I'll take yet a third stab at explaining. There is no formally deterministic way of rendering raw sensor data into an RGB pixel raster. All such renderings are by nature subjective and empirically derived. Even calibrated renderings of known sample targets(various color cards, etc), in fixed lighting, are to a certain extent, applicable only to that specific shot. Various software, approaches this in various ways, with various terminology. I guess you could define the sensor/camera mfgr's rendering software as the default, but that's basically just semantics
So yes, anything you do in Lr, always starts with a rendering to 'some' profile. Depending on cameras and Lr versions, there may be multiple profiles provided to choose from, and you or third parties can always create custom profiles with the tools Adobe provides.
If you haven't done any image post processing, then the exported image adjustments will be confined to:
a) the inherent modifications of resampling, in this case downsampling.
b) the inherent modifications of the ProPhotoRGB to sRGB profile conversion.