[juaristi22]

Fixes #1822.

The dividend scale held a £37,500 basic-rate limit until 2026 and a £150,000 additional-rate threshold until 2026. This adds the missing statutory dates so historical dividend tax uses the corresponding UK bands under Income Tax Act 2007 s. 13.

Tax year beginning 6 April	Limit added	Source
2016	£32,000 basic-rate limit	Finance (No. 2) Act 2015 s. 6
2017	£33,500 basic-rate limit	Finance Act 2016 s. 2
2018	£34,500 basic-rate limit	SI 2017/1184 art 2
2019	£37,500 basic-rate limit restored	Finance Act 2019 s. 5
2021	£37,700 basic-rate limit	SI 2021/111 art 2
2023	£125,140 additional-rate threshold	Finance Act 2023 s. 6

For one adult aged 40 in England, the genuine-income regressions give:

Year and annual income	Previous dividend tax	Updated dividend tax
2018, £40,000 pay and £10,000 dividends	£762.50	£1,512.50
2025, £49,000 pay and £2,000 dividends	£363.75	£313.75
2025, £130,000 pay and £10,000 dividends	£3,206.25	£3,738.25

The dividend nil-rate amount occupies band capacity. For the £49,000/£2,000 example, the Personal Allowance leaves £36,430 of taxable wages and £1,270 of basic-band capacity: £770 × 8.75% + £730 × 33.75% = £313.75. Tests also assert final income_tax using independently derived expectations.

All existing rates, original and 2026 dated entries, the optional extra bracket, formulas, Gift Aid extensions and reform interfaces remain unchanged. The README records the historical dates and scope limits. This does not synchronize post-2026 projected thresholds or certify the pre-2016 dividend-tax-credit regime.

The inherited 2016 Personal Allowance discrepancy remains separate in #2359 / #2442. Existing fixed-order allowance allocation also does not always implement the most-beneficial split for mixed income. The new 2016, 2017, 2021 and 2026 final-tax cases therefore use £160,000 of dividends with a naturally zero allowance. The selected 2018, 2019 and 2025 mixed cases have independently checked allocations. No intermediate taxable-income or allowance input is forced.

Validation: final fixtures recorded 19 failures / 6 passes with the unchanged base dividend parameter, then 111 passed with the corrected parameter (25 new cases and 86 existing regressions; 24.30s pytest / 40.580s bounded invocation). The combined run covered dividend bases/allowances, Gift Aid, property band use, final income tax, fiscal conversion, generated-parameter processing and two dividend Scenario compatibility controls. Per-case outcomes and exact hashes were recorded from collection. Earlier fixture-oracle premises were corrected before the final paired run; those earlier logs remain separate and are not additional unique coverage. Formatting, Ruff and diff checks passed.

Local Python 3.13.14 / Core 3.32.27 differs from locked Core 3.33.2. Full locked CI, the documentation build and private population analysis were not run locally. No population impact is claimed.

Reproducer: policyengine_uk/tests/test_dividend_band_dates.py::test_raw_dividend_limits_change_on_the_statutory_tax_year_date

Reproducer: policyengine_uk/tests/policy/baseline/gov/hmrc/income_tax/dividend_band_dates.yaml::2025 dividends with wages 49000 and dividends 2000

axiom: needed

Maintainer follow-up: verify RuleSpec/Axiom parity for the six dated dividend band limits. No encoder or parity verification has been run. Draft #1963 documents the lag as an inherited limitation; its owner can integrate this separate fix.

[vahid-ahmadi]

Looks right, fine to approve. One follow-up is worth filing.

@juaristi22, the dated limits now match rates/uk.yaml and the cited Acts and Orders for 2016-17 to 2025-26. The £125,140 additional-rate threshold now applies from 6 April 2023.

1. should, policyengine_uk/parameters/gov/hmrc/income_tax/rates/dividends.yaml:57: the same lag comes back from 2031-32. uk[1].threshold is now indexed by create_threshold_indexation.py, but the dividend scale is not. In the processed parameters I get £38,500 against £37,700 in 2031, and £41,900 against £37,700 in 2035. ITA 2007 s13 ties dividends to the same basic rate limit, so this is the bug in #1822 again, for future years. It is outside this PR's stated scope. It needs a follow-up issue (add the dividend limits to the indexation, or derive them from uk), or a line in this PR saying it is deliberately left.
2. nit, dividends.yaml:57 and :111: 2015-16 still reads £37,500 and £150,000, while uk.yaml gives £32,000 for that year. The README says the pre-2016 tax-credit regime is out of scope, so this is fine for now. A note in the issue would help.
3. nit, policyengine_uk/parameters/gov/hmrc/income_tax/README.md: policyengine-core uses the text after a folder README's first line as that node's description. The whole gov.hmrc.income_tax node would then be described by dividend-band notes. A ## Dividend bands section in a README for rates/, or a shorter description, would fit better.

What I checked

• Read the issue, the PR body and the full diff (including test_dividend_band_dates.py and the YAML file), and compared each new date with uk.yaml and the cited instruments.
• Worked two cases by hand. 2018, £40,000 pay and £10,000 dividends: 28,150 taxable pay leaves 6,350 of basic band; 2,000 nil-rate, then 4,350 x 7.5% + 3,650 x 32.5% = £1,512.50. 2025, £130,000 pay and £10,000 dividends: no personal allowance, all dividends above £125,140; 9,500 x 39.35% = £3,738.25. Both match. A case the PR body doesn't work through, 2017 with £160,000 of dividends: 28,500 x 7.5% + 116,500 x 32.5% + 10,000 x 38.1% = £43,810. This also matches.
• Ran policyengine-core test .../dividend_band_dates.yaml: 9 passed. pytest test_dividend_band_dates.py: 16 passed. With dividends.yaml reverted: 7 YAML cases and 12 pytest cases fail.
• ruff check and ruff format --check on the new test file: clean.
• Hosted CI on 83de1ef is green.
• Process: the changelog fragment is .fixed.md; the Reproducer and axiom lines are present; the Fix evidence check is green.
• Not checked: RuleSpec parity, and population effects.

🤖 Generated with Claude Code

[vahid-ahmadi]

Approving, per my review above. The shoulds and nits can follow.