Stop deciding a household's UC health element rate by its position

[MaxGhenis]

Fixes #2250

Stacked on #2248, the protected LCWRA amount; the diff below is relative to it. Retarget to main once #2248 merges.

What was wrong

add_universal_credit_reform gave the new-claimant LCWRA rate by a seeded draw over every benefit unit in a simulation, household situations included:

• Position decided the award. A household's award depended on its position among the other benefit units. On main, 40 identical single LCWRA claimants in one situation get £217.26 at positions 1, 2, 15, 25 and 32, and the protected amount elsewhere.
• No way to state it. No situation could state that a claimant is new, so a lone household always got the protected amount.

The law

UC Regs 2013 reg. 36 gives the new-claimant rate to claimants other than a pre-2026, severe conditions criteria or terminally ill claimant. That is a fact about the claimant, which a household calculation should be able to state.

Changes

• uc_receives_new_claimant_health_element (new, benefit unit, bool, default false).
• uc_reform.py decides the status per year in _new_claimant_health_element_status, reading the simulation's own branch (sim.get_array):
  • A value the situation or dataset supplies is used as given.
  • Data that supplies none keeps the seeded draw, the same draw as before, now recorded as an input.
  • A situation that supplies none gets the protected amount (false).

Invariants (tested)

• Position independence. In a situation, a household's health element does not depend on how many other benefit units the simulation holds or where it sits among them. 40 identical households all get what a lone household gets.
• Supplied status wins. A supplied status is used as given, in situations and in data.
• Microsimulation unchanged. On data that does not supply the status, the seeded draw is unchanged. Estimated on the local enhanced FRS 2023-24, 2026-27 LCWRA is £23.3207bn and UC £62.1568bn, identical to Pay pre-2026 UC health element claimants the legislated protected amount #2248; 2027-28 matches as well.

Tests

test_uc_rebalancing.py:

• Position. test_household_award_does_not_depend_on_its_position (40 households). On main, the same check finds 5 of 40 at £217.26.
• Stated status. test_new_claimants_use_fixed_health_element now states the status rather than forcing the draw.
• Data. test_data_without_a_status_keeps_the_seeded_draw and test_data_that_supplies_the_status_overrides_the_draw use in-memory datasets.
• Branches. test_a_branch_can_state_its_own_status: a branch's status, and a nested branch's inherited status, give the new-claimant rate while the parent is unchanged. It fails on the default-branch lookup.

Local runs:

• test_uc_rebalancing.py, code_health/test_cached_arrays_not_written_in_place.py and test_income_support_esa_entered_directly.py: 71 passed.
• UC YAML policy tests: 414 passed.
• Ruff check and format: clean.

Impact

None in microsimulation. Household situations without the status now always get the protected amount; previously about 11% of benefit units got the new-claimant rate in 2026, depending on position.

axiom: n/a: input plumbing, and no rule or amount changes. rulespec-uk uk/regulations/uksi/2013/376/36.yaml already takes the claimant's pre-2026 status as an input (claimant_is_pre_commencement_lcwra_claimant).

🤖 Generated with Claude Code

[MaxGhenis]

Merging on the gates.

• Head: e3a1baf3, which merges main (with Pay pre-2026 UC health element claimants the legislated protected amount #2248 merged as 9cddf051) into the reviewed af865c6c. The PR was stacked on Pay pre-2026 UC health element claimants the legislated protected amount #2248 and is now based on main. Its added and removed lines against main are identical to the reviewed change set (186 lines, same content).
• Checks: all green on this head, including Test.
• Mergeable: MERGEABLE, no change request. Marked ready from draft.
• Independent review: GPT-6.1 Sol, two rounds.
  • Round 1 requested changes: Holder.get_array ignored branches, so the PR now uses sim.get_array.
  • Round 2 approved af865c6c.
• Local check on this head: test_uc_rebalancing.py 18 passed.

🤖 Generated with Claude Code