If you change the sequence, you change the system. Designing Gold → Silver → Bronze → Ingestion—with the semantic model as the contract and the lake/warehouse as implementation—doesn’t just alter build tasks. It reshapes how data mesh and traditional architectures behave. The same platform primitives are available to both, but the incentives shift: domains and central teams stop arguing about “which pipeline stage we’re in” and align on “which product contract we’re honoring.”
Below is how the flip lands in each world—what truly changes, what stubbornly stays the same, and what actually gets better.
In a data mesh: contracts become the currency between domains
Mesh has always promised domain‑owned data products with clear interfaces. Leading with gold finally makes that real. The semantic model—grain, conformed dimensions, measures, RLS, and SLOs—becomes the public API of a domain product. In Fabric that API isn’t a slide deck; it’s a living asset: a Direct Lake model over OneLake Delta that’s fast enough for interactive work without an extra import hop. The model—and its performance—drive how Delta tables are partitioned, laid out, or materialized.
Silver turns into light, declarative enforcement of the contract. Instead of multi‑month “silver programs,” domains express reusable transforms as Materialized Lake Views (MLVs) in the lake for star‑ready facts—or fall back to Warehouse objects where relational semantics or T‑SQL procedures are the better fit. Because MLVs materialize to Delta in OneLake, they’re directly consumable by Direct Lake models.
Bronze shrinks and specializes. Domains land only what their contract needs through OneLake shortcuts (virtualize external stores) and Mirroring (low‑latency replication of source systems). Bronze isn’t a museum of everything; it’s a precise inlet. That keeps domain boundaries clean and decoupled.
For real‑time, domains treat Eventhouse as streaming bronze and a product engine: Eventstreams deliver events into Eventhouse for sub‑second KQL analytics, and OneLake Availability projects those same tables into Delta in OneLake so the gold model can use them alongside batch facts—no parallel “streaming stack” to keep in sync.
Mesh governance becomes more computational than conversational: Fabric Domains help teams publish and discover products by business area; shared semantic models (Power BI/Fabric) make those contracts discoverable and re‑usable across workspaces with endorsements and build permissions. The political work of mesh—who owns which KPI—doesn’t vanish, but the unit of reuse is the model, not a crate of “gold tables.”
In a traditional EDW: central standards turn into shared services
Centralized programs don’t disappear; they reposition. The EDW still curates conformed dimensions, SCD rules, and compliance controls—but expresses them as shared semantic models and reusable MLVs, not as mandatory trips through fixed pipeline layers. Downstream domain teams consume those contracts directly (Live/Shared models across workspaces) and compose local products on top—thin, “headless” reports, notebooks via Semantic Link, or exports—without forking logic.
The platform’s two physical surfaces become complements rather than stages: the Warehouse when you want full T‑SQL surface area (DML, procedures, materialized views) on Delta; the Lakehouse when you want Spark‑first ELT with a read‑only SQL analytics endpoint for views and governance. Central teams can standardize when to pick which, but the choice is driven by consumption patterns, not ceremony.
Bronze stops sprawling. Central ingestion focuses on Mirroring key operational systems and Shortcuts for external lakes—keeping inventory lean, auditable, and just‑in‑time. For streaming, the EDW doesn’t bolt on a separate toolchain; it blesses Eventhouse + OneLake Availability so events show up as Delta next to mirrored tables, and the same conformed dimensions and measures apply in daily and intraday contexts.
What changes (in both worlds)
- Sequence and ownership of truth – The product contract (semantic model) is authored first and enforced downward. That elevates data product engineers (viz/semantic modelers) from “report writers” to the people who define “done.”
- Nature of silver – Silver becomes thin, declarative, and close to consumption—Materialized Lake Views in lake, targeted procedures/views in Warehouse—rather than a broad, one‑size curation phase.
- Bronze’s job – Bronze is “land only what the contract needs” via shortcuts, mirroring, and eventstreams → Eventhouse, not a catch‑all stash.
- Streams and batch meet – Eventhouse powers live analytics and, via OneLake Availability, emits Delta so gold can unify batch and real‑time under the same measures.
What stays the same
- You still need conformed language – Someone must adjudicate “customer,” “household,” “AUM,” “policy,” “order.” Upside‑down design doesn’t dodge definitions; it pulls them forward and makes them executable.
- Quality, lineage, security, and cost still matter – Row‑level security, entitlements, audit, lineage graphs, and capacity management remain. They’re easier to govern when the model is the API, but the responsibilities don’t go away.
- Specialization persists – Mesh still needs platform engineers; centralized shops still need domain SMEs. What moves is the center of gravity, not the need for talent.
What gets better
- Lead time to value – Direct Lake lets you model first and avoid an extra import hop; you ship a real semantic contract early, then grow only the silver/bronze needed to satisfy it.
- Fewer copies, clearer lineage – Shortcuts and Mirroring keep bronze lean; MLVs declare transformations; Eventhouse publishes streams to OneLake without bespoke CDC, so the estate stays smaller and easier to reason about.
- Federated without fragmentation – Mesh domains publish semantic models others can build on across workspaces; central teams certify, endorse, and govern those assets—contracts travel; tables follow them.
- Choice without chaos. Warehouse and Lakehouse are fit‑for‑purpose surfaces rather than mandatory steps; teams choose based on workload characteristics and governance needs.
Closing thought
Data mesh and traditional EDW used to argue about where ownership should sit. When you design from gold downward, the more interesting question becomes what is being owned: a reproducible, governed answer. Fabric’s primitives—Direct Lake, Materialized Lake Views, Shortcuts, Mirroring, and Eventhouse—make both camps play the same game: ship the contract, minimize motion, and let every surface reuse the same truth. That’s a healthier architecture, whether your strategy says “mesh” or “monolith.”
EduDataSci - Educating the world about data and leadership 