Add UK bus-spending imputation plan and DfT calibration targets

[vahid-ahmadi]

Problem

The published Populace UK population miscalibrates both bus consumption variables (verified against the incumbent enhanced-FRS build and DfT Annual Bus Statistics, at the same simulated year):

variable	Populace UK 2023	enhanced-FRS	DfT anchor
bus_fare_spending	£7.4bn raw, 17.7% of hh, £1,442/spender	£3.9bn, 27.4%, £456	~£4.0bn
bus_subsidy_spending	£1.0bn, 18.2%	£3.5bn, 33.1%	~£3.5bn

Fare is ~2× too high and over-concentrated; subsidy is ~3.6× too low — exactly the failure mode that occurs when the survey imputation isn't anchored to DfT. The incumbent build applies a DfT calibration (calibrate_bus_fare_spending / calibrate_bus_subsidy_spending); that step was never ported to the Populace UK build.

What this adds (mirrors build/us)

• build/uk/bus_calibration_targets.py — a TargetRegistry with DfT-anchored household-sum targets for bus_fare_spending (~£4.0bn) and bus_subsidy_spending (~£3.5bn): England DfT BUS05a/BUS05b totals uplifted to UK by the ONS mid-2023 population ratio (1.18). Feeds the calibration solver and aggregate_admin_gate.
• build/uk/bus_source_stages.json — a source manifest declaring the two imputation stages (LCFS fares, ETB subsidy) as weighted-QRF draws, each followed by a support_clip to the donor's realized range (kills the over-concentrated high-value tail).
• build/uk/bus_imputation.py — uk_bus_plan(), donor graph, stage names; mirrors us_plan (transforms injected by the caller — no stubs/fallbacks).
• tests/test_uk_bus.py — plan assembly, donor citations, manifest contract, target value/provenance.

Notes

• No incumbent data-package reference (passes test_no_incumbent_data_package_references_in_live_tree); survey microdata is supplied as plain tables at call time.
• Executable QRF/clip handlers are injected by the build runtime, as for the US stages — this PR adds the declarative plan + targets, not the build orchestration, so it does not by itself regenerate populace_uk_2023.
• Follow-up: rail_subsidy_spending and the road-fuel litre proxies share the same missing-anchor class of bug.

Full populace-build test suite passes locally (uv run pytest), ruff clean.

🤖 Generated with Claude Code

[MaxGhenis]

Holding off on merging this as-is. The imputation plan shape is fine, but the calibration targets need to be Ledger-backed before they land in Populace: _DFT_ENGLAND_FARE_RECEIPTS, _DFT_ENGLAND_NET_GOVERNMENT_SUPPORT, and the England-to-UK uplift should be source facts/derived facts with lineage in Ledger, and Populace should reference those facts rather than embed target values in Python. This keeps the no target values in the live repo rule intact and makes the dashboard/provenance surface auditable.

[MaxGhenis]

The diagnosis and the plan shape are right: published populace UK bus fares are ~2x the DfT total and subsidy is well under, and anchoring to DfT BUS05 with a QRF-plus-support-clip imputation is the correct fix. The DfT figures check out — England fare receipts £3.4bn and net government support £3.0bn for the year ending March 2025 (note DfT revised these tables on 24 March 2026, and the precise tables are BUS05aii and BUS05i/ii).

But this has to be rebuilt against the architecture that landed after it opened, and it conflicts with main today (agent-assisted review):

1. Spec-only packages. build/uk is now resource-only — build/uk/__init__.py moved to build/uk_runtime, so editing it is a modify/delete conflict, and test_spec_only_country_packages.py fails on any .py under build/uk. bus_imputation.py and bus_calibration.py are executable and belong in build/uk_runtime; bus_source_stages.json can stay as a spec.

2. TargetSpec no longer takes aggregation. bus_calibration_targets.py passes aggregation="sum" and test_uk_bus.py asserts it — both break on rebase. The green CI here is stale (it ran when the branch was 107 commits behind main).

3. Targets must be Ledger-backed, per my earlier comment and the ledger_targets/LedgerTargetReference surface that has since landed. Don't embed the £3.4bn fare receipts, the £3.0bn net support, or the 68.3/57.7 uplift as Python literals — the DfT facts and the England-to-UK derivation belong in Ledger with lineage, referenced from here.

Two substantive points for the rebuild:

• Vintage: the population is 2023 but the targets are period-2025 DfT. Calibrate 2023 to the year-ending-March-2023 figure, or say why 2025 is the right anchor.
• The England-to-UK population uplift is crude — England receipts already include London, the dominant bus market, so scaling by population likely overstates the UK total, and Scotland/Wales fund their own subsidy. A per-nation or GB anchor would be better; at minimum caveat it.

Also, the executable calibrator grew this past its "declarative plan + targets" description — I'd split it out or move it to runtime as part of the rework. Direction is good; this is a rework, not a tweak.