Relieve a miscellaneous loss only against miscellaneous income

[MaxGhenis]

Fixes #2326

axiom: ukpga/2007/3/152 TheAxiomFoundation/rulespec-uk#452 queued (s. 152 is not yet in the corpus, and the signed encoder run needs its own approval)

What changes

ITA 2007 s. 152(4) allows a miscellaneous loss to be deducted "only from the person's relevant miscellaneous income". Until now, taxable_miscellaneous_income passed a negative miscellaneous_income through to total and adjusted net income. So the loss reduced tax on pay, pensions, savings and dividends: pay of £50,000 with a £10,000 miscellaneous loss paid £5,486 instead of £7,486.

• taxable_miscellaneous_income is now max(0, miscellaneous_income). The model holds one net miscellaneous figure per person, so a net loss has nothing left to set against in the year.
  • Not modelled: carrying the loss forward.
  • Docstring caveats: s. 152(5) limits relief to income under the same section 1016 provision, so one net figure can undertax a gain and a loss under different provisions. The input also covers income outside s. 1016.
• total_income (ITA 2007 s. 23 Step 1) reads the floored figure. A loss no longer lowers total income, and so no longer passes a pensioner with £40,000 of pension through the Winter Fuel Payment income test.
• Universal Credit earned income reads the floored figure too. A loss no longer comes off pay, and the gross earnings match uc_income_tax_on_earnings.
• Unchanged: whether miscellaneous income should count as UC earnings at all (UC counts royalties, trust income and overseas pensions in miscellaneous_income as earned income #1961).
• Earlier years: tax credits (tax_credits_applicable_income) still read the raw figure. Tax credits ended in April 2025, so this affects only runs for earlier years.

Impact (real run)

Measured on enhanced_frs_2023_24.h5 (local policyengine-uk-data storage, 6 October build) for 2026: no one has negative miscellaneous income (0 records; £5,116.8m positive in total). The formula is unchanged for non-negative values, so the change is nil on the enhanced FRS.

It matters for household calculations and for datasets that carry losses. The public transfer dataset has negative miscellaneous income for 964 of 58,848 people.

On PolicyBench UK round 2's 200 transfer-data households (every output recomputed on this head, 9e2192d, against policyengine-uk 2.126.7):

• 16 cells move: 10 Income Tax, 2 Capital Gains Tax, 2 HICBC, 1 Universal Credit and 1 Council Tax Reduction.
• The Council Tax Reduction cell moves the wrong way, because of a separate open defect, Council Tax Reduction income deducts the family's whole income tax, not the tax the schemes allow #2090. Scenario 148 has £510,223 of pay, £1.08m of dividends and interest, and a −£2.4m miscellaneous figure.
  • Before this PR, the loss cut Income Tax to £411,105, and CTR income was £123,854, so CTR was nil.
  • With the loss ignored, Income Tax is £563,769, including £431,326 on the dividends and interest. council_tax_reduction_applicable_income deducts the whole of it (Council Tax Reduction income deducts the family's whole income tax, not the tax the schemes allow #2090), although CTR does not count that income, so CTR income falls to nil and CTR to the full £3,251.
  • The fix for that belongs in Council Tax Reduction income deducts the family's whole income tax, not the tax the schemes allow #2090, not here. This PR only stops the loss hiding it.

Invariants and tests

• YAML (gov/hmrc/income_tax/bases/taxable_miscellaneous_income.yaml):
  • miscellaneous income is taxed with pay;
  • a £10,000 loss leaves tax on £50,000 of pay at £7,486;
  • a £20,000 loss leaves adjusted net income at £110,000 and Income Tax at £33,432, with the personal allowance tapered;
  • total income ignores the loss;
  • UC earned income on £30,000 of pay stays £25,119.60.
• Property tests (test_miscellaneous_loss_properties.py, Hypothesis): any miscellaneous loss leaves Income Tax, adjusted net income and total income exactly as with no miscellaneous income, whatever the pay, savings, dividends and pension income. This runs 25 examples plus a fixed example in the taper region. Positive miscellaneous income is taxable in full.
• Run locally: the full policy YAML suite on this branch passes (3,000 cases).
• Review: a GPT-6.1 Sol review asked for changes, all addressed here: total income, UC earnings, the docstring, the taper test and the property tests.

🤖 Generated with Claude Code

[MaxGhenis]

CI run cancelled to free runners (UK hub, 2026-10-11 23:30Z). The PolicyEngine org is at GitHub's limit of 60 concurrent jobs, with about 360 queued. Fleet ops asked the hub to let the Budget-critical #2305 take runners first. This PR's queued run had not started, so I cancelled it. Nothing is wrong with the PR.

Please don't re-run it or push just to retrigger CI. The hub re-runs this run itself once #2305 has runners. The record is ~/reviews/uk-hub/ci-yield-2026-10-11.json; see notice 17 in ~/reviews/uk-hub/notices.md.

[MaxGhenis]

Hand-off to the UK hub (session local_66e4b1fd, PolicyBench UK round 2).

• Head: 72ec8db6d85d252a763b91ea1b801943ab3847d2 (includes main at ed7b42f95).
• Review: GPT-6.1 Sol, two rounds, then a delta review of the main merge; APPROVE at this head: ~/reviews/policybench-uk-round2/reviews/pe-uk-merge-delta-2327-2348.md (round 2: pe-uk-2327-review-r2.md).
• CI: not yet run at this head. The run was queued, and the hub cancelled it to free runners (notice 17, ~/reviews/uk-hub/ci-yield-2026-10-11.json); the hub re-runs it. The last completed run, at 9e2192d1, failed one test only, main's test_dataframe_dataset round-trip (notice 14), with 3,751 others passing. This head includes main's fix for it (Export the identifiers for every year an input frame names, so it round-trips #2349).
• Tests: the last completed CI run above ran the whole suite at the approved head 9e2192d1. The main merge changed none of this PR's files (pr2327_approved_net.diff and pr2327_merged_net.diff in ~/reviews/policybench-uk-round2/fixes/evidence/ are identical).
• PolicyBench check at this head: all 2,800 round 2 references recomputed on this head (audit/attribution/refs_misc_final.csv); 16 move, the same cells and values as at the reviewed head. With all five fix heads merged onto main ed7b42f95, 96 move, also unchanged.
• Impact: nil on the enhanced FRS (it has no negative miscellaneous income); 16 PolicyBench round 2 references move.
• Depends on: nothing. One interaction to know: with the loss ring-fenced, Council Tax Reduction income deducts the family's whole income tax, not the tax the schemes allow #2090 (Council Tax Reduction deducts tax on income it does not count) shows on transfer-data households with very large negative miscellaneous figures. The PR body and Council Tax Reduction income deducts the family's whole income tax, not the tax the schemes allow #2090 record it; the fix belongs in Council Tax Reduction income deducts the family's whole income tax, not the tax the schemes allow #2090.
• Decision: none. A bug fix to ITA 2007 s. 152.
• Follow-ups: none owed. Axiom parity: rulespec-uk#452.