The Veracity engine measures how accurately AI models represent source web pages during grounded generation by calculating a fidelity score based on survived, lost, and distorted facts using Gemini's url_context tool.
flowchart TD
subgraph Inputs
SRC[Source Web Page]
RSP[AI Response Segments]
end
subgraph VeracityEngine[Veracity Engine]
V_PARSE[Fact Extraction & Matching]
subgraph ClaimBuckets[Claim Categorization]
SURV[Survived Facts]
LOST[Lost Facts]
DIST[Distorted Facts]
end
end
SCORE[Aggregate Fidelity Score]
SRC --> V_PARSE
RSP --> V_PARSE
V_PARSE --> SURV
V_PARSE --> LOST
V_PARSE --> DIST
SURV --> SCORE
LOST --> SCORE
DIST --> SCOREReferenced by