01Workload Boundaries & Engine Selection
Microsoft Fabric provides multiple compute engines (Spark Lakehouse, Synapse T-SQL Warehouse, KQL Real-Time Analytics, and Data Factory pipelines) over a single OneLake SaaS data lake. Engine selection must be driven by workload characteristics: choose Lakehouse for un-structured/semi-structured PySpark processing, Warehouse for T-SQL relational data warehousing with DDL/DML isolation, and Real-Time Intelligence for sub-second telemetry streams.
Implementation & Verification Checklist
- Document SLA, latency, concurrency, and recovery objectives for each data pipeline
- Map raw ingestion to Bronze, conformed cleaning to Silver, and business curated models to Gold
- Use OneLake zero-copy shortcuts to reference external Azure ADLS Gen2 or AWS S3 buckets without data duplication
- Establish domain ownership boundaries before workspace provisioning
02Zero-Copy DirectLake & Semantic Layer Design
Power BI DirectLake mode queries OneLake Delta Parquet files directly at in-memory VertiPaq performance without importing data or executing slow DirectQuery SQL translations. Ensure Gold tables use V-Order Parquet compression and conformed dimensions so DirectLake models remain in-memory without falling back to DirectQuery mode during peak concurrent reporting loads.
Implementation & Verification Checklist
- Enforce V-Order compression on Gold Delta Parquet tables
- Verify that semantic model measures avoid unsupported DAX functions that trigger DirectQuery fallback
- Implement conformed dimension keys across all domain semantic models
- Validate row-level security (RLS) policies over DirectLake datasets
03Fabric IQ & AI Agent Ontology Foundations
To prepare data estates for Copilot and Fabric IQ autonomous AI agents, semantic models must maintain strict concept consistency. Primary keys, foreign key joins, entity names, and calculation logic must be standardized across Bronze, Silver, and Gold layers to prevent AI hallucinations during natural language queries.
Implementation & Verification Checklist
- Standardize entity naming and business measure definitions across domain workspaces
- Maintain clear column descriptions and Purview sensitivity labels for AI indexing
- Implement automated data quality steward loops to audit and correct AI query exceptions