About & Methodology

The Accountability Index brings together data from 92 official sources across the UK's oversight landscape into one place. Inspectors, auditors, ombudsmen, coroners, select committees, regulators: each publishes its findings separately, in different formats, on different websites. The index connects them.

It covers 2,645 public inquiry recommendations, 6,328 coroner reports, 2,007 NAO recommendations, and 23,046 select committee conclusions, alongside performance data, inspection ratings, enforcement actions, complaints data, and more.

Nothing in it is private data. It is the existing public record, structured and cross-referenced.

The Accountability Index is built and maintained by Matt Lewsey. Matt spent over 20 years in the UK Civil Service working on public inquiries and inquests. His roles included Director of the Grenfell Tower Inquiry, Director of the Hillsborough Law duty of candour, Secretary to the Hillsborough Independent Panel, Head of Inquiries and Inquests at the Ministry of Defence, and Head of the Cabinet Office's central inquiries unit. He also led the design and initial setup of the UK Covid-19 Inquiry.

Why this exists

The UK has strong independent oversight. Auditors, inspectors, ombudsmen, select committees, coroners, and public inquiries all produce findings and recommendations. But the data is published across dozens of separate websites in different formats and rarely linked together. The Accountability Index exists to close that gap.

Independence and funding

The Accountability Index is not affiliated with, endorsed by, or funded by the UK government, any political party, or any public inquiry. It is an independent project.

AI tools are used in the creation and maintenance of the index. Google Gemini classifies and cross-references data at scale. Anthropic Claude assists with development. All AI-generated content is clearly labelled. The underlying data is drawn entirely from official sources.

Public index and paid work

The public index is free to use. Paid work helps fund the project, but it does not change the public record: clients cannot pay to have data removed, rewritten, prioritised, or suppressed.

Contact

matt@accountabilityindex.uk  ·  LinkedIn

The index draws on data from official sources across the UK's oversight landscape, organised into nine categories: recommendations and oversight, health and care, policing, prisons and detention, children and education, ombudsmen and complaints, transport and water, enforcement and transparency, and public spending.

How data is collected

Data is imported from official sources using automated pipelines running on a scheduled basis throughout the day and week. Most sources update daily; others update monthly or when new data is published. Sources include government APIs (Parliament, NHS Digital, CQC, judiciary.uk), structured data downloads (CSV, Excel, PDF), and web scraping where no API is available. Each source has a dedicated importer that normalises the data into a common schema.

Linking findings to organisations

Where records name an organisation (a coroner's PFD report addressed to an NHS trust, for example), we use automated matching to link them to bodies in our database. This matching uses a central resolver with an alias table covering common name variations. Match rates vary by source: typically 80 to 95 per cent for structured data, lower for free-text fields. Where a match cannot be made confidently, the record is left unmatched rather than assigned incorrectly.

View all data sources →

The index is AI-assisted. AI is how we can do this at pace and at a scale no team of analysts could sustain. The platform reads and normalises the flow of oversight publications continuously, cross-references findings across sources, and surfaces patterns that would otherwise remain invisible.

But AI classifies and connects. It does not generate facts. Everything the index surfaces is the existing public record: coroner reports, inspection findings, ombudsman decisions, audit reports, parliamentary scrutiny. Every data point can be traced back to its primary official source. Human judgement sits on top for interpretation, narrative, and the decisions about what matters.

Specifically, AI is used for: classifying records against themes, identifying cross-source patterns, generating summaries of coroner concerns and PFD response classifications, and matching records to organisations. All AI-generated content is clearly labelled. Where AI has been used to summarise a document, the primary source is linked so users can verify it themselves.

Themes identify recurring issues that appear across multiple data sources. When different oversight bodies independently flag the same concern about the same organisation, the theme system surfaces that pattern.

How themes are generated

Themes are created using a two-step AI process. First, AI analyses recommendations, inspection findings, and coroner reports to identify clusters of related concerns. Second, each record is classified against the discovered themes, creating cross-references between sources that would otherwise sit in separate silos.

Themes are grouped into 274 specific issues across 16 parent categories, covering areas including patient safety, use of force, data protection, safeguarding, and institutional culture. They are generated by AI and should be treated as indicative. False positives are possible, particularly where terminology is ambiguous.

What themes show

Each theme page shows which organisations have been flagged by multiple independent sources for the same issue. A single inspection finding may not be significant. If a coroner, an ombudsman, and a select committee have all raised the same concern about the same body, that convergence is.

View all themes →

The index supports full-text search across all records: inquiry recommendations, coroner reports, inspection findings, ombudsman decisions, NAO reports, and more. Results are ranked by relevance and can be filtered by source type.

The index also supports semantic search, which finds conceptually related records even where the exact search terms do not appear. Searching for "communication failures in emergency care" will surface relevant records from coroner reports, CQC inspections, and PHSO decisions that describe the same issue in different language. This is particularly useful for cross-source pattern finding.

If you spot an error or know of a data source we should include, please get in touch. Factual errors are corrected as a priority. Disagreements about the interpretation of evidence are investigated and updated if additional sources support a change.

All significant changes are logged on the changelog.

matt@accountabilityindex.uk