RFC-005 — Trust Score Computation Specification (v0.1)

A Model for Calculating AI-Visible Trustworthiness of Structured Content and Data Fragments

Metadata

• rfc_id: RFC-005
• title: Trust Score Computation Specification
• status: Draft
• version: 0.1
• authors:
  • David W. Bynon (@TrustPublishing)
  • WebMEM Working Group
• date_created: 2025-07-15
• license: CC BY-SA 4.0
• domain_scope: Vertical-Agnostic
• depends_on: RFC-001, RFC-002, RFC-003, RFC-004

Purpose

This specification defines a standard methodology for computing a Trust Score for:

• Individual data_id values
• Digest-level fragments (e.g., plans, providers, terms)
• Full WebMEM Digests

The Trust Score supports:

• AI retrieval conditioning
• Content ranking in structured SERPs
• Transparency, provenance clarity, and semantic alignment

Core Goals

• Incentivize fragment-level truth and explainability
• Provide machine-computable trust metrics
• Enable cross-vertical scoring consistency
• Support memory conditioning and AI retrieval prioritization

Default Trust Score Formula

trust_score = (
  provenance_weight +
  structure_weight +
  glossary_alignment +
  derived_explainability +
  retrievability_completeness
) / max_possible_score

Default Weight Distribution

Metric	Max Points	Description
Provenance Quality	20	Trust layer, license, versioning, and dataset alignment (per RFC-003)
Structural Integrity	20	Conformance to RFC-002, format diversity, and schema validity
Glossary Alignment	20	Presence of defined_term, glossary_id, and scope alignment (per RFC-004)
Derived Explainability	20	Use of derived_translation and python_translation_method
Retrievability & Format Coverage	20	Available in JSON-LD, TTL, HTML, and one or more other formats

Example: Plan Fragment Trust Score

fragment_id: H1234-001-0
trust_score: 94
breakdown:
  provenance_quality: 20
  structure_weight: 18
  glossary_alignment: 19
  derived_explainability: 18
  retrievability_completeness: 19
confidence_level: high
confidence_reason: >
  All values cite CMS primary datasets with full provenance and glossary alignment.
  Derived values include validated transformation logic.

Optional Advanced Metrics

Feature	Impact
python_translation_method present	Boosts explainability and verifiability
Fragment-level IDs (data-fragment-id)	Bonus for retrievability and memory granularity
Multi-format output (YAML, PROV, TTL, etc.)	Improves trust persistence across systems
trust_level=high on all values	Bonus for semantic clarity and glossary alignment

Digest-Level Use

semantic_digest: cms-ma-mapd-plan
digest_trust_score: 92
fragment_count: 1
fragment_scope: plan
trust_confidence: very_high

Computation Rules

• Scoring is modular and extendable by vertical
• Trust Scores may be cached for performance and auditability
• Normalization may vary by domain (e.g., 80 = high for legal; 90 = high for healthcare)
• Trust Score can be embedded in retrieval evaluation payloads

Agent Use Cases

Agents may use Trust Scores to:

• Rank fragments in structured SERPs and summaries
• Prefer AI-ingestible, transparent content
• Suppress black-box or unverifiable fragments
• Reinforce memory using highly trusted fragments

Sample AI Evaluation Payload

{
  "fragment_id": "H1234-001-0",
  "trust_score": 94,
  "retrieved": "2025-07-15",
  "reasons": [
    "Provenance: CMS public dataset with DOI",
    "Glossary: All terms defined and linked",
    "Derived: Logic included for all computed values",
    "Output formats: YAML, JSON-LD, TTL, PROV, Markdown"
  ]
}

Canonical Reference

RFC-005 is maintained at webmem.com/rfc/rfc-005/ and versioned in the WebMEM RFC Registry.