from the exact date would have pushed this rule the other way. Births from October to December 1954 entered it, which a pre-fix run showed as one 10.6k-weight benefit unit gaining £83m of Pension Credit in 2027-28.

• The pension-age rules are untouched without a date input. Against main, for every record in every year 2024-2030, the following are identical: state_pension_age, months_since_state_pension_age, is_SP_age, state_pension_credit_qualifying_age, has_attained_state_pension_credit_qualifying_age, meets_savings_credit_age_requirement and state_pension. The data carries no date_of_birth, so routing them through the shared birth instant changes nothing there.
• Ordering by date of birth changes Universal Credit by at most £0.003bn in any year, and no poverty count. This was measured against Set State Pension age from date of birth, including the rise to 67 #1899's head with the rankings reverted to whole age.
• Poverty counts move in whole survey households, one to eleven a year, so the year-to-year pattern is lumpy.
  • Relative poverty shows the same. Its line, 60% of the person-weighted median, is identical before housing costs in every year. After housing costs it is identical except in 2026-27, where it rises £1.
  • Before housing costs: -173k and -101k people in 2024-25 and 2025-26, then between -79k and 0.
  • After housing costs: -192k and -114k, then between -77k and 0.

The scripts, CSVs and household-crossing counts are in ~/PolicyEngine/_pr_worktrees/dob-cutoffs-axiom/impact/: v3_main.csv, v3_head.csv, v3_summary.txt, v3_*_flips.npz, v3_saving_*.npz, v3_pa_*.npz and v3_lines_rel_*.txt, v3_saving_hh2.txt and, for the UC recipient-children row, v3_hh_*.npz and v3_saving_hh.txt (whose household columns are withdrawn), from run_one_v2.py, flips.py, saving_diag.py, saving_hh_diag2.py with saving_hh_aggregate.py, pension_age_diag.py and lines_rel.py. A PROVENANCE.md there records the commits and the dataset blob. Every figure is from a real run; none is scaled or interpolated. Earlier runs against #1899's head and against main at 325d585e5 gave the same child-cutoff figures to within £0.01bn.

Caveats

• One record over several years. Ages are held fixed across years in the data, so a record's date of birth moves a year later each period. Transitional protection is therefore a property of each period's cohort, not of a child followed over time.
• Siblings of the same age. Set State Pension age from date of birth, including the rise to 67 #1899 places each person in their year of age independently, by weight within age and sex. Twins therefore get different dates of birth and can fall either side of the cutoff. The multiple-birth exception (UC Regs 2013 reg 24A(1)(b)(ii) and Sch 12 para 2) isn't modelled either way.
• CTC family element. Tax Credits Act 2002 s.9(2)(a) limits the CTC family element to claimants responsible for a child born before 6 April 2017, but CTC_family_element pays it to every CTC claimant. That is a separate, pre-existing gap for 2017-18 to 2024-25, and it's queued as a follow-up; the Axiom side is in Encode TCA 2002 s.9(2)(a), (3A), (3B) and CTC Regs 2002 regs 7(2)–(2A) and 9 (family element and child limit by date of birth, 2017-18 to 2024-25) TheAxiomFoundation/rulespec-uk#372.
• Other exceptions. The other statutory exceptions to the limits (UC Regs 2013 Sch 12: adoption, non-parental caring arrangements, non-consensual conception; CTC Regs 2002 regs 10 to 14) aren't modelled.

Removed parameters

gov.dwp.universal_credit.elements.child.limit.start_year and gov.dwp.tax_credits.child_tax_credit.limit.start_year held the year 2017, and gov.dwp.pension_credit.mixed_age_couples.saving_latest_birth_year held 1954 (added by #1940). They are replaced by the dates born_before (20170406) and saving_latest_birth_date (19540205), and the removed fragment says how to write a reform now. In the local checkouts of policyengine.py, policyengine-api, policyengine-api-v2, policyengine-app-v2, microcosm and policyengine-uk-data, the only other mention is policyengine-app-v2's mock metadata (app/src/mocks/UK_Metadata.json). A saved policy that changes any of them will stop loading, rather than silently doing nothing. The fragment type is removed (a minor release), following #1899's removal of the State Pension age year parameters and this repo's github-prs.md; downstream packages pin exact versions.

Independent review

An Opus 5.5 reviewer (Subfleet, review/standard) requested changes. Each finding and what was done:

Finding	Severity	Resolution
A date_of_birth set for one person gave the others 0, so a 35-year-old parent got State Pension age −1925 and counted as over it; unset siblings ranked first and counted as born before the cutoff	blocking	Reproduced. utils.dates.birth_instant now resolves every rule's birth instant: the input where given, otherwise from age and months_since_last_birthday. There are YAML regressions and a Hypothesis property.
A date input inconsistent with the age grid gave a State Pension age measured from two births (5 January 1960 gave 65.75)	should-fix	Reproduced. state_pension_age and months_since_state_pension_age now share one instant, which gives 66, attained 5 January 2026.
"Household situations unchanged" does not hold for fractional ages	should-fix	Now scoped to whole ages in the body and changelog. The UC flag was added to the household differential test.
Poverty-crossing and poverty-line figures had no saved output	should-fix	Saved, with provenance.
removed vs breaking	should-fix	Kept removed, following #1899's precedent (see above).
Invariant 4's year list; reference independence; revoked regs 24A and 24B	nit	Corrected here. The code comment and rulespec-uk#371 now note that reg 24B was revoked from 6 April 2026.
The UC age_exemption reform never lifts the limit	pre-existing	Split into a separate follow-up.

A round-2 review approved, on the condition that the tests pass (they do; see "Tests run"). Its one should-fix: a date_of_birth input was never checked, so 20170229 became 1 March, and 2017 was read as a date in year 1, which put a child over State Pension age. It also wasn't checked against age. Such inputs now raise a ValueError (test_an_invalid_date_of_birth_is_rejected and test_a_date_of_birth_that_contradicts_age_is_rejected). Its wording nits, the confounded before/after comparison and the BHC/AHC mix-up in the poverty sentence, are corrected above.

A final review by GPT-6.1 Sol (Subfleet, review/hard) approved 86523cf3f after merging main and the mixed-age change. It re-ran all the suites, 280,000 no-input comparisons against main, and 73,049 calendar round trips, and confirmed the 5 February 1954 cutoff is inclusive for both sexes. Its should-fixes, all done in c6db31973:

• reject negative date_of_birth inputs;
• scope the "unchanged" claim to the child rules;
• note that reg 24A(3) only took its date wording from 1 February 2019.

Its figure nits (rounding, poverty ranges, the Pension Credit population, the saving breakdown, and a relative-line check that had used the absolute line) are corrected above from new runs.

A Sol delta review of c6db31973 approved, with three documentation nits:

• AHC rounding;
• the provenance of the relative lines;
• the UC reference title's description of reg 40.

All three are fixed, the last in 297a210f8. Main then moved 255 commits, so 297a210f8 merges it and routes #1907's new pension-age helpers through the shared birth instant (see the summary). That merge is what the next review covers.

The Subfleet daemon hung, so that review ran in-session: two independent Opus 5.5 reviewers, each with its own lens.

• Engine and merge lens: APPROVE. It found:

  • 0 differences against main across 102 arrays, covering six pension-age variables for 4,228 people over 2010-2046, with whole and fractional ages, edge months and both sexes;
  • 0 mismatches against the statute through the date-input path, for 10,186 people over 13 years;
  • all targeted tests passing.

  Its nits are fixed in 20e67683d: the unused age-only helper is removed, the input-path test is extended, and the State Pension docs cover a direct date_of_birth.

• Claims lens: REQUEST_CHANGES, on this description only. Its points are corrected above from my own re-runs:

  • Housing Benefit and net income for the mixed-age couples;
  • the UC child row counted only among recipients;
  • main's 1954 rule, which covered births from 7 October 1953 to 6 October 1954;
  • the birth_year readers and changelog wording.

A final in-session review of 20e67683d ran with the same two lenses.

• Code lens: the property test barely exercised SPC attainment on the input path, and the invariant text was stale. Fixed in dc2ae4e22, as above, along with the Class 4 docs sentence.
• Description lens: the 2025-26 household net income for the mixed-age households was wrong (-£12.6m). My diagnostic had mapped household values to benefit units with map_to, which splits a household's value across its members. Recomputed at household level, it is -£7.4m. The poverty entrants are now explained, and the provenance is completed.

Tests run

At head dc2ae4e22; the head commit dcf391214 only corrects a test comment (the 1960 example's qualifying age, 66 years and 8 months):

• YAML: policyengine-core test policyengine_uk/tests/policy -c policyengine_uk, every policy YAML: 2,419 passed.
• pytest: 85 passed across test_date_of_birth.py, test_state_pension_age.py, test_state_pension_credit_qualifying_age.py, test_mixed_age_couple_properties.py and test_pension_age_severe_disability_premium_properties.py.
• Mutation check: making has_attained_state_pension_credit_qualifying_age ignore a date input fails both pinned examples.
• Not run locally: the full pytest suite, which CI runs.

axiom: uk/regulation/uksi/2002/1792/schedule/IIA encoded-correct (takes the eldest child's born-before-6-April-2017 status as an input judgment; companion tests post_2017_child_has_no_eldest_child_increase and mixed_children_with_pre_2017_disabled_and_severely_disabled_amounts) | TheAxiomFoundation/rulespec-uk#371 queued (UC Regs reg 24A(3) and 24B, SI 2014/1230 reg 43, reg 36 first-child row) | TheAxiomFoundation/rulespec-uk#372 queued (TCA 2002 s.9(2)(a), (3A), (3B); CTC Regs regs 7(2A) and 9) | TheAxiomFoundation/rulespec-uk#370 queued (SI 2019/37 art. 4(1) saving, extended with the 5 February 1954 condition)

🤖 Generated with Claude Code

[MaxGhenis]

Coordination note from #2051 (Child Tax Credit family element, #2050), which adds a fourth reader of the 6 April 2017 cutoff on main.

• CTC_family_element now applies TCA 2002 s.9(2)(a) / CTC Regs reg 7(2)(a): the family element is included only where a child or qualifying young person was born before 6 April 2017. The cutoff is a new YYYYMMDD parameter, gov.dwp.tax_credits.child_tax_credit.eligibility.family_element_born_before (20170406, unit: date), in the same format as this PR's born_before parameters.
• On main the formula compares birth_year * 10_000 + 406 (6 April of the birth year) with it. That equals birth_year < 2017, the reading is_CTC_child_limit_exempt uses. When this PR rebases onto main after Hold the Child Tax Credit family element at £545 and apply the 6 April 2017 birth-date condition #2051, the conversion is one line: birth_day(person, period) < ctc.eligibility.family_element_born_before. Its whole-age invariant (born before 6 April 2017 iff period - age < 2017) would then cover this reader too.
• Hold the Child Tax Credit family element at £545 and apply the 6 April 2017 birth-date condition #2051 also treats a birth_year of 0 as not given and falls back to the period less age. A situation that enters birth_year for some people leaves the others at 0, and main's three existing cutoff readers count them as born before 2017. This PR's helper avoids that for date_of_birth; the fallback in Hold the Child Tax Credit family element at £545 and apply the 6 April 2017 birth-date condition #2051 goes away with the conversion.

[MaxGhenis]

Follow-up from #2051's review: when CTC_family_element moves to birth_day, three of #2051's YAML cases (policyengine_uk/tests/policy/baseline/gov/dwp/tax_credits/ctc_family_element.yaml) need date_of_birth inputs instead of birth_year, because the helper ignores entered birth years:

• "a child aged 7 with an entered birth year of 2016";
• the two "birth year entered for one child only" cases.

#2051's birth_year > 0 fallback in the formula also goes away. Its property test generates entered birth years too (birth_year_offset in test_ctc_family_element_properties.py), so that generator would switch to date_of_birth.

[MaxGhenis]

Merging on the gates at head dcf391214.

CI and mergeability. gh pr checks exits 0 with all 7 checks passing (the Test job took 42m17s). The PR is MERGEABLE, its merge state is CLEAN, and it is up to date with main (8090f1c19).

Independent reviews. These are not GitHub reviews; their outputs are in ~/PolicyEngine/_pr_worktrees/dob-cutoffs-axiom/ and the session's workflow journals.

Round	Reviewer	Head	Verdict
r1	Opus 5.5 (Subfleet)	42b358e2	request changes (partial-input bug), fixed
r2	Opus 5.5	ef35551d	approve
r3	GPT-6.1 Sol	86523cf3f	approve
r4	GPT-6.1 Sol	c6db31973	approve
r5	two-lens in-session Opus	297a210f8, the main merge with #1907	engine approved; description findings fixed
r6	two-lens in-session Opus	20e67683d	findings fixed
r7	description lens	dc2ae4e22	approve
final	in-session Opus	dcf391214	approve

What the final check executed

• Women's qualifying-age assertion: no counterexample over 878,455 person-years.
• Date path against months path: a maximum difference of 0.0 over 1.76 million combinations.
• Mutants: both caught by the pinned examples.
• Tests: passed.

Why this needs no decision from Max. It corrects the model to the cited law. Per Max (2026-10-02), bug fixes that move aggregates merge on gates; the impact is stated in the PR body.

Axiom.

• Sch IIA is already encoded correctly.
• Encode UC (Transitional Provisions) Regs 2014 reg 43, UC Regs reg 24A (6 Apr 2017 – 5 Apr 2026) and reg 24B; condition the reg 36 first-child amount on a child born before 6 April 2017 TheAxiomFoundation/rulespec-uk#371, More consistent phrasing in descriptions #372 and Frequent flyer levy #370 are queued.