CI/CD inconsistency for Power Bi Reports

Hi mohan072​,

Thanks, that helps. If the semantic model in bi-uat is definitely connected to the Lakehouse in data-uat, and the report is also bound to the correct UAT semantic model, then I would stop looking at binding for now.

The next thing I would isolate is whether the semantic model itself is healthy after deployment.

Since you are using the REST APIs, one thing worth checking is the refresh sequencing. A semantic model refresh request is asynchronous, so the API call succeeding does not necessarily mean the refresh has finished successfully. Microsoft documents this under enhanced refresh with the Power BI REST API.

I would make sure Job 2 does not deploy the reports until the UAT semantic model refresh status has actually reached Completed, rather than just continuing after the refresh request was accepted.

I would also try a very simple DAX query directly against the UAT semantic model using the Execute Queries REST API.

Something as simple as:

EVALUATE TOPN(10, 'YourTable')

If that fails, then the report is probably only exposing a semantic model/data-access problem underneath it.

If the DAX query succeeds but the report visuals still fail, then I would compare the deployed report definition with Dev next, because at that point the model, Lakehouse connection and data path have all been proven independently.

One other thing I would check if this is Direct Lake is the identity being used when the model queries the Lakehouse. Microsoft notes that Direct Lake access can depend on either the report user having ReadData on the Lakehouse or the semantic model using a fixed identity with access to the source. Simply giving the CI/CD identity Contributor access does not necessarily prove that the runtime query identity has access. The Fabric permission model.

So my next three checks would be:

1. Confirm the semantic model refresh actually reaches Completed.
2. Execute a simple DAX query directly against the UAT semantic model.
3. If it is Direct Lake, verify the runtime identity/connection has access to the UAT Lakehouse.

That should tell us whether the failure is still in the semantic model layer or genuinely only in the deployed report.

AI-assisted drafting: AI was used to help structure and phrase this response. I reviewed and validated the technical content before posting.