Weekly FRS amounts are annualised with 365.25/7 but policyengine-uk's weekly rules divide by 52 #531

Problem

policyengine-uk-data and policyengine-uk use different weeks-per-year factors for the same amounts.

• uk-data annualises weekly FRS amounts with WEEKS_IN_YEAR = 365.25 / 7 (about 52.18; policyengine_uk_data/datasets/frs.py). That covers earnings, benefits, rents and, from Carry the rent boarders and lodgers pay the householder as person inputs #506, the rent boarders and lodgers pay.
• policyengine-uk converts annual amounts back to weekly with WEEKS_IN_YEAR = 52 (policyengine-core commons/formulas.py) before applying weekly rules. Examples are the Housing Benefit and council tax reduction earnings and income disregards, the £20 sub-tenant and board-and-lodging disregards, and non-dependant deduction bands.

So any weekly FRS amount reaches the model's weekly tests about 0.34% (365.25 / 364) higher than the survey recorded. A lodger paying exactly £20 a week, which the law disregards in full (HB Regs 2006 Sch 5 para 22), is counted at about £0.07 a week (£3.57 a year). Independent review of policyengine-uk#2002 and policyengine-uk-data#506 raised this.

Options

1. Agree one producer/consumer contract. For example, uk-data could annualise amounts that feed weekly tests with 52, or policyengine-uk could convert with 365.25 / 7 where a rule is weekly.
2. Keep the gap and document it as a known approximation, with its size.

Either way, the choice should apply across all weekly-tested inputs, not just #506's two columns. Measure the effect on HB, CTR, Pension Credit and non-dependant deductions with a real build before choosing.

🤖 Generated with Claude Code