This is what I understand happens (I could well be entirely wrong as most of ths is guesswork from observation rather than knowing stuff for sure)
When you upload a PBIX file to PBI-SSSR Server it creates an entry in the ReportServer DB dbo.catalog table. It also creates two entries in dbo.CatalogItemExtendedContent table. One of these is the visuals, m code etc, the other is the data blob.
If you take a PBIX that has imported data and rename it to .zip you can Unzip it. Inside you will find a file called DataModel. This is a super compressed blob of data. This is the thing that gets split out as the data blob in the dbo.CatalogItemExtendedContent table. Everything else is in the visuals blob.
I assume this is so that you can setup a scheduled refresh of the data and it just overwrites the data blob.
When you view a PBIX report the SSRS server magically stitches them back together somehow. Actually what it looks like it does is fire up a captive instance of SSAS tabular (msmdsrv.exe) and feeds it the data blob. It sends the visuals to your web browser and spins up some web services that your visuals send queries to. The web services forward the queries to the captive SSAS instance and ferry the results back to your browser. So your visuals are effectively connected to this temporary data source and as you slice and dice your charts the queries are sent to the captive SSAS instance via the web services. Once you've spun your report down the captive SSAS instance is quietly killed off, or after a timeout of no activity on the captive SSAS Instance/web services. Some of this is guesswork as there's very little hard info out there. But you can see instances of msmdsrv.exe spinning up and down as users browse PBIX reports.
If you are using direct query or live query data sources its a bit different as you don't get a captive SSAS instance so I assume there's just some web servcies that ferry the queries to the actual data source/s.
When you hit refresh in your web browser while looking at a visual you are asking for a new copy of the visuals and for the data to be reloaded. If the data blob has been updated by a scheduled refresh or upload you would see new numbers. Equally if someone has loaded up a new version of the PBIX with revised visuals you would see that. If the PBIX is connected to live data you would get all the queries issued again to the data source and hopefully see some new numbers, assumimg the data source has been updated somehow.
However if nothing has changed you just spin up a new copy of exactly the same visuals and data and issue the same queries which hopefully return exactly the same results (it would be bad if they didn't), hence the appearance of it not doing very much.
We have some dashboards of our call logging system that are connected to the Live backend SQL DB. When we push that refresh button the numbers move/change when the report re-renders.
