I can also confirm this bug in 0.9.8.9 sv 2288 using Burkhard's dataset.
I've noticed that the bug occurs for the x,y,y_error data with instrumental weighting but if one plots and fits only the x,y data, the results of the fit are consistent whether done using the fit wizard or from the Analysis->Fit Linear menu option. In that case, the fit parameters also match those returned by LibreOffice Calc. Ed