Bias in pv left-right delta x

When running the alignment for the VELO we see a bias of ~ 20 microns in PV left-right delta x plot when using database with no misalignments. The alignment performs as expected but each time converges at this misaligned position due to the existent bias.

This plot is produced in TrackVertexMonitor and it is an output of the reconstruction. I compared the same plot using the DC page on lhcbpr. On gcc8 all plots are unbiased and on gcc9 until 12/12/2019 pv left-right delta x is unbiased. All dates after this give a biased plot.
Slide that refer to the biased and unbiased distibutions is attached. Brunel_lhcbpr_DC.pdf

The bias has been introduced by the change from PrPixelTracking to VeloClusterTrackingSIMD in Rec!1801 (merged) .