Scoring and data methodology
Use this page to interpret the dashboard badges. V1 is UK-only, official-first and trend-based. A status describes the measured direction of one metric on its stated coverage; it is not an overall political judgement about the country.
How to read a status
Statuses combine direction, history and confidence. They should be read with the latest period, coverage badge and drawer caveats.
5Y and 10Y comparisons both improve, or the metric explicitly permits a shorter history.
Improving describes this metric on its stated coverage, not the country overall.
The trend improves, but history, coverage or measurement caveats cap confidence.
Do not present this as a full 10-year improvement signal.
The signal is flat, contextual, mixed, or the 5Y comparison improves while 10Y deteriorates.
Use the drawer caveats before summarising the trend.
The 5Y comparison deteriorates while 10Y improves, or a partial-history deterioration is present.
Treat this as deterioration with uncertainty, not a severe flag.
5Y and 10Y comparisons both deteriorate.
Red is reserved for consistent deterioration across the medium and longer windows.
There is no reliable automatic signal because data are unavailable, too short, or affected by a methodology break.
Grey should stay neutral unless a human override is added later.
Comparison windows
Comparisons are period-aware. The scorer prefers exact period matches and marks each returned comparison with a quality value.
Monthly
Compare the latest YYYY-MM to the same month 1, 5 or 10 years earlier.
2026-04 compares with 2025-04, 2021-04 and 2016-04.
Quarterly
Compare the latest YYYY-Qn to the same quarter in the prior year window.
2026-Q1 compares with 2025-Q1, 2021-Q1 and 2016-Q1.
Annual
Compare YYYY to YYYY-N.
2025 compares with 2024, 2020 and 2015.
Fiscal year
Compare fiscal-year strings to matching fiscal years.
2024-25 compares with 2023-24, 2019-20 and 2014-15.
If an exact period is missing, the scorer may use the nearest prior period only inside a narrow tolerance: up to 3 months for monthly data, 1 quarter for quarterly data, and 1 year for annual or fiscal-year data.
Availability badges
A usable 10Y comparison and a usable since-2019 comparison both exist. Observation count alone is not enough.
The metric has usable observations, but not enough period-matched history for full 10-year confidence.
The dashboard has no usable numeric observations, or too few points to score responsibly.
The observed series has a known methodology break. Automatic scoring stays grey.
Data policy
Official and derived-official observations are the only production data. Sample and mock observations are labelled separately and exist only for layout, interaction, scoring and pipeline tests.
Official mode starts only when the processed manifest records official or derived-official observations indata/processed/*.json.
Scope and source caveats
- Coverage can be UK, Great Britain, England, England and Wales, or mixed when the official source does not publish the same concept for the whole UK.
- Crime sources are not interchangeable: CSEW is a victimisation survey, police data are recorded-offence counts, and Home Office outcomes describe case-resolution categories.
- Business sources are not interchangeable: Companies House records register events, while ONS Business Demography estimates the business population.
- Contextual metrics are included for interpretation, but they stay yellow because a single direction is not always good or bad.
Before citing a metric
- Check the latest period and observation count.
- Check the coverage badge: UK, GB, England, England and Wales, or mixed.
- Read the availability badge before trusting the status colour.
- Open the drawer and review caveats, methodology notes and source URL.
- Use the exact metric wording. Do not generalise one metric to the whole dashboard.
Data pipeline
- Define metric semantics, direction, coverage and source metadata.
- Download raw files into source-specific folders under
data/raw. - Normalize observations with source adapters.
- Score processed observations with the period-aware rules above.
- Validate processed outputs before using the dashboard for factual claims.