GPS is very good for determining speed. A dedicated unit for such task like V-Box samples speed with 0.01km/h accuracy at 10ms intervals (done by some prediction and integration). Mainly used to measure some dynamics performances for vehicles.
Sure it might have some error is the road is not flat and the HDOP is high (btw there is an app to measure the HDOP for gps). But generally if the speed indicator on the average phone gps is not jumping at constant speed on a level road (phone gps recomputes speed every second or two.. so to ensure a constant speed look at the cycocomputer.. not calibrated yet but whatever speed is showing must be constant).. then that value is bang on to about plus/minus 0.1km/h good enough and if higher speed is involved the better. Because computing speed averages several points in time and gives an average over that time. since speed was constant during that time that average is about what it should be. Position wise it has more error like 10-15-100m even more if weird reflections from buildings are involved, but speed is computed and averaged from more points so in average is more reliable.

To calibrate over distance can be a better approach but it needs a reliable source of distance traveled. (gps can go haywire to display trip distance.. it jumps back and forth on different roads, it's a mess; road indicators and google maps indications are rougly but not necesarly for the path of the bike.. the bike may do some waves from a straight line and who knows if a turn is involved what line is considered for that distance) - either are good enough, but leaves a question mark that for the sake of science it can be done better.
For the most of us, any method by measuring roll-out, or comparing distances on road markings are good enough.

In the 5th wheel device, accuracy is very high (around 2600impulses/ revolution of the wheel - it has 1000 imp/rev at the encoder side and it has a gear multiplier in between of 1:2.6 - in contrast to one impulse/revolution for cyclocomputer), and precision is given by us by correctly determining the circumference of the wheel to the fraction of millimeter. This sort of accuracy is needed at high speed since the car can reach 250km/h with this measuring device and if a millimeter is off then all performances will be off. Not acceptable. Standardized tests for measuring key performances like acceleration time to 100km/h has to be done to the hundredth's of a second. The speed indicator also have to have a high degree of accuracy of at least 0.01km/h at maximum speed. (5th wheel as a measuring device is very good since the bike wheel has a constan load from some springs, it has very low deformation in contrast to car tyre, is not braked nor powered so it's not affected by those regimes, etc. - old fashioned equipment that still works with high accuracy and precision if one takes care to measure the calibration factor by some precise method above)

So this measuring device actually uses a bicycle wheel. Here is the best example where measuring the circumference of a bike wheel is serious business.