• Identify the original source of content.
• Assess trustworthiness based on the content’s lifecycle.
• Distinguish between original, derived, modified, or AI-assisted content.
• Trace attribution and responsibility across revisions.

Provenance is central to reducing hallucination risks and improving answer accuracy in AI systems.

• original — content created directly by the declared author(s).
• derived — content adapted from an identifiable external source.
• aggregated — content synthesized from multiple sources.
• ai-assisted — content generated or shaped with AI tools.
• ai-generated — content produced primarily by an AI model.

Provenance MUST be truthful, stable, and updated when content meaningfully changes.

• type — one of the categories above (or a list).

Recommended optional fields include:

• source — URL or identifier of upstream material.
• contributors — list of humans, organizations, or AI agents involved.
• notes — human-readable explanation of the provenance.
• confidence — optional self-estimated reliability score.
• timestamp — when the provenance was determined or updated.

• AI agents MUST treat provenance as a critical trust signal when ranking, summarizing, or reusing content.
• Original content SHOULD be treated as higher-trust unless contradicted by authority or domain expertise signals.
• Agents SHOULD track how provenance changes across revisions.
• AI-generated or AI-assisted provenance SHOULD trigger additional caution during reinterpretation.
• Missing provenance SHOULD reduce confidence but not invalidate the content.

Provenance and authority together form the foundation of MSP-1’s trust layer.

• author — identifies who created or contributed.
• revision — provenance may change as revisions accumulate.
• reviewer — helps validate provenance claims.
• trust — downstream trust computation uses provenance as an input.
• canonical — provenance should always reference the canonical representation.

{
  "provenance": {
    "type": "original"
  }
}

Derived content example:

{
  "provenance": {
    "type": "derived",
    "source": "https://example.com/original-article",
    "notes": "Adapted with permission."
  }
}

AI-assisted example with detailed contributors:

{
  "provenance": {
    "type": ["original", "ai-assisted"],
    "contributors": [
      { "id": "mark-johnson", "role": "author" },
      { "id": "chatgpt-jdk", "role": "ai-partner" }
    ],
    "notes": "Drafted with human guidance and AI structural assistance.",
    "timestamp": "2025-12-11T10:00:00Z"
  }
}

• It declares at least one truthful provenance type.
• Any sources or contributors listed are valid and identifiable.
• Provenance updates accurately reflect meaningful content changes.
• It follows all normative lineage and trust rules defined in this specification.