UK FRS stages evaluate policy parameters at the survey period: declare a policy-year rule (uk-data#475, uk-data#478)

[juaristi22]

Parent: #665 (workstream E, #145); registered on #796. Incumbent filings: PolicyEngine/policyengine-uk-data#475 and PolicyEngine/policyengine-uk-data#478, fixed on the incumbent side by PolicyEngine/policyengine-uk-data#480; the week-constant point from PolicyEngine/policyengine-uk-data#476.

What

Three stage paths read policyengine-uk parameters at the frame period (2024, the survey year), where the incumbent did the same:

1. uk_runtime/frs_disability.py: categories from parameters(2024).baseline.gov.dwp, which is the FY 2023-24 rate table (AA higher 101.75), because the baseline clone is taken before the fiscal-year conversion; flags from parameters(2024).gov.dwp (FY 2024-25, 108.55). The docstrings record the split as value-bearing parity with the incumbent; uk-data#475 now identifies it as an incumbent defect. The contemporaneous rates for April 2024 to March 2025 fieldwork are the plain gov tree at 2024. Not in spine_swap_signed_differences.json.
2. uk_runtime/spi_income.py::_refresh_disability_derived_inputs repeats the split for the SPI refresh of the same columns, and divides by the model's 52-week constant where frs_disability divides by 365.25/7 (uk-data#476's only real content). The two stages disagree on the categories path by 0.34%.
3. uk_runtime/frs_education_grants.py (frs_education_grant_split): DSA capacity is a zero vector because 2024 < DISABLED_STUDENTS_ALLOWANCE_FIRST_MODELED_YEAR (uk-data#478). disabled_students_allowance_eligible_expenses is an all-zero input, excluded from parity as zero-share; the stage notes sign it as inert.

Fix

• A declared year_rule on parameter-reading operations, as student_loans already declares (year_rule: calibration_year): observed-amount thresholds (disability categories and flags) at the survey fiscal year's converted tree (gov@2024); modelled capacities (DSA gate, eligibility predictors, maximum) at the calibration year.
• One week constant across frs_disability and the SPI refresh.
• Fix-and-sign: re-mint the parity fixture and the signed-differences entries; re-measure disability category and flag shares and the two TCL disability cells (UK spine: child disability under-identification blocks the TCL disability cells (−56.5% / −40.0% across three spine vintages) #808 lineage); record the DSA column's new mass.

Acceptance

• source_stages.json declares the year rule for both stages; the runtime resolves parameters from it, never from the frame period alone.
• Category/flag shares and the DSA column measured on a licensed build, with the deltas vs the incumbent signed.
• No stage reads .baseline.gov unless the declaration says why.

Not filed separately

• uk-data#476: frs_disability.py already uses one constant; only the SPI refresh above.
• uk-data#477: covered by frs_release.json, frs_release.py validation, test_uk_frs_release.py:44 (spine TIME_PERIOD == release time_period), and the tab sha256 check at the root; Root identity: every raw microdata input hash-pinned and referenced to its Chronicle registration #848 registers the pin in Chronicle.
• uk-data#479: microcosm has no base-year materialisation and never ported the four scalers; it is an input to Adjudicate the incumbent's four post-calibration aggregate scalers #790.

🤖 Filed with Claude Code