# SCC Nexus Provenance Standard

## Objective

Every important public research object should be traceable backwards from the published claim or figure to the source records and processing steps that produced it.

## Minimum provenance fields

Where applicable, each source or derived research object should record:

- stable internal record ID;
- source title and publisher/creator;
- canonical source URL or persistent identifier (DOI, PMID, accession, Companies House number, dataset ID, etc.);
- source type;
- publication/observation date;
- retrieval timestamp in UTC;
- licence or reuse status where known;
- raw-source checksum when a local artefact is retained;
- extraction method and software version;
- transformation or analysis version;
- reviewer/validation status;
- parent record IDs for derived objects;
- output checksum where practical.

## Hashing

SHA-256 is the default integrity hash for retained raw and generated artefacts unless a stronger project-specific requirement applies. A checksum establishes byte-level integrity, not truth or scientific validity.

## Immutability and corrections

Raw evidence should not be silently overwritten. If a source changes, retain or identify the prior version where licensing and storage permit. Corrections to derived outputs should receive a new version and a dated changelog entry.

## Machine-readable lineage

Derived public datasets should include parent identifiers or a manifest sufficient to reconstruct lineage. Human-readable reports should link to their methodology, source register and versioned evidence package where available.

## AI provenance

If an AI system materially assists extraction, classification, summarisation or drafting, record the task class, model/tool identity where available, date, relevant configuration or prompt version, and the subsequent human or deterministic validation step. AI assistance must not be represented as independent corroboration.