I contacted Chris Webb (SQL Server MVP) about this, who got back to me.
The answer is: the premise of the question was wrong. Default processing does not detect changes in the relational source.
In fact, if you have done a full process on your cube, and the data changes in the relational source, Default process will not load the new data into the MOLAP storage.
The only time default processing loads data from the relational source, is when a partition is unprocessed, or one of the indexes is out, or for some other reason certain parts of the cube have been left in an 'unprocessed' state.
So if you're scheduling a SQL Agent job to load data into the cube, and you want to use Default, you need to do something like a Process Update or a Process Add first, followed by a Process Default.
Here is the MS article: https://docs.microsoft.com/en-us/sql/analysis-services/multidimensional-models/processing-options-and-settings-analysis-services