Why this layer matters

Semantic Integrity is about definitions: what a metric means, how it is calculated, and where its boundaries are. When the same business concept has multiple measures, or when naming and units are inconsistent, AI answers diverge even if the data is correct. This layer focuses on reducing ambiguity and making semantics explicit in the model.

What breaks when this layer is weak

• AI returns different values for the same metric depending on the question phrasing.
• Executives see conflicting numbers across reports and natural‑language answers.
• Automated insights highlight the wrong drivers because the metric definition is unclear.

Common symptoms

• Multiple measures for “Revenue,” “Bookings,” or “Active Customers.”
• Measures with unclear units (percent vs basis points).
• Hidden filters embedded in DAX that change meaning without being visible.
• Measure names that don’t match the business definition.

Root causes

• Metric definitions are stored in documents instead of the model.
• Teams create local measures to solve immediate needs without reusing canonical ones.
• Naming conventions are inconsistent across datasets and reports.
• No owner for key metrics or definition changes.

What good looks like

• One canonical measure per business metric, with documented meaning and unit.
• Names reflect business definitions and calculation boundaries.
• Default aggregations are explicit and correct for each field.
• Semantic changes are reviewed and communicated.

Remediation checklist

• Create a canonical metrics list with owners.
• Deprecate duplicates and map reports to canonical measures.
• Add descriptions for measures, columns, and tables.
• Document units, time windows, and exclusions in the model metadata.

Metrics to track

• % of measures with descriptions

• of duplicate metrics per business concept

• % of measures using a canonical base
• Average metric name conformity to naming standards

Diagnostic bridge

This layer most often deteriorates under the semantic-reliability, change-reliability diagnostic domains. Priority operating controls include Canonical metric definitions, Governed semantic ownership, Definition change reviews.