Now that the dbt-mcp server has been released, there are a lot more compelling reasons to support the semantic layer.
By default, because of how configuration changes in the source data make for widely divergent dbt-GA4 package configurations, we can only do a bare-bones sample configuration (probably with the session tables, user tables, and a page_view fact model and an events fact model).
In the discussion on using the custom event macro, we briefly mention that using the macro consistently will let us build a configurator in Jupyter notebook that can read BigQuery and generate the settings.
We could also do this for semantic layer except that we would need to read the Git project and generate the semantic layer since we don't know which fact models that you created and the settings pulled from BigQuery in that notebook will already be in the Git project so we won't need to access BigQuery again to get them. Also, what we find in BigQuery may not represent what you actually want in the semantic layer as events and parameters can be temporarily broken and we don't want to automatically pick up broken settings.
Now that the dbt-mcp server has been released, there are a lot more compelling reasons to support the semantic layer.
By default, because of how configuration changes in the source data make for widely divergent dbt-GA4 package configurations, we can only do a bare-bones sample configuration (probably with the session tables, user tables, and a page_view fact model and an events fact model).
In the discussion on using the custom event macro, we briefly mention that using the macro consistently will let us build a configurator in Jupyter notebook that can read BigQuery and generate the settings.
We could also do this for semantic layer except that we would need to read the Git project and generate the semantic layer since we don't know which fact models that you created and the settings pulled from BigQuery in that notebook will already be in the Git project so we won't need to access BigQuery again to get them. Also, what we find in BigQuery may not represent what you actually want in the semantic layer as events and parameters can be temporarily broken and we don't want to automatically pick up broken settings.