How Deepheem classifies sources and validates findings

Knowing that a source exists is not the same as knowing whether it supports your claim. Here's exactly how Deepheem classifies sources by authority type and validates each finding against retrieved text.

Source classification

Every source retrieved in an investigation is classified by authority type. The classification appears on every source card in your report, so you always know what kind of source you're dealing with before you rely on it.

Judicial Authority — Court judgments, case law (e.g. BAILII, UK Supreme Court)
Official Legislation — Statutes and statutory instruments (e.g. legislation.gov.uk)
Regulatory Authority — Regulators and enforcement bodies (e.g. FCA, CMA, Sentencing Council)
Government Source — Official government publications and datasets (e.g. GOV.UK, ONS, Home Office)
Academic Source — Peer-reviewed journals and research institutions
Authoritative Source — Established, independently verified publishers (e.g. Reuters, Full Fact)
Secondary Commentary — Analysis, press releases, trade publications
Low Reliability — Unverified sources, forums, anonymous content

Finding validation

Once sources are retrieved and classified, each finding generated by the investigation is assessed against the actual retrieved text. The system does not infer beyond what the sources explicitly state. Each finding receives one of four statuses:

Supported — At least one retrieved source contains text that directly and explicitly confirms this claim
Partially Supported — A source confirms part of the claim, or confirms the entity exists but not the specific conduct or outcome being asserted
Unsupported — No retrieved source addresses this claim in any way
Contradicted — At least one retrieved source directly conflicts with this claim

How findings affect the verdict

Findings marked Unsupported or Contradicted are excluded from the verdict automatically. The verdict is rewritten using only Supported and Partially Supported findings. If the majority of findings are excluded, the verdict explicitly acknowledges that the evidence was insufficient to confirm the claims — rather than inferring a confident conclusion from incomplete data.

This means you always know whether a verdict is based on strong retrieved evidence or limited source coverage — before you act on it.