Now for the stuff that looks really bad.  What we are taking advantage of here is a thing called characteristic functions.  These are functions that will reduce to either 0 or 1.  This is where the table called #pivot comes in.  (Don't get the name of the table confused with the operation we are perfomring.   This is a table fold and not a pivot.  But I have called this a pivot table, because it helps me conceptualize what is going on and keep things straight.  You can think of the sysreferences table like a sheet of paper.  Take a thumb tack and place it in the top left corner.  Now rotate the paper so that the long side of perpendicular to the floor.  This is what sysreferences looks like.  Now pivot the paper 90 degrees so that the short side is now perpendicular to the floor.  This is what sysreferences is going to look like.)

Keep in mind we are joining this to the #pivot table we created that has 16 rows in it numbered 1 to 16.  Now to understand this, we will walk through and evaluate each one to determine the final outcome.  (Yes, the join we have specified will produce a cross product between sysreferences and #pivot, but this is exactly what we want.) 

The first row in #pivot is 1.  We feed this into the first of the three calculations: r.rkey1 * (1 - abs(sign(p.Value - 1))).  p.Value - 1 = 0 because 1-1=0.  sign(0) = 0.  abs(0) = 0.  And 1-0 = 1.  We are then left with r.key1*1. 
Now the next one: r.rkey2 * (1 - abs(sign(p.Value - 2))).  p.Value - 2 = -1 because 1-2 = -1.  sign(-1) = -1.  abs(-1) = 1. 1-1=0.  We are then left with r.rkey2 * 0.  The same holds true for the next  14 calculations.   The result of this first one being r.rkey1 * 1 + 0 + 0 + 0 + 0 + 0 + 0 + 0 + 0 + 0 + 0 + 0 + 0 + 0 + 0 = r.rkey1. 

Now we feed in 2 which is the next value in #pivot.  r.rkey1 * (1 - abs(sign(p.Value - 1)))  p.Value - 1 = 1 because 2 -1 = 1.  sign(1) = 1.   abs(1) = 1.  1 - 1 = 0 so we are left with r.rkey1 * 0
Now the next one: r.rkey2 * (1 - abs(sign(p.Value - 2)))  p.Value - 2 = 0 because 2 - 2 = 0.  sign(0) = 0  abs(0) = 0  1 - 0 = 1 so we areleft with r.rkey2 * 1.   From this we can see that rkey3 through rkey16 will reduce to zero.

So as you can see, the table #pivot acts as a selector to zero out all but one of the columns.  This will produce and end result of 16 rows for this table that look like the following from this section:        pkcolid = r.rkey1 * (1 - abs(sign(p.Value - 1))) + r.rkey2 * (1 - abs(sign(p.Value - 2))) + r.rkey3 * (1 - abs(sign(p.Value - 3))) +
    r.rkey4 * (1 - abs(sign(p.Value - 4))) + r.rkey5 * (1 - abs(sign(p.Value - 5))) + r.rkey6 * (1 - abs(sign(p.Value - 6))) + r.rkey7 * (1 - abs(sign(p.Value - 7))) +
    r.rkey8 * (1 - abs(sign(p.Value - 8))) + r.rkey9 * (1 - abs(sign(p.Value - 9))) + r.rkey10 * (1 - abs(sign(p.Value - 10))) + r.rkey11 * (1 - abs(sign(p.Value - 11))) +
    r.rkey12 * (1 - abs(sign(p.Value - 12))) + r.rkey13 * (1 - abs(sign(p.Value - 13))) + r.rkey14 * (1 - abs(sign(p.Value - 14))) + r.rkey15 * (1 - abs(sign(p.Value - 15))) +
    r.rkey16 * (1 - abs(sign(p.Value - 16)))

Find Foreign Keys 1 2 3 4 5