• Weight content based on trust indicators.
• Identify reviewed, authoritative, or validated material.
• Distinguish high-integrity sources from unverified or low-confidence ones.
• Apply appropriate safeguards when generating output.
• Understand the trust framework of a site or resource.

Trust declarations are essential to MSP-1’s mission of machine-readable clarity and transparency.

• Be truthful and non-deceptive — accuracy of trust indicators is mandatory.
• Reflect actual verification processes, not merely marketing claims.
• Be stable unless further verification or de-verification occurs.
• Align with provenance, reviewer, and authority metadata.

Trust is not assigned by AI systems; it is declared by resource owners and validated through metadata consistency.

• self-asserted — declared by the author or owner without external validation.
• verified — reviewed or validated by a qualified reviewer or process.
• authoritative — recognized as an official or canonical source.

Implementers MAY extend with domain-specific trust levels (e.g., “peer-reviewed,” “certified,” “fact-checked,” etc.).

• trust — a trust level string OR an object with structured trust details.

Recommended fields include:

• reviewer — entities responsible for verification.
• provenance — supporting lineage evidence.
• scope — what the trust applies to (section, page, site).
• confidence — optional numeric weighting (0–1 or 0–100 scale).
• notes — explanation of trust assignment.

Trust SHOULD be applied at the lowest meaningful granularity (e.g., section-level).

• AI MUST treat trust as a structured signal informing answer reliability.
• Authoritative resources SHOULD receive the highest weighting.
• Self-asserted resources SHOULD be weighted lower unless supported by provenance.
• Conflicts between trust metadata and provenance/reviewer metadata MUST be flagged.
• Trust levels MUST cascade unless explicitly overridden at a lower level.
• Missing trust metadata SHOULD result in neutral weighting rather than negative weighting.

Trust helps AI systems avoid hallucination by grounding responses in validated, high-confidence sources.

• provenance — trust often derives from provenance quality.
• reviewer — reviewers validate claims and increase trust.
• authority — authoritative entities can elevate trust levels.
• version — trust may change across major versions.
• site — site-level trust sets a baseline for all pages.

{
  "trust": "verified"
}

Extended trust structure:

{
  "trust": {
    "level": "verified",
    "reviewer": { "id": "editorial-team" },
    "notes": "Reviewed for accuracy and compliance."
  }
}

Authoritative trust assignment:

{
  "trust": {
    "level": "authoritative",
    "authority": { "id": "msp-1-org" },
    "provenance": { "type": "original" }
  }
}

Section-level trust example:

{
  "section": {
    "id": "interpretation",
    "trust": {
      "level": "verified",
      "reviewer": { "id": "ai-audit" },
      "notes": "Verified for accuracy of semantic rules."
    }
  }
}

• Trust declarations are truthful, consistent, and non-deceptive.
• Trust levels align with supporting metadata.
• No contradictions exist across site, page, or section trust declarations.
• All normative rules for trust semantics are satisfied.