Vertical Depth Score

Measures the concentration of domain-specific terminology and technical jargon that generalist models rarely encounter in broad training data. This includes proprietary acronyms, regulatory terms (e.g., 'IFRS 17', '21 CFR Part 11'), and niche taxonomies. High density signals that content was authored by a practitioner, not a generalist writer.

• Calculation: Ratio of verified domain terms to total token count
• Signal: Differentiates expert content from surface-level summaries
• Example: A legal brief using 'res judicata' and 'collateral estoppel' correctly scores higher than one using only 'case closed'

Evaluates the depth of regulatory framework integration within content. This component scores documents that correctly reference specific statutes, compliance standards, and jurisdictional nuances. Generative engines prioritize content that demonstrates awareness of legal constraints, as this reduces hallucination risk in regulated verticals.

• Key signals: Citation of specific regulatory clauses (e.g., 'GDPR Art. 17')
• High-value patterns: Discussion of regulatory grey areas and enforcement trends
• Anti-pattern: Generic statements like 'comply with all applicable laws'

Quantifies the complexity of entity-to-entity relationships expressed within the content. Instead of merely mentioning entities, high-scoring documents define the predicates connecting them—effectively contributing new triples to knowledge graphs. This component rewards content that explains how and why entities interact.

• Graph contribution: Novel predicates between known entities
• Example: Stating 'Company A acquired Company B' is low-depth; explaining the regulatory rationale and market impact is high-depth
• Measurement: Count of unique, non-obvious relationship types per document

Assesses the degree to which tacit practitioner knowledge is converted into explicit, machine-readable content. This includes decision trees, troubleshooting flowcharts, and heuristics that experienced professionals use but rarely document. Generative models treat this as high-value information gain because it fills a critical gap in their training corpora.

• Examples: 'If X fails, check Y before escalating to Z'
• Format: Stepwise logic, conditional branching, failure-mode documentation
• Value: Transforms unwritten intuition into retrievable knowledge

Measures how precisely content aligns with the current state of practice in a domain. This component penalizes outdated references and rewards content that acknowledges recent regulatory changes, technology deprecations, and evolving standards. It includes explicit markers for deprecated knowledge to prevent AI models from surfacing obsolete guidance.

• Positive signals: References to recent enforcement actions, updated standards
• Negative signals: Recommending deprecated APIs or superseded frameworks
• Mechanism: Timestamped assertions with version-context metadata

Evaluates the application of methodologies from adjacent domains to solve problems within the target vertical. This component rewards content that demonstrates sophisticated reasoning by importing frameworks, models, or findings from one field into another—creating novel insight that generalist models cannot synthesize independently.

• Example: Applying financial risk modeling techniques to cybersecurity threat assessment
• Signal strength: Correlated with unique information ratio
• Measurement: Detection of cross-domain terminology clusters and reasoning patterns