When using classic Newton's second law we have
$$ \vec{F} = m \vec{a} $$
$$ \vec{F} = G \frac{M m}{r^2}{e_{r}} $$
$$ \vec{a} = \left( \frac{d^2 r}{dr^2} - r \left( \frac{d \phi}{dt} \right)^2 \right) \hat{e_{r}} $$
Because eccentricity of a GPS orbit is around e = 0.005 I believe we can treat GPS orbits as circular and write
$$ G \frac{M m}{r^2} = m r \left( \frac{d \phi}{dt} \right)^2 $$
Replacing angular velocity
$$ \frac{d \phi}{dt} = \frac{v}{r} $$
finally we get
$$ v = \sqrt{\frac{G M}{r}} $$
Now, GM ($\mu$) taken from GPS specification is $3.986005 * 10^{14} [m^3/s^2]$, $r$ taken from decoded ephemeris is about $26414660 [m]$. All this gives us
$$ v = 3885 [m/s]. $$
But the same ephemeris data, when decoded according to GPS Interface Specification, Table 30 (SV Velocity) gives values around
$$ v = 2774 [m/s] $$
for SV velocities.
At the beginning I thought that my procedure to calculate SV velocity is wrong. But later on found on the web (https://www.gps.gov/technical/icwg/meetings/2019/09/GPS-SV-velocity-and-acceleration.pdf) some other calculations where SV velocity is in the range 2600 to 3100 [m/s] with avarage 2880 [m/s]. And the question is. Can someone explain why I get such big difference when using classic Newton's law (3885 m/s) and when using ephemeris data (2774 m/s)? Most likely my understanding is wrong, but still cannot figure out what is wrong.
Satellite position, velocity and acceleration as decoded from ephemeris can be found here https://github.com/lukasz-wiecaszek/gr-gnss/blob/master/examples/kepler.dat