Dataflow field and table descriptions propagated to OneLake catalog and downstream Fabric items

[AlexPowers]

Fabric customers rely on a shared metadata contract across ingestion, modeling, analytics, and AI. Descriptions for tables and fields exist across multiple Fabric engines, including Delta, SQL Warehouse, Semantic Models, and KQL, but today these descriptions are typically defined after ingestion and do not consistently propagate downstream.

This idea proposes enabling table and column descriptions to be authored in Dataflows during the ELT process and automatically propagated into Delta table and column comments, Fabric SQL and Warehouse metadata, Semantic Model field descriptions, and the OneLake Catalog. By setting descriptions at ingestion time, Fabric can establish a single authoritative metadata foundation that travels with the data across all Fabric experiences.

This ensures consistent discovery, understanding, and trustworthy AI-assisted analytics, while creating a durable metadata layer that benefits governance, self-service BI, Copilot experiences, and AI workflows without requiring duplication or post hoc documentation.

Delta and OneLake (Foundational Layer)
Delta supports native table and column comments, making it a natural system of record for metadata.

Example concepts:

• Table-level comments describing dataset purpose and scope
• Column-level comments describing business meaning and usage

Ask in this idea: Allow Dataflows to set Delta table and column comments during write operations so metadata is stored with the Delta files.

---

Fabric SQL Engine (Lakehouse SQL Endpoint)
Fabric SQL can read metadata from Delta and expose it through SQL metadata views and extended properties.

Supported concepts

• Table descriptions
• Column descriptions

These descriptions are visible in SQL tooling and data exploration experiences and can be surfaced to Copilot.

Ask in this idea: Inherit Delta comments authored by Dataflows and prevent metadata loss unless explicitly edited.

Fabric Data Warehouse
Fabric Data Warehouse supports first-class metadata similar to SQL Server and Synapse.

Supported concepts

• Table descriptions
• Column descriptions
• Schema descriptions

Ask in this idea: When Dataflows write to Warehouse-backed tables, descriptions should flow automatically and not require re-authoring.

Semantic Models (Power BI)
Semantic Models rely heavily on field descriptions for usability and Copilot effectiveness.

• Table descriptions
• Column descriptions
• Measure descriptions

These descriptions are shown in field wells and are critical for self-service users.

Ask in this idea: Auto-populate and keep Semantic Model descriptions in sync with OneLake metadata unless intentionally overridden.

KQL Database in Fabric
KQL supports documentation at both the table and column level.

• Table-level documentation
• Column-level documentation

This metadata improves query autocomplete, exploration experiences, and AI-assisted querying.

Ask in this idea: When Dataflows land data into Eventhouse or KQL databases, apply descriptions automatically and treat them as part of schema evolution.

OneLake Catalog
OneLake Catalog should treat table and column descriptions as first-class metadata.

• Surface descriptions consistently across Lakehouse
• Surface descriptions across SQL and Warehouse
• Surface descriptions across Semantic Models
• Surface descriptions across KQL

Key ask: Dataflow-authored descriptions should flow into OneLake once and become the canonical metadata source that downstream tools consume.