Another thing you might find interesting is that you can generate a report on how good your reference station is. It's actually in Sparkfun's documentation, so you're probably aware, but if not: https://learn.sparkfun.com/tutorials/how-to-build-a-diy-gnss... Specifically the part where you collect data with u-center and upload the results to https://webapp.csrs-scrs.nrcan-rncan.gc.ca/geod/tools-outils... for analysis was very interesting.
Imagine you go on a road trip (along the surface of the earth). How far have you driven? In spherical coordinates, that's just changing two angles. In Cartesian coordinates, it's an ugly mess. Doesn't hurt that it's a lot easier to measure angles in surveying than distance.
However, certain GIS systems like QGIS and arcGIS are designed for making maps and have to display things in a 2D space. Thus, they have a projection mapping the spherical coordinates to Cartesian canvas coordinates and back again. This leads to unintuitive behavior, but it's mathematically hard to do better.
Now, the user interfaces and the terminology and the subtly disastrous inconsistencies between different data sources? Hot flaming garbage, all of it. These aren't problems with the underlying data models though.
Actually no, I'm complaining about the exact opposite. I want to be working in spherical native coordinates, but QGIS seems to treat "degrees" as just another fixed unit of length measurement rather than an angular measurement!
For example I had the measurement projection set to "cartesian", which I would have expected to either give me the linear distance from (X,Y,Z) to (X,Y,Z), or the linear distance between the projection of those two points on an approximation of the earth's surface (either sphere or ellipsoid). Instead, it was treating (Lat, Lon) as if they were (X,Y) coordinates on a flat map and doing Pythagorean theorem on the angular measurements, resulting in the longitudinal distance being off by a factor of cos(latitude) !
I can see no paradigm in which such a result would ever be desired, apart from QGIS fundamentally working in terms of linearized projections, with WGS84 coordinates being added on as an afterthought.
It's amazing what you can do with consumer-priced gear these days. I set up a Sparkfun ZED-F9P breakout board as a fixed beacon on my roof, and then their "RTK Facet" as the rover to do precise measurements to create a map. I could have done the basic thing I needed to by hiring a surveyor or eyeballing things with a tape measure, but this is much more general.
The GNSS software world seems to be a mess though, ripe for a paradigm shift. For example, QGIS seems to be based on flat projections with transformations rather than 3d-native - from what I can tell, QGIS seems to consider the "degree" to be a unit of length measurement! This leads to ridiculous things like being able to accidentally measure a nonsensical "cartesian" distance between two points that differs from the actual distance by a factor dependent on latitude.
I've still got to tidy up my own pipeline that lets me do things like turn N (point, distance) samples into a single point. I would have thought that type of operation would be common, but thinking about how surveyors work I guess they're usually locating points optically, rather than trying to position a GPS receiver at the point to be measured.