Difficult Datum

It has to be a point ? I can see that getting wonky. You have a line and a tangent point. Anything mid point created in the center would probably be mid point perp to B. In this case, B will only be a 0.0 and will not control rotation as it cannot. Midpoint from C and midpoint from D will then create a " line " for rotation that will be used as CD. I would consider creating CD , rotate to that then take single point on both "sides" and create midpoint for B. Of course all the target points are coming from each other so what you going to establish 1st ?

Is the top right right line, the other side of the width comprising B, a line or a radius?

If line:
Measure the four datum target points and datum B lines. Create midpoints between C1/D1 and C2/D2, create a line between them.
Create a midline between the two features that comprise datum B. Intersect that with the CD line.
Alignment is level to A, rotate to B (can/may/must), origin to BCD point and datum A

If radius
Measure everything as above, except the top line is now measured as a radius
construct a line perpendicular to the datum B line, passing through the radius center
If that line has a vector of <0,-1,0>, pierce the arc with it. If the line is 0,1,0, reverse it, then pierce
construct a line parallel to datum B passing through that point.
Midline for datum B

Or if you're scanning, you can do a high density scan of the radius (call it RAD1 for this example). Measure datum B at the bottom, rotate to it and zero.
Create an arrayed variable
ASSIGN/MAXY_HITS==RAD1.HIT[1..RAD1.NUMHITS].Y
ASSIGN/MAXY_PNT==MAXINDEX(MAXY_HITS)
GENERIC/POINT
NOM RAD1.HIT[MAXY_PNT}.TX,RAD1.HIT[MAXY_PNT].TY,RAD1.HIT[MAXY_PNT].TZ
ACTUAL RAD1.HIT[MAXY_PNT].X,RAD1.HIT[MAXY_PNT].Y,RAD1.HIT[MAXY_PNT].Z
et cetera

That's the highest point from your scan.

I establish the bottom line of B first since most of the basics come from there. The only ones that I cannot get from there are the target points for Datum A. I could rotate to C-D but I don't think a point as a secondary datum is acceptable. Thats where I'm stuck.

I think these would work for a vertical line, but I need a horizontal midline for the my Y origin. Unless I misread, I don't think these methods can do that.

You could do as Lee.drt stated and get the high point from a scanned then use that high point Y result in a generic line. This will give you two parallel lines that you can create a mid-line from and set datum B

It may not be 100% correct but in this instance I would probably align to A/C/D and since the radius is so large and such a small segment of the complete circle I would take 2 equally distant points on both sides of the radius using the C/D midline. (i.e. +10mm and -10mm) Then I would take a point at the peak of the radius and offset that line to the peak..... Then use that line to create your midline for datum B. You just have to be careful about your points on the radius and ensure they are hitting correctly and evenly.

To locate your target points for C and D you can use the bottom half of Datum B since you have a basic to the midplane and from the midplane you have basics to your C/D targets. Once you get your initial alignment established I would follow up with a secondary alignment to more closely and accurately locate all of your datum features.

Other than that I cant think of a way to create a midplane from a radius and a plane.

It has occurred to me that the only way to do this is to have a tangent line to the radius that is parallel to the bottom of the width. However, I have no idea how to construct a tangent line.

I think that would work. I need a tangent line basically and this method should get me that. I'll give it a shot.

You have two datum target points for datum D (11.88 up from datum B). Datum D also needs to be a width feature - in this case a 1D width. You need to create an alignment levelled to A, rotated and translated to B. You then need to measure the R15 on each side of the part - I would use linear open scans and then construct curves from the scans. You can then intersect the curves with a generic plane (11.88 from B) to get your D1 & D2 points. Once you have D1 & D2, construct a 1D width feature and define it as datum D.

D I can do. B is the problem, getting a midline from a line and a radius. I need a tangent line to the curved end of width B I think.