Source Authority Graph

The foundational layer of the graph models distinct real-world entities as discrete nodes. These nodes represent authors, institutions, domains, publications, and datasets. Each node carries a unique identifier and a dynamic set of attributes, such as an author's H-Index or a domain's registration age. This granular representation allows the system to assign and track credibility at the most atomic level, moving beyond simple domain-level trust to evaluate the specific researcher or lab behind a claim.

Connections between nodes are defined as directed or undirected edges, each carrying a specific weight that quantifies the nature and strength of the relationship. Edge types include:

• Cites/IsCitedBy: A directional link with a weight influenced by the citing source's own authority.
• AffiliatedWith: Links an author to an institution, inheriting institutional trust.
• CoAuthors: An undirected link that can propagate authority between researchers.
• PublishedIn: Connects a paper to a journal, weighted by the journal's Bibliometric Impact Factor. These weighted pathways are the conduits for authority propagation algorithms.

The graph's core algorithm iteratively calculates authority scores by distributing trust through the network. Inspired by Citation Graph Rank, a node's authority is a function of the quantity and quality of its inbound links. A citation from a highly authoritative source confers more trust than one from an unknown or low-tier source. This process is recursive: the authority of the citing node was itself calculated from its own inbound links, creating a self-reinforcing but mathematically grounded hierarchy of trust across the entire graph.

To prevent static bias toward older, well-cited sources, the graph integrates temporal signals. A Source Recency Weight applies a decay function to the authority score of nodes and edges based on their age. A groundbreaking paper from a decade ago will have its raw authority modulated downward if it hasn't been recently cited or updated, ensuring that fresher, replicated research can compete. This mechanism is critical for domains like medicine and technology where information obsolescence is rapid.

The final authority score for a node is not a single number but a composite vector derived from multiple independent signals. The graph aggregates:

• Topological Authority: Score from the recursive link structure.
• Peer-Review Validation Flag: A binary boost for vetted academic work.
• Authoritative Domain Boost: A positive signal for .gov, .edu, and known institutional repositories.
• Source Tier Classification: A hierarchical prior based on editorial rigor (e.g., Tier 1 for primary research). These signals are combined by a Trust Scoring Algorithm to produce a robust, multi-faceted credibility metric.

The graph maintains a live, dynamically updated Retracted Source Blacklist and a Predatory Journal Filter. When a paper is retracted, its node is not deleted but flagged, and its authority score instantly drops to zero, immediately invalidating all downstream citations that depend on it. This triggers a cascading recalibration, where any source that heavily cited the retracted work may see its own Citation Integrity score temporarily downgraded, reflecting the broken evidence chain.

An algorithmic assessment of a source's importance within a network of citations, analogous to PageRank. Authority is derived from the quantity and quality of inbound links from other credible sources.

• Nodes represent entities (authors, domains, papers)
• Edge weight reflects the credibility of the citing source
• Iterative computation propagates authority through the graph
• Used to rank sources before factoring in content relevance

A quantitative metric that evaluates the trustworthiness of a cited source based on multiple signals:

• Author expertise: H-index, institutional affiliation, publication history
• Domain authority: TLD trust (.gov, .edu), backlink profile, age
• Historical accuracy: Rate of retractions, corrections, disputed claims
• Editorial rigor: Peer-review status, predatory journal classification

The composite score feeds directly into the Source Authority Graph as a node's initial weight before graph propagation.

A hierarchical categorization system that ranks sources into tiers based on editorial rigor and authority:

• Tier 1: Primary research, peer-reviewed journals, official government data
• Tier 2: Established news media, institutional reports, recognized textbooks
• Tier 3: Corporate blogs, personal websites, conference presentations
• Tier 4: Social media, forums, unverified user-generated content

Tier classification serves as a prior probability in the authority graph, influencing how quickly trust propagates from a node.

A temporal decay function applied to a citation's authority score, prioritizing recently published or updated sources to ensure information freshness.

• Exponential decay models reduce weight for older sources
• Field-specific half-lives: CS papers decay faster than mathematics
• Last-updated timestamps can reset or adjust the decay curve
• Prevents outdated but highly-cited sources from dominating results

Works in tandem with the authority graph by applying a time-based multiplier to propagated scores.

A verification technique that checks for agreement among multiple independent, high-quality sources to confirm a claim. Increases confidence through corroboration.

• Requires at least N independent sources confirming the same fact
• Sources must be from distinct authorship and publication chains
• Detects circular citations where sources reference each other
• High consensus strengthens the authority graph edges between agreeing nodes

Acts as a validation layer on top of the graph structure, preventing echo chambers from inflating authority scores.

A measure of the completeness and logical validity of the path from an AI's output claim back through its citations to the foundational, verifiable data.

• Traces each citation to its primary source via Citation Chaining Protocol
• Flags broken chains where intermediate sources misrepresent originals
• Validates that each link in the chain logically supports the next
• High integrity chains receive a multiplier boost in the authority graph

Ensures the graph doesn't reward sources that are well-connected but poorly grounded in primary evidence.