Skip to content
Parity matrix — what Loom delivers vs Fabric, grade, and honest gaps

CSA Loom — Parity Matrix

Comparative positioning note

This document is written from the perspective of Microsoft Azure, Cloud Scale Analytics, and CSA Loom. Any description of third-party or competing products, services, pricing, or capabilities is derived from publicly available documentation and sources believed accurate at the time of writing, and is provided for general comparison only. We do not claim expertise in, or authority over, any non-Microsoft product or service; the respective vendor's official documentation is the authoritative source for their offerings, which may change over time. Nothing here is intended to disparage any vendor — where a competing product has genuine advantages, we aim to note them honestly. Verify all third-party details against the vendor's current official documentation before making decisions.

The keystone artifact: for every Microsoft Fabric capability, what CSA Loom delivers, the parity grade, and where to read the detail.

Parity grades:

  • Exact — functional equivalent shipped; UX may differ; semantics match
  • Approximate — functional intent matched; honest gap in performance, UX polish, or specific feature subset
  • Honest gap — not delivered; documented explicitly
  • v1.1 / v2 — tracked; landing in a later release

Storage + namespace

Fabric CSA Loom Grade Detail
OneLake unified namespace ADLS Gen2 + Console unified path tree + Shortcuts service Approximate engine-layer enforcement, not storage-protocol; see OneLake parity
OneLake shortcuts (cross-cloud) Custom shortcuts service over ADLS / S3 / GCS Approximate 20–80 ms latency overhead matches Fabric
OneLake Security (RLS / CLS / object-level, storage-protocol enforcement) UC roles (Commercial) / Purview overlay (Gov) + per-engine RLS/CLS Approximate engine-layer, not storage-protocol
External Data Sharing across Entra tenants Delta Sharing protocol from UC / Synapse Approximate functional; UI gap

Workspaces + capacities

Fabric CSA Loom Grade Detail
Workspace + Capacity model Loom workspace (RG) + Loom capacity (Databricks + Synapse Serverless + ADX + Power BI Premium aggregation) Approximate concept mapped; not single F-SKU billing
Domains / Subdomains Console domain hierarchy + per-domain settings Exact metadata abstraction; full parity
Workspace Identity UAMI per workspace; Console-managed Exact
Workspace Outbound Access Protection (OAP) NSG egress + Azure Firewall app rules Approximate functional intent matched
Capacity smoothing / bursting Per-service auto-scale + Console capacity dashboards Approximate no unified CU pool semantics

Governance + lineage

Fabric CSA Loom Grade Detail
OneLake Catalog Console "Catalog" pane backed by UC + Purview (Commercial / GCC) or Purview-primary (Gov-IL4) or Atlas (IL5) Approximate discovery UX simpler than Fabric's
Purview integration Native (Microsoft Purview is the catalog overlay regardless of Loom) Exact
Automatic lineage across items Spark + ADF + Synapse lineage published to Purview Approximate requires disciplined Purview registration
Sensitivity labels (MIP) propagation MIP labels in Purview → UC tags (one-way connector) Approximate one-way (Purview → UC); not bidirectional
Workspace RBAC + admin model Entra groups + Console role bindings + per-engine RBAC Approximate multi-engine roles vs single Fabric role

Data Factory + ingestion

Fabric CSA Loom Grade Detail
Data Factory pipelines Azure Data Factory pipelines Exact same engine 1:1
Dataflows Gen2 ADF Mapping Data Flows Approximate same engine; different authoring UI
Copy Job ADF Copy activity Exact
dbt Job (Fabric DF native) dbt Core in ADF / Databricks Workflows Exact dbt-azuresynapse / dbt-databricks adapters
On-Premises Data Gateway ADF Self-Hosted IR Exact
VNet data gateway ADF Managed VNet + Private Endpoints Approximate functional parity

Mirroring

Fabric source CSA Loom support Grade Detail
Azure SQL DB / Azure SQL MI Debezium SQL Server connector + Spark Streaming Approximate same protocol surface; UX gap on first-touch
Cosmos DB Cosmos Spark connector with change feed Approximate sub-minute latency matches
SQL Server 2016-2025 (on-prem / VM) Debezium SQL Server + Self-Hosted IR Approximate
Azure DB for PostgreSQL Debezium Postgres connector Approximate
Snowflake Custom poller via Snowflake streams API Approximate no native Debezium
Oracle Debezium Oracle with LogMiner Approximate
Open Mirroring publishers (SAP / SAP Datasphere / Qlik / Striim / Informatica) Honors Open Mirroring landing-zone protocol (__rowMarker__) Exact publishers can drop Parquet directly
Fabric SQL DB auto-mirror n/a (no Fabric SQL DB in Loom) Honest gap not applicable

Data Engineering / Lakehouse

Fabric CSA Loom Grade Detail
Lakehouse item Databricks lakehouse on Delta + UC catalog (Commercial) / Hive (Gov) Exact functional
Spark notebooks + environments Databricks notebooks + cluster policies Exact in Commercial; Approximate in Gov no Photon-via-SQL Warehouse in Gov
Native Execution Engine (NEE) Databricks Photon (Commercial only) Approximate NEE is Microsoft-proprietary; Photon is Databricks-proprietary
Materialized Lake Views (MLVs) Databricks DLT (Commercial) or scheduled CREATE OR REPLACE TABLE Jobs (Gov) Approximate
User Data Functions Azure Functions (Premium EP1 in Gov; Flex Consumption Commercial) Exact
Spark Job Definitions Databricks Jobs (CLI / API) Exact

Data Warehouse (Polaris)

Fabric CSA Loom Grade Detail
Polaris engine Databricks SQL Warehouse (Commercial) / Synapse Serverless (Gov) Approximate
T-SQL DML (INSERT / UPDATE / DELETE / MERGE) on Delta Databricks SQL Warehouse (Commercial) Exact in Commercial; Honest gap in Gov (Synapse Serverless is read-only; writes via Databricks Spark)
Stateless / horizontally elastic Databricks Serverless SQL (Commercial) Approximate
Cross-warehouse / cross-lakehouse queries UC three-level naming (Commercial); Synapse external-table cross-DB (Gov) Approximate
SQL Analytics Endpoint Synapse Serverless over Delta Exact functionally
AI Functions in T-SQL (sentiment, translation, etc.) Databricks ai_query() (Commercial only); Notebook UDFs calling AOAI (Gov) Approximate in Commercial; Honest gap in Gov

Real-Time Intelligence

Fabric CSA Loom Grade Detail
Real-Time Hub Console "Real-Time Hub" pane Approximate UX simpler than Fabric's curated discovery
Eventstream (no-code stream ingestion) Azure Stream Analytics jobs Approximate
Eventhouse + KQL DB Azure Data Explorer (same engine) Exact KQL queries portable
KQL Queryset Console "Queryset" pane + Cosmos store Exact
Real-Time Dashboard ADX dashboards (embedded in Console) Exact
OneLake availability (KQL DB → Delta export) ADX ContinuousExport to ADLS Gen2 Exact
Fabric Maps Azure Maps / Mapbox / PMTiles in Console (v2 deferred) v2

Data Activator / Reflex

Fabric Reflex primitive CSA Loom Activator Grade Detail
increasesAbove(threshold) NRules rule Exact
decreasesBelow(threshold) NRules rule Exact
is above NRules rule Exact
is below NRules rule Exact
changesTo(value) NRules rule Exact
andStays(duration) NRules + Redis state machine Exact
noPresenceOfData(seconds) Cron-based stale detection on Redis Exact
everyNthTime(n, seconds) NRules occurrence counter Exact
Per-object state tracking Redis hash keyed by entity ID Approximate exactly-once semantics unverified vs Fabric
Action surface (Teams / Email / Power Automate / Notebook / Pipeline / UDF / webhook) Function App dispatcher Exact
Visual rule designer Console "Activator" pane (v1: functional; v1.1: polished) Approximate Fabric's drag-drop UX more polished
Latency profile 5-30 s end-to-end Exact matches Fabric Reflex

Data Science / ML

Fabric CSA Loom Grade Detail
Notebook ML Databricks notebooks Exact in Commercial; Approximate in Gov
MLflow Databricks-managed MLflow (Commercial); OSS MLflow on AKS (Gov interim) Approximate in Gov
Model Registry UC MLflow (Commercial); OSS MLflow (Gov) Approximate in Gov
Model Serving Databricks (Commercial); Azure ML or AKS-MLflow (Gov) Approximate in Gov
Vector Search Databricks (Commercial); Azure AI Search vector (Gov) Exact functionally
SynapseML Pip-installed in Databricks notebooks Exact
AI Functions library (Spark DataFrame) Custom apps/fiab-ai-functions/ PyPI package Exact
Semantic Link (read/write Power BI semantic models from notebook) semantic-link-labs (open-source) via Power BI XMLA Exact

Data Agents (formerly AI Skills)

Fabric capability CSA Loom delivery Grade Detail
Agent over Lakehouse (NL2SQL) nl2sql tool in Loom Data Agents Approximate
Agent over Warehouse (NL2SQL) Same nl2sql tool with warehouse data source Approximate
Agent over Power BI semantic model (NL2DAX) nl2dax tool Approximate NL2DAX maturity gap vs NL2SQL
Agent over KQL DB / Eventhouse (NL2KQL) nl2kql tool Approximate
Microsoft Graph queries graph_search tool Exact
Custom AI Search index grounding custom_search tool Exact
Per-source example queries (few-shot grounding) Cosmos DB store of Q→Query pairs Exact
Verified answers (pinned Q→A pairs) Cosmos DB store Exact
Field descriptions UC tags / Purview classifications + custom annotations Approximate
Identity-passthrough execution OBO via MSAL BFF Exact
Foundry integration (one Foundry agent attaches one Loom Data Agent) Exact in Commercial; Honest gap in Gov until Foundry Agent Service Gov-GAs
Sensitivity / RAI policy binding Purview Access Restriction Policies (preview) Approximate
External Python client Custom REST client Approximate

Direct Lake — the hardest item

Fabric capability CSA Loom delivery Grade Detail
Direct Lake on SQL Endpoint (DL/SQL) Premium Import + warm-cache materializer Approximate 5-30 s freshness (honest gap vs sub-second)
Direct Lake on OneLake (no fallback) Not delivered Honest gap requires VertiPaq transcoder ownership
Framing (vs refresh) TOM partition-scoped refresh via Direct-Lake-Shim service Approximate seconds-to-minutes vs Fabric's seconds
V-Order Parquet sort Not implemented; Delta tables written via Databricks (no V-Order) Honest gap only matters when Fabric reads our tables; mitigated by OneLake shortcut + Fabric re-compaction
Composite models Power BI Desktop native composite-model authoring Exact
GCC: any Direct Lake at all Not deliverable Honest structural gap F-SKU unavailable in GCC; structural; not timing-fixable

See Direct Lake parity for the detailed mechanics and the honest discussion of the freshness gap.

Power BI in Loom

Fabric CSA Loom Grade Detail
Power BI semantic models Power BI Premium (P-SKU in GCC; F-SKU in GCC-H/IL5) Exact
Power BI Reports + Dashboards Power BI Premium Exact
Composite models Power BI Desktop native Exact
Power BI Direct Lake (GCC-H / IL5) Direct-Lake-Shim warm-cache Approximate
Power BI Direct Lake (GCC) Not deliverable Honest structural gap
Org Apps Power BI Apps Exact

Copilot in Loom

Fabric workload Copilot CSA Loom Copilot persona Grade Detail
Notebook Copilot (slash commands) /explain, /fix, /comments, /optimize in Loom notebook magics Approximate
NL2SQL in Warehouse pane Loom Data Agents NL2SQL tool Approximate
NL2DAX in Semantic Model designer Loom Data Agents NL2DAX Approximate
NL2KQL in Real-Time pane Loom Data Agents NL2KQL Approximate
Fix-with-Copilot on Spark failures Notebook integration Approximate
Cross-workload context continuity Per-pane in v1; cross-pane in v1.1 v1.1

Lifecycle (Git + Deployment Pipelines)

Fabric CSA Loom Grade Detail
Workspace ↔ Git binding Console "Git" pane → Databricks Repos + ADF Git + TMDL Git Exact functionally
Deployment Pipelines (dev/test/prod) Console "Deployment Pipelines" pane Approximate fewer stages; simpler promotion
Variable Libraries Cosmos-backed Variable Library Exact
fabric-cicd Python lib / Fabric CLI deploy fiab-cli wrapping azd + Bicep + workspace-create REST Approximate
Branched Workspaces (preview at FabCon 2026) v1.1 v1.1

Fabric IQ family (v2 deferred)

Fabric IQ item CSA Loom delivery Grade
Ontology v2 Cosmos + AI Search vector index sketched
Plan (agent orchestration recipes) v2 Container App executor sketched
Graph / FabricGraph v2 Cosmos for Apache Gremlin or TinkerPop on AKS
Operations Agent v1.1 extension of Loom Copilot with write-tools
Maps v2 Azure Maps / Mapbox / PMTiles
Fabric Databases (SQL DB, Cosmos DB, HorizonDB) v2 HorizonDB-equivalent via Postgres Flexible Server

Operations + Monitoring

Fabric CSA Loom Grade Detail
Monitoring Hub Console "Monitoring" pane aggregating App Insights + Log Analytics + Sentinel Approximate
Capacity / CU dashboards Synthesized CU-equivalent dashboard Approximate summed from DBU + DPU + vCore + Power BI memory
Activity Log Azure Activity Log + per-engine audit logs Exact
Audit + DSPM Purview Unified Catalog + DSPM (Gov GA July 2026) Approximate

Compliance + Sovereignty (the headline differentiator)

Boundary Microsoft Fabric CSA Loom
Azure Commercial / GCC (note) GA (Commercial only) GA
GCC-High / IL4 Forecasted Available v1
DoD IL5 Forecasted v1.1
Azure Government Secret / IL6 Forecasted Not authorized (out of scope)
HIPAA BAA (Commercial + Gov) Commercial only All boundaries
CMMC L2 / L3 Commercial only GCC-High via FedRAMP High inheritance
StateRAMP Pending Available via FedRAMP High inheritance
ITAR (in GCC-High) Not yet in Gov Available in GCC-High

GCC runs on Azure Commercial

GCC ("Government Community Cloud") customers run their workloads on Azure Commercial regions under an M365 GCC identity boundary — they do not sit on Azure Government. That means a GCC customer's audit boundary covers Azure Commercial resources, and the Microsoft Fabric Commercial GA covers them. However, the M365 GCC tenant identity rules block direct use of Fabric's tenant-level SP flows that customers in pure Commercial enjoy. Loom is the bridge: the same Bicep that deploys against Azure Commercial deploys against the same regions under GCC identity, and the post-deploy bootstrap issues the AAD app-roles that the GCC identity boundary requires. Bottom line: Azure Commercial and GCC are both GA for CSA Loom, because both audit boundaries are FedRAMP-High-on-Commercial.

Summary by parity grade

Grade Count of capabilities
Exact ~30
Approximate ~40
Honest gap ~5 (Direct Lake on OneLake no-fallback; GCC Direct Lake; V-Order; some Foundry-only Gov features)
v1.1 ~6
v2 ~7 (Fabric IQ family + Fabric Databases)

For the workload-by-workload deep design, see Workloads. For the per-boundary deployment dispatch, see Reference architecture. For the honest treatment of the gaps, see Direct Lake parity and per-workload pages.