Catalog Fidelity Index (CFI)

Merchandising-grade quality scores do not predict whether an autonomous agent can safely cite a SKU. Teams need a defensible index that decomposes into explainable drivers auditors and merchants can act on.

CFI blends corroboration-aligned fidelity, semantic relevance to intent classes, coverage and freshness, and explicit conflict detection across supplier feeds, golden records, and policy text. Thresholds vary by agent use-case severity.

Operational as a gating signal for agent-eligible surfaces; regressions route to owners with driver-level explainability. Correlates with reduced hallucination incidents in pilot evaluations.