Confidence Scoring in the Perception Layer
Every node and edge in the AI Perception Graph carries a confidence score. Understanding how confidence is assigned, propagated, and used in downstream scoring is essential for interpreting AI visibility analysis.
Confidence in the Perception Graph is a probabilistic measure: the likelihood that a given node or edge accurately represents what an AI system would understand from the available content. A perception_node with confidence 0.9 means: given the available DOM signals, there is very high probability that an AI system would correctly identify and represent this entity. A confidence of 0.4 means: there is significant uncertainty — the entity exists but the signals are too weak or inconsistent for reliable AI inference.
Node Confidence vs Edge Confidence
Node confidence and edge confidence are distinct scores with different computation methods. Node confidence reflects entity recognizability: how clearly the entity is identified in the source content. Edge confidence reflects relationship certainty: how strongly the content establishes the specific relationship type between two nodes. An edge cannot have higher confidence than the lower of its two endpoint nodes — a relationship between two poorly-identified entities cannot be a high-confidence relationship.
How Confidence Propagates Through the Graph
Confidence propagates through the Perception Graph along derivation chains. When a perception node is derived from other perception nodes (one entity is understood through its relationship to another), the derived node's confidence is bounded by the confidence of its sources. This propagation prevents high-confidence claims from being built on low-confidence foundations — a critical property for trust analysis.
●Confidence floors: SiteNexis applies a minimum confidence of 0.2 for any entity or relationship that is explicitly declared in schema markup, even if body text signals are absent. Schema is a strong intent signal — the publisher explicitly chose to declare this relationship. Complete absence of schema and body text signals produces a confidence floor of 0.
Confidence in the Scored Output
The Entity Confidence Score reported in the SiteNexis dashboard is an aggregated measure of perception node confidence across all primary entities on the domain. It is not the average of all perception node confidences; it is weighted by the PageRank authority of the pages from which each entity was extracted. Primary entities on high-authority hub pages carry more weight in the aggregate than secondary entities on low-authority peripheral pages.
Perception vs Fact Layer — Part 8 of 10