Split out of #476 at Vahid's request on #480.
Both disability derivations threshold reported amounts at rate − GBP 1/week against the DWP rates in force during the survey fiscal year (disability_benefits.py, after #480). That GBP 1 gap is smaller than one April uprating step: AA higher, DLA care highest and PIP daily living enhanced all moved 101.75 → 108.55 (GBP 6.80/week) in April 2024. A respondent who reports the previous year's rate for the whole year (101.75 × 52.1786 = GBP 5,309.17) falls below the enhanced-flag threshold (GBP 5,611.81) and the HIGHER/ENHANCED category threshold (107.55/week): they lose is_enhanced_disabled_for_benefits, which drives the severe-disability premia, and drop a category band. The error is one-directional: it only removes the flag and demotes the category.
Latent on the current vintage. #480's before/after receipt on the licensed FRS 2024-25 tabs shows no respondent in the band between the two rate vintages (categories and flags identical under both threshold sets), so the shipped dataset is unaffected today. It bites on any vintage that carries stale-rate reporters (fieldwork straddling an uprating, or respondents quoting last year's award letter).
test_categories_and_flags_share_the_survey_fiscal_year_rates (#480) encodes today's behaviour: a full-year reporter of the previous year's rate is LOWER and not enhanced. Any change here revisits that test.
Options:
- Express the safety gap for the top band against the previous fiscal year's rate (
previous − GBP 1), which accepts one uprating step of staleness; the next band down is ~49% lower, so there is no promotion risk.
- Keep the GBP 1 gap and document that stale-rate reporters are demoted.
Option 1 needs a receipt on a vintage that has such reporters; the FRS 2024-25 tabs cannot distinguish them.
🤖 Filed with Claude Code
Split out of #476 at Vahid's request on #480.
Both disability derivations threshold reported amounts at
rate − GBP 1/weekagainst the DWP rates in force during the survey fiscal year (disability_benefits.py, after #480). That GBP 1 gap is smaller than one April uprating step: AA higher, DLA care highest and PIP daily living enhanced all moved 101.75 → 108.55 (GBP 6.80/week) in April 2024. A respondent who reports the previous year's rate for the whole year (101.75 × 52.1786 = GBP 5,309.17) falls below the enhanced-flag threshold (GBP 5,611.81) and theHIGHER/ENHANCEDcategory threshold (107.55/week): they loseis_enhanced_disabled_for_benefits, which drives the severe-disability premia, and drop a category band. The error is one-directional: it only removes the flag and demotes the category.Latent on the current vintage. #480's before/after receipt on the licensed FRS 2024-25 tabs shows no respondent in the band between the two rate vintages (categories and flags identical under both threshold sets), so the shipped dataset is unaffected today. It bites on any vintage that carries stale-rate reporters (fieldwork straddling an uprating, or respondents quoting last year's award letter).
test_categories_and_flags_share_the_survey_fiscal_year_rates(#480) encodes today's behaviour: a full-year reporter of the previous year's rate isLOWERand not enhanced. Any change here revisits that test.Options:
previous − GBP 1), which accepts one uprating step of staleness; the next band down is ~49% lower, so there is no promotion risk.Option 1 needs a receipt on a vintage that has such reporters; the FRS 2024-25 tabs cannot distinguish them.
🤖 Filed with Claude Code