Source Tier Classification

The highest tier reserved for sources with the most rigorous editorial and provenance standards. These sources provide direct, verifiable evidence.

• Direct Primary Sources: Original research datasets, clinical trial results, and first-hand accounts.
• Institutional Repositories: Content from .gov, .edu, and established standards bodies (e.g., IETF, ISO).
• Peer-Reviewed Literature: Journals with robust editorial boards and high Bibliometric Impact Factors.
• Verification: These sources serve as the ground truth for Factual Entailment Ratio calculations.

Sources with strong editorial oversight and a track record of accuracy, but which primarily synthesize or report on primary information rather than generating it.

• Established Media: Major newspapers and wire services with documented retraction policies.
• Industry Analysts: Reports from firms with a verifiable methodology and historical data.
• Subject-Matter Experts: Content from recognized authors with a high H-Index Weighting.
• Signal: Receives an Authoritative Domain Boost but is cross-referenced against Tier 1 for critical claims.

Sources with minimal to no editorial gatekeeping, where authority is highly variable and must be individually verified. This tier is subject to the highest scrutiny.

• Social Media Posts: Content from platforms like X or Reddit, requiring Cross-Reference Consensus before use.
• Personal Blogs: Self-published content without institutional backing.
• Unverified Forums: Community-driven sites where identity and expertise are not validated.
• Risk: Automatically flagged by the Predatory Journal Filter and Retracted Source Blacklist if applicable. Carries a high Hallucination Risk Index if used in isolation.

The lowest tier, encompassing sources with no verifiable provenance or those known to be artificially generated. These are typically excluded from evidence chains.

• Anonymous Content: Information with no identifiable author or originating entity.
• Known Synthetic Media: AI-generated text and imagery not explicitly watermarked as authentic.
• De-indexed Sources: Domains that have been algorithmically demoted for systematic misinformation.
• Action: Triggers a Source-Output Divergence Metric check and is automatically assigned a negative Source Credibility Score, effectively blacklisting it from citation.

Tier classification is not static; it incorporates a temporal dimension to ensure information freshness and detect post-citation changes.

• Source Recency Weight: Applies a decay function to authority, prioritizing recent Tier 1 and 2 sources for time-sensitive topics.
• Citation Drift Detection: Continuously monitors cited sources for updates or alterations that could invalidate the original claim.
• Reference Provenance Hash: A cryptographic snapshot taken at the time of citation to immutably verify the source's state, regardless of its current tier status.

A source's tier is reinforced or degraded by its position within the broader citation ecosystem, analyzed through graph-based algorithms.

• Citation Graph Rank: Elevates sources that are frequently cited by other high-tier sources, analogous to a trust-based PageRank.
• Co-Citation Analysis: Strengthens the relationship between sources that are consistently cited together in credible documents.
• Source Diversity Index: Penalizes over-reliance on a single source or a narrow cluster, ensuring a high Evidence Chain Integrity by demanding corroboration from multiple independent Tier 1 or 2 sources.

An algorithmic rule that assigns maximum authority weight to direct, first-hand accounts over secondary interpretations. This heuristic distinguishes between a raw clinical trial dataset (Tier 1) and a press release summarizing it (Tier 3).

• Implementation: Applies a multiplicative boost (e.g., 2.5x) to sources classified as primary
• Key differentiator: Detects whether content represents original research or derivative commentary
• Example: A patent filing receives higher weight than a blog post discussing the patent

A positive signal applied to citations from established, high-trust top-level domains and institutional repositories. Domains like .gov, .edu, and recognized archives (e.g., arXiv.org, PubMed Central) receive automatic tier elevation.

• Mechanism: Pre-computed domain authority list with tier assignments
• Granularity: Operates at the subdomain level (e.g., nist.gov vs. generic .gov)
• Dynamic updates: Domains can be downgraded if they host compromised or retracted content

A temporal decay function that modulates a source's authority score based on its publication date. Freshness requirements vary by domain: medical research may decay within 24 months, while mathematical proofs remain authoritative for decades.

• Formula: Exponential decay with domain-specific half-life parameters
• Configuration: Half-life of 18 months for clinical research, 60 months for engineering
• Safeguard: Landmark papers (highly cited, foundational) receive decay exemptions

A binary indicator confirming whether a source has undergone formal peer review prior to publication. This flag serves as a strong heuristic for academic rigor and is a primary differentiator between Tier 1 (peer-reviewed journals) and Tier 2 (preprint servers).

• Detection: Cross-references ISSN/DOI against known peer-reviewed journal databases
• Nuance: Distinguishes single-blind, double-blind, and open peer review
• Limitation: Peer review status alone does not guarantee correctness; retracted papers retain the flag until explicitly blacklisted

A classifier designed to identify and down-weight or exclude sources from publications characterized by fraudulent editorial practices. These outlets mimic legitimate journals but lack genuine peer review, often charging authors for rapid, unvetted publication.

• Detection signals: Aggressive solicitation patterns, fake editorial boards, rapid acceptance timelines
• Integration: Feeds into the Retracted Source Blacklist for automatic citation invalidation
• Maintenance: Requires continuous updates as predatory publishers evolve their tactics

The application of the H-index—a metric measuring both productivity and citation impact—to weight the credibility of an individual author's cited work. An author with an H-index of 45 has published 45 papers each cited at least 45 times.

• Application: Modulates source tier classification at the author level
• Normalization: Field-adjusted to account for varying citation cultures across disciplines
• Caveat: Used as a supplementary signal, not a standalone authority measure; early-career researchers may have low H-indices despite rigorous work

An algorithmic weighting rule that gives higher authority to direct, first-hand accounts over secondary interpretations. Within tier classification, this rule ensures that:

• Tier 1 is reserved for original research, raw datasets, and primary documentation
• Tier 2 includes peer-reviewed meta-analyses that synthesize primary work
• Tier 3 covers tertiary summaries and journalistic reporting

The rule prevents a well-written summary from outranking the foundational research it cites.

A positive signal applied to citations from established high-trust domains. This heuristic directly influences tier placement:

• .gov and .edu domains receive automatic Tier 1 or Tier 2 consideration
• Recognized institutional repositories (e.g., arxiv.org, nih.gov) bypass initial spam filters
• New or obscure domains must earn authority through citation graph rank before ascending tiers

The boost is not absolute—a retracted paper on a .edu domain still gets downgraded.

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. This metric:

• Prevents citation farms from gaming tier classification
• Ensures that a source cited by many Tier 1 papers gains authority transitively
• Helps distinguish between a widely-cited seminal work and an isolated but rigorous study

Graph rank and tier classification operate as mutually reinforcing signals.

A temporal decay function applied to a citation's authority score. Even a Tier 1 source loses relevance if it is outdated for time-sensitive claims. The weight operates as a modifier:

• A 10-year-old Tier 1 paper on a fast-moving field may be downgraded below a recent Tier 2 meta-analysis
• Foundational, timeless works (e.g., established mathematical proofs) are exempt from decay
• Recency thresholds vary by domain—medical research decays faster than philosophy

This ensures tier classification remains contextually appropriate.