It seems even better than that. My phone can tell me which side of a small road (2m accuracy) I am on 95% of the time. Maybe it uses a WiFi database and dead reckoning to improve on pure GPS.
AGPS is designed to give a quicker Time-To-First-Fix (TTFF) but the accuracy is unchanged.
The application here (driving) constrains the solution to 2D (i.e. known altitude from the map), and hence can since that variable is known quite accurately it can improve the accuracy of the other variables.
Receivers tend to be built with augmentation systems. This uses a network of ground stations that measure the variation in the GPS signal and transmit corrections over an additional satellite. This helps the device account for atmospheric interference.
GPS and location systems designed for driving will 'snap' to the correct side of the road.
They're aware of what they're being used for and optimise for that. Similarly, because driving is 2D solution instead of 3D, the altitude will be used derived from the map, not the signal.
Which is why GPS systems designed for driving are fundamentally not suitable for use in (say) planes which need a true 3D solution.
Another thing to note on the accuracy front is that the US is about to start launching the next GPS IIIA generation, which has higher resolution than the current generation.
While the features of IIIA are nice, it will be a while until all the existing ones are replaced by IIIA.
The more immediate nice thing about the IIIA launches are that it adds more satellites that can transmit the L2 signal. Currently there are 18 healthy IIR-M and IIF satellites which can transmit L2. I believe you need about 24 minimum for full coverage.
Once we have full L2 and we get civilian dual band L1/L2 receivers it will get rid of one of the biggest sources of inaccuracy which is ionospheric delay. A dual band L1/L2 receiver will be able to measure ionospheric delay directly. This means the delay calculated will be for that specific point instead of the area covered by a WAAS ground station. Plus you won't need a direct view of the WAAS satellites.
As long as you have an unobstructed view of the equator. To get 2-4m accuracy with commercial GPS you must use WAAS. The sattelites that tansmit the WAAS signal are geostationary.
So if you have buildings, hills or mountains blocking your view you're back to 7-15m accuracy. It gets worse thr furer north you go as your angle to the WAAS sattelites gets closer to he ground.
Plus it's North America only. Not sure if Garmin units support EGNOS which is the system in Europe.
EGNOS and WAAS use the same frequencies and same protocol. They're essentially two implementations of the same thing. As far as a receiver is concerned, they are identical. My Garmin eTrex 20 receives both without issue.
http://www.gps.gov/systems/gps/performance/accuracy/
> Real-world data from the FAA show their high-quality GPS SPS receivers attaining better than 2.168 meter horizontal accuracy, 95% of the time.