It seems that the field extractions written into the data model (the JSON which stores it) are stored just there, and not within the general props of the sourcetype. In order to "backfill", I might apply the regular expressions to the sourcetype with the Fields submenu of the manager. Note that if you start with the complete set of field extractions on the sourcetype before creating your data model, the model's "auto-extracted" field list should show all of the fields on the sourcetype (assuming the sample result set is large enough to tickle all of the extractions).
Personally, I'd do field extractions first, and then the data model. But I'm firmly rooted in Splunk 4.x, 5.x, etc. 🙂