Default Semantic Models Are Gone: Cleaning Up the Orphans
Fabric stopped creating default semantic models and turned the existing ones into ordinary models. Here is how to find them, decide which to keep, and govern the rest.
Core Engineering Takeaways
- •Since September 5, 2025, Fabric no longer creates a default semantic model when you create a warehouse, lakehouse or mirrored item.
- •By November 30, 2025, existing default semantic models were disconnected from their items and became independent semantic models.
- •Many workspaces now contain models nobody consciously designed or owns — some with reports depending on them.
- •Inventory them, keep and properly own the ones in use, and remove the rest.
1. What changed
Fabric used to create a default semantic model automatically for every warehouse, lakehouse and mirrored item. It was convenient for quick exploration, but it was limited, tied to its parent item, and it cluttered workspaces with models nobody designed.
Microsoft sunset them in two steps:
- Since September 5, 2025, default semantic models are no longer created automatically for new warehouses, lakehouses or mirrored items.
- By November 30, 2025, all existing default semantic models were disconnected from their item and became independent semantic models.
Nothing was deleted. Instead, every workspace that had lakehouses or warehouses now has ordinary semantic models that used to be invisible plumbing.
2. Why this is worth an afternoon
Former default models are easy to ignore, but they cause real problems:
- Confusion. Users find two models with similar names and do not know which is the certified one.
- Hidden dependencies. Some reports were built directly on a default model because it was there. Those reports now depend on a model nobody owns.
- Inconsistent numbers. A default model has no curated measures; two reports built on it can define "revenue" differently.
- Governance gaps. Unowned models are not documented, not reviewed and not covered by your naming and certification rules.
3. Find them
Former default models usually share the name of their lakehouse or warehouse, so they are easy to spot once you look:
- List semantic models per workspace (the admin portal, the Fabric REST APIs, or Semantic Link in a notebook).
- Flag models whose name matches a lakehouse, warehouse or mirrored item in the same workspace.
- Use the workspace lineage view to see which reports and dashboards depend on each one.
- Check usage metrics for those reports to see whether anyone still opens them.
4. Script the inventory
For more than a handful of workspaces, script it. In a Fabric notebook, Semantic Link can list items so you can match semantic models to lakehouses and warehouses by name. A starting point to adapt to your environment:
import sempy.fabric as fabric
items = fabric.list_items() # current workspace; pass workspace=... for others
stores = items[items["Type"].isin(["Lakehouse", "Warehouse", "MirroredDatabase"])]
models = items[items["Type"] == "SemanticModel"]
suspects = models[models["Display Name"].isin(stores["Display Name"])]
print(suspects[["Display Name", "Id"]])Treat the output as candidates, not verdicts: a model can share a lakehouse's name by design. The lineage view and usage metrics make the final call.
5. Decide: keep, replace, or remove
Every former default model falls into one of three buckets.
No dependent reports, no usage — remove it. It is noise. Deleting it simplifies the workspace and removes a source of confusion.
Reports depend on it, and a curated model exists — replace it. Rebind the reports to the proper, certified semantic model and then remove the former default model. This is also the moment to fix any measure definitions that drifted.
Reports depend on it, and it is the only model — own it. Give it a real name, an owner and descriptions; add the measures people actually need; review its relationships; and bring it under your normal process, ideally in Git as PBIP/TMDL. At that point it is simply a semantic model like any other.
Do not create new models the old way out of habit. New lakehouses and warehouses no longer come with a model, which is a good prompt to design one deliberately — usually one well-modelled Direct Lake or Import model per subject area rather than one per lakehouse.
6. Keep it from happening again
The default-model clean-up is a symptom of a broader question: who decides which semantic models exist? A few lightweight rules prevent the next wave of orphans:
- One owner per model, recorded in the description.
- Endorse (promote or certify) the models people should build on.
- Keep important models in source control so changes are reviewed — the PBIP readiness checker shows how ready your team is.
- Review the workspace inventory quarterly.
If you would like help designing the replacement models, a fixed-scope sprint can cover the clean-up and the new model together.
Clean models start with a health check
Score a model's relationships, measures and documentation in the browser, or have an engineer review the replacement design.
Get the next Engineering Note
Practical DAX, semantic model and Fabric notes. 1–2 a month, unsubscribe any time.