Repost. Answer from Manfred Fuchs;

The limits for the measurement error are +/-2knots for the GT31 (1Hz data) and +/-4knots for the GW60 (5Hz). The large errors can be due to the strong deceleration or going very fast over larger chop at the end of the runs that cannot be reliably recorded by the GT31. A GW60 (not worn at the wrist, but at the upper arm with a strap-extension) would have been better in that situation.

If one increases the error limits in the filter settings to e.g. +/-3knots, unreliable points are taken into account and better 10s averages result, however, the errors of the runs (+/- column in the results section) increase also as expected. For example the second fastest run has the largest errors (up to +/-2.7knots) and is only run #9 when applying the agreed error limits and make a big difference in the 5x19s average unfortunately...

Track is stil under investigation by Yann. I hope i can stil post my 5x 10 seconds average of 47,62 knots.