Add UK Chronicle targets for issue 196

[juaristi22]

Fixes #196

Summary

• Add the HMRC CGT size-of-gain 2025 source package with nine TY2023-24 individual gain bands and both taxpayer-count and gains facts.
• Add the ONS households-by-type country 2025 source package for the Scotland 2025 couple-households-with-3+-dependent-children fact.
• Register both packages in the UK bundle and source package aliases, and allow consumer target profiles to select rows by dimensions key/value mappings for Microcosm.
• Update the UK source checklist with the exact UK national gaps: CGT gain bands, housing cost aggregates, Scotland household composition, salary-sacrifice contributions, benefit cap reduction #196 ported facts and the computed/no-publisher-measure exclusions.

Verification

• UV_CACHE_DIR=/Users/mariajuaristi/Documents/Codex/2026-08-24/can-x20/work/.uv-cache uv run --with ruff ruff format --check chronicle/source_package.py chronicle/bundle.py policyengine_chronicle/consumer.py tests/test_chronicle_consumer.py tests/test_chronicle_source_package.py
• UV_CACHE_DIR=/Users/mariajuaristi/Documents/Codex/2026-08-24/can-x20/work/.uv-cache uv run chronicle validate-package hmrc-cgt-size-of-gain-2025 --year 2025
• UV_CACHE_DIR=/Users/mariajuaristi/Documents/Codex/2026-08-24/can-x20/work/.uv-cache uv run chronicle validate-package ons-households-by-type-country-2025 --year 2025
• UV_CACHE_DIR=/Users/mariajuaristi/Documents/Codex/2026-08-24/can-x20/work/.uv-cache uv run pytest tests/test_chronicle_source_package.py -k "cgt_size_of_gain or households_by_type_country"
• UV_CACHE_DIR=/Users/mariajuaristi/Documents/Codex/2026-08-24/can-x20/work/.uv-cache uv run pytest tests/test_chronicle_consumer.py -k "dimensions_selector or dimension_value"
• UV_CACHE_DIR=/Users/mariajuaristi/Documents/Codex/2026-08-24/can-x20/work/.uv-cache uv run pytest tests/test_source_package_alias_drift.py
• UV_CACHE_DIR=/Users/mariajuaristi/Documents/Codex/2026-08-24/can-x20/work/.uv-cache uv run chronicle build-bundle --source hmrc-cgt-size-of-gain-2025 --source ons-households-by-type-country-2025 --out /tmp/chronicle-196-bundle --replace
• UV_CACHE_DIR=/Users/mariajuaristi/Documents/Codex/2026-08-24/can-x20/work/.uv-cache uv run chronicle build-consumer-artifact --facts /tmp/chronicle-196-bundle --out /tmp/chronicle-196-artifact --replace

[vahid-ahmadi]

Automated review pass (Claude Code, high effort, diff only — no execution). Three findings, one of which binds a fact to the wrong series.

1. packages/ons/households_by_type_country_2025/source_package.yaml:510 — source_concept points at the wrong package

source_concept: ons.families_households_table7 looks copy-pasted from the separate ons-families-households-2025 package (UK-wide Table 7 of Families and households), while every other field on this measure points at the regions/countries workbook.

That matters because consumer selection resolves source_concept, and concept falls back to source_concept when there is no canonical alignment (policyengine_chronicle/consumer.py:578-582). So this single Scotland fact is silently folded into the UK Table 7 series: a Microcosm selector on ons.families_households_table7 picks up an extra Scotland-only row, while the selector documented for this package (ons.households_by_type, docs line 146) resolves only via concept and never reaches it. Nothing fails loudly — the fact just binds to the wrong series, which downstream reads as a UK figure.

2. policyengine_chronicle/consumer.py:555-560 — the new dimensions Mapping branch matches subsets, not sets

The branch does subset matching (all(dimensions.get(name) == expected ...)), whereas every other dimension selector in this function is exact-set identity. A selector like {"cgt_gain_band": "gain_12300_to_24999"} therefore also matches any row dimensioned by that band and a further dimension, and an empty {} matches every row in the candidate set. The extra rows aren't rejected — they flow into the period/assertion choice downstream, so the selector silently resolves against a larger candidate set than it names.

Related: a mis-typed dimension name yields zero matches rather than an error. That is the same silent-dead-target failure mode as the microcosm children_count batch — 8 references bound to a non-existent variable and simply never compiled. Given this layer feeds those targets, failing loudly on an unknown dimension name seems worth the strictness.

3. policyengine_chronicle/consumer.py:561-565 — duplicated list branch, and a now-dead _selector_value case

The list branch inside the new dimensions block duplicates the pre-existing list handling in _selector_matches — same code, same comment verbatim — and that path is fed by _selector_value's if key == "dimensions": return sorted(...) at 590-591, which is now unreachable for the only key that used it. The diff leaves a dead branch plus a duplicated comparison that has to be kept in sync with its twin.

---

Checked and clean, given the recent history in this lane: the CGT band edges are contiguous and gap-free from £12,300 up to the open-ended £5,000,000+ top band, with no overlaps; both manifests' source_url, filename, R2 key and sha256 agree with each other, so no repeat of the #188 doubled-path defect; no duplicate package or record_set_spec_id values; and the remainder of the consumer.py diff is line-reflow.

[juaristi22]

Addressed Vahid review in 329b757.

Fixed:

• Gave the ONS country household package its own source-specific source_concept: ons.households_by_type_regions_countries_table7.
• Updated the ONS package smoke test and UK checklist so Microcosm consumers can select the Scotland row by source_concept without folding it into the UK Table 7 series.
• Made dimensions mapping selectors exact row matches, rejected empty mappings, and added explicit errors for unknown mapped dimension names.
• Removed the duplicate dimensions-list selector branch while preserving list selectors as exact dimension-name-set matches.

Verification:

• uv run --with ruff ruff format ...
• uv run --with ruff ruff check ...
• uv run --locked pytest -q tests/test_chronicle_consumer.py
• focused ONS/consumer review regression tests
• uv run --locked pytest -q tests/test_chronicle_bundle.py::test_build_bundle_writes_merged_consumer_contract
• Both GitHub Chronicle checks passed.