7.4.7 Quality assessment and validation: Caveats
The algorithm used is intended to fit orbital solutions to the RV curves of binaries. It is doing so but is not protected against some anomalies. In addition, it is not (for the moment) able to properly deal with higher order systems. Besides, it should be noted that all decisions are taken in a statistical way; this means that the output is statistically correct but that some incorrect individual fits can survive near the borders where the decision are taken (e.g. between an orbital fit and a trend solution for example). The global weak point of the algorithms could be the determination of the period even if hereabove it is shown that the success rate is at least 80%.
The RV dataset suffers from all the problems not attenuated by the main spectroscopic processing and is not particularly protected against outliers, misclassified double-lined objects, intrinsic variables, emission-line objects and other problematic situations (e.g. badly normalised continuum, etc…).
There is probably at a low number of transits a tendency to overestimate the eccentricity which is a general behaviour in such RV fits. Later in the process, the spectroscopic solutions may be combined with photometric or astrometric ones and in these cases the resulting output is certainly very secure. If the user of the catalog only finds the results of the spectroscopy fits, it is advisable to restrict the sample to some kind of gold sample. This can be done by performing additional selections in the catalog. For example, large number of transits, higher significance, periods in the range over ten days which seems better, period with small errors, solutions with negligible phase gaps, etc….