Credibility Index

The index is computed by fusing heterogeneous signals into a single score. No single metric defines credibility.

• Factual Accuracy: Measured via automated fact-checking against a verified knowledge base
• Citation Integrity: Evaluates the quality and relevance of cited sources, not just quantity
• Author Expertise: Derived from the Authority Vector of the content creator in the specific topical domain
• Content Freshness: Weighted by a Reputation Decay Function to prioritize current information

The Signal Aggregation Layer normalizes these disparate inputs before fusion.

A Credibility Index is always a normalized, dimensionless score. Raw signal values—such as citation counts, years of expertise, or factual error rates—are transformed onto a common scale.

• Common ranges: 0.0 to 1.0 or 0 to 100
• Enables direct comparison across disparate sources (domains, authors, documents)
• Uses Trust Score Normalization techniques like min-max scaling or Z-score standardization
• Eliminates unit dependency, making the score universally interpretable

This normalization is what distinguishes an index from raw metrics.

The index is not static. It is continuously recalculated as new signals arrive, reflecting the evolving trustworthiness of a source.

• Real-time updates: New citation violations or factual corrections immediately impact the score
• Temporal decay: Older positive signals lose weight via the Trust Decay principle
• Anomaly detection: Sudden score drops trigger Trust Score Anomaly Detection alerts for potential compromise
• Feedback loops: User reports and editorial reviews feed back into the model

Staleness is the enemy of credibility assessment.

Credibility is domain-dependent. A Nobel laureate in physics has no inherent credibility in medieval literature. The index is therefore calculated within a defined topical ontology.

• Each source maintains separate credibility scores per knowledge domain
• Leverages Entity Linking and Resolution to map content to specific topics
• Prevents authority laundering—transferring expertise from one field to another
• The Trust Score Ontology defines the hierarchical topic structure

A single global score is a flawed oversimplification.

Every Credibility Index must be auditable. The score itself is meaningless without a clear lineage of how it was computed.

• Information Lineage Tracking records every signal input and transformation
• Confidence Weighting assigns probabilistic coefficients to each contributing signal
• The Trust Score Schema standardizes the output structure for interoperability
• Enables third-party verification and Trust Score Validation against ground-truth datasets

An opaque score is indistinguishable from an arbitrary one.

The continuous index is mapped to discrete actions through Trust Score Thresholding. This converts a nuanced score into operational decisions.

• High credibility (>0.85): Content eligible for featured snippets and direct answers
• Moderate credibility (0.50–0.85): Indexed but requires additional corroboration
• Low credibility (<0.50): Flagged for human review or demoted in rankings
• Zero credibility: Source is de-indexed or blocked from AI training pipelines

Thresholds are calibrated through Trust Calibration against real-world outcomes.