I've run into this problem repeatedly. The issue appears to be that the radial option must be chosen for fill direction. The bevel of the weld preparation causes the fillet weld, the direction of the weld being perpendicular to the bevelled surface. Unfortunately, you can't choose which surface is used for perpendicular projection. (Though, in this case it still wouldn't be correct geometry in the model.)
The work-around is to edit the branch pipe. Sketch the trunk pipe diameter at the appropriate location and extrude as a surface "through all" of the branch pipe. Use the surface tool to split the branch pipe's outer surface using the new surface extrusion as the cutting tool. You may hide the extruded surface if you choose. Returning to the assembly, the branch pipe now exhibits a "cut mark" that matches the trunk pipe's surface. When making the new weld, select the beveled hole surfaces (bevel and inner rim) as one surface group and the matching surface on the branch pipe, which is now limited to the cut mark you made previously, as the second surface group. A radial projection should now give the correct geometry.
You may need to change which group is which, as the weld projection may "invert" the weld as it generates from one group to the other. (Top rim on one side gets matched up to bottom rim of the other side of the weld in an x-patterned cross section. The fault is visible in the weld preview state.) When you see it, you'll know what I mean.
Be aware that if the trunk geometry changes, the branch will also have to be changed to suit or the weld will fail or become bizarre, assuming you're not driving the change parametrically across the assembly some other way.
I use a similar surface cutting technique for exterior pipe bevels inside slip-on flanges in order to get a correct perpendicular/radial groove weld.
