Split War Widow's Pension (FRS code 9) out of bsp_reported #1222

Summary

bsp_reported is built from FRS benefit codes 6 and 9 (packages/microcosm-build/src/microcosm/build/uk_runtime/frs_spine.py:825, benefit.isin((6, 9))). The FRS 2024-25 data dictionary labels code 6 "Bereavement Support Payment / Widowed Parent's Allowance" and code 9 "War Widow's/Widower's Pension". These are uprated differently:

• Bereavement Support Payment (BSP) rates have been frozen since April 2017 (SI 2017/410 reg 3; SI 2026/148 art 19 says each amount "remains unchanged").
• Widowed Parent's Allowance rises with CPI under art 18 of the same order (£150.90 to £156.65 a week for 2026/27).
• War Widow's Pension rises with September CPI under the Service Pensions (Amendment) Order.

In the raw FRS 2024-25 benefits table, grossed with gross4, code 9 is £1,077m a year and code 6 is £123m, of which about £100m is at the BSP monthly rates (£23.01 and £80.55 a week) and £16m at the 2024/25 WPA rate (£148.40). So about 92% of bsp_reported is benefits that are uprated by CPI.

PolicyEngine/policyengine-uk#2381 has the full evidence. PolicyEngine/policyengine-uk#2382 adds war_widows_pension_reported, a CPI-uprated input counted wherever bsp is. bsp_reported stays on CPI until the datasets carry the split, because freezing it now would understate war widow's pension by about £84m in 2029-30.

What this repo needs

1. Write code 9 to war_widows_pension_reported and only code 6 to bsp_reported in frs_spine.py. bsp_reported is also listed in frs_spine.py:258, graph.py (lines 309, 859 and 979), spi_benefit_coherence.py:122 (SPI_RESTORED_REPORT_COLUMNS), spi_spine.py:217 and spi_support.py:164; war_widows_pension_reported should be carried the same way.
2. Ideally, also take WPA out of bsp_reported so it is BSP only. VAR2 on code 6 rows is meant to carry the Wid question (1 BSP, 2 WPA), but in the 2024-25 EUL file it does not match the amounts (most £23.01 rows have VAR2 = 2), so check it before relying on it; matching the amount to the WPA rate is a fallback.
3. Keep the split in step with policyengine-uk-data, which needs the same change, so both builds give the same columns.

The policyengine-uk freeze should only follow a dataset release that does step 1.

🤖 Generated with Claude Code