fiab-0001: Fabric feature scope¶
Status: Accepted Date: 2026-05-22 Locked decision ref: LD-3
Context¶
Microsoft Fabric ships roughly a dozen workloads + a growing set of adjacent capabilities (Fabric IQ Ontology, Graph, Plan, Operations Agent, Maps, Fabric Databases / HorizonDB, etc.). CSA Loom needs to draw a scope boundary: which Fabric capabilities does Loom v1 deliver parity for, which lands in later releases, and which is out of scope entirely?
Two constraints shape the answer:
- Audit-boundary availability of the underlying Azure services. v1 supports Commercial + GCC + GCC-High. Some Fabric workloads require Azure services that aren't authorized at all Gov boundaries (e.g. Azure Container Apps not at IL4+; Microsoft Purview not at IL5; Foundry Agent Service Gov-GA unconfirmed).
- Public information depth on Fabric-only items. Some recent Fabric items (Fabric IQ Ontology, Graph, Plan — first-class at FabCon March 2026) don't yet have enough public architectural information to design honest parity against.
Decision¶
Three-release scope:
v1 (target: 6-9 months from foundation)¶
Cloud boundaries: Commercial + GCC + GCC-High.
Fabric workloads / capabilities with v1 parity: - OneLake (approximate — engine-layer vs storage-protocol enforcement) - Workspaces + capacity model (approximate) - Domains / Subdomains (exact) - Workspace Identity (exact) - Data Factory pipelines + Dataflows Gen2 + Copy Job + dbt Job (exact / approximate) - Mirroring (approximate — same publisher contract; honest UX gap) - Lakehouse + Spark notebooks + environments (exact in Commercial; approximate in Gov) - Materialized Lake Views (approximate — DLT in Commercial; scheduled Jobs in Gov) - User Data Functions (exact) - Warehouse / Polaris (approximate — Databricks SQL Warehouse Commercial; Synapse Serverless Gov) - SQL Analytics Endpoint (exact functionally) - Data Science / MLflow / Vector Search (approximate) - Real-Time Intelligence (exact engine via ADX; approximate UX) - Data Activator / Reflex (approximate — Loom Activator Engine) - Mirroring (approximate) - Direct Lake parity via warm-cache materializer (approximate — honest 5-30s freshness gap) - Data Agents (approximate — extension of apps/copilot/) - Copilot in Loom (approximate — per-pane personas) - OneSecurity (approximate — engine-layer enforcement) - Git Integration + Deployment Pipelines + Variable Libraries (exact functionally)
v1.1 (+3 months)¶
- DoD IL5 support — Marketplace publishing engagement with Microsoft federal; customer-managed plan only; Atlas-on-AKS catalog; HSM-CMK storage
- Power BI embedded Console panes — Console panes surfaceable inside Power BI workspaces
- Remaining 17 industry examples (v1.1 ports wave 2)
fiab-migrateforward-migration CLI- Operations Agent (Loom Ops Copilot — takes action on Loom resources)
- Mirroring source expansion (Open Mirroring publisher SDK + partner-onboarded paths)
- UC managed Gov promotion track — when UC managed reaches Gov-GA, migrate Gov customers from Purview-primary to UC-managed + Purview overlay
v2 (+6 months after v1.1)¶
- Fabric IQ family — Ontology, Graph, Plan, Maps
- HorizonDB-equivalent (Postgres-in-Loom)
- Fabric Maps parity (Azure Maps / Mapbox / PMTiles)
- AI Skills auto-config (auto-generated example queries)
Out of scope (never, or out of Loom's mission)¶
- DoD IL6 / Azure Government Secret — csa-inabox not authorized in this boundary; sponsor-specific deploys only
- Direct Lake on OneLake (no-fallback) parity — engineering- impossible without owning the VertiPaq transcoder
- Multi-cloud destinations — Loom is Azure-only by design
- Replacing Power BI with an OSS BI tool — Power BI is the BI surface; substituting removes the forward-migration story
- Replacing Databricks with OSS Spark — Databricks (Photon, UC managed) is the strategic compute target
- Public-cloud Marketplace for Azure operated by 21Vianet (China) — separate sovereign cloud; separate publisher relationship
- Anti-Microsoft positioning — Loom is Fabric-aligned, not Fabric-competing
Consequences¶
Positive¶
- Customers get a clear contract: "v1 covers these workloads at this parity grade; here's what comes later"
- Engineering scope is bounded; no scope creep from "but Fabric has X" asks
- Honest gaps (Direct Lake sub-second, Fabric IQ family) are documented up front so customers aren't surprised
- v1.1 + v2 give a 12-month forward visibility
Negative¶
- Some federal customers in IL5 wait 3 months past v1 GA
- Fabric IQ family enthusiasts wait 6+ months
- Quarterly freshness rescans needed to keep scope decisions current (Build, FabCon, Ignite cadences)
Neutral¶
- Build 2026 (June 2-3) and future Microsoft events may shift these scope decisions; quarterly review at every major Fabric event
Alternatives considered¶
| Alternative | Why not chosen |
|---|---|
| All boundaries day-one (incl. IL5) | +3 months scope; IL5 publisher-managed-plan complexity adds risk |
| Commercial only v1; Gov v2 | Misses the strategic target — Loom is FOR Gov |
| Top-4 workloads only v1 | Too sparse for federal customers to commit to; sub-MVP |
| Fabric IQ family in v1 | Public info too thin in 2026-Q2 to design honest parity |
| Direct Lake "on OneLake" (no-fallback) parity | Engineering-impossible without VertiPaq transcoder ownership |
References¶
- PRD:
temp/fiab-prd/03-scope.md - Amendments:
temp/fiab-prd/AMENDMENTS.md§A2 - Research:
temp/fiab-research/01-fabric-capability-surface.md,02-gov-boundary-availability.md,03-fabric-only-internals.md - Parent ADR:
docs/adr/0010-fabric-strategic-target.md