My experience is that the toe and heel of the angle, channel, beam, etc., are treated as if they had no radius.
I assume the reasoning is that the allowable stress on the cross section of the fillet weld is conservative enough to accommodate any incomplete fusion that may occur in the corner of the radius.
That being said, I am surprised at what I read as acceptable in AWS D1.1 with regards to the amount that a fillet weld may be undersized. The amount a weld may be "undersized", i.e., the final dimension of an undersized weld permitted, in recent editions of the code are much different than what was permitted before the 2000 and later editions. The standard was no more than 1/16th inch for 10% of the length, now as great as 1/8th inch for fillets 5/16 inch or larger.
Let me think about the ramifications:
A root opening between members is 1/8 inch, the weld size specified is 5/16 inch. I increase the weld size because of the root opening to 7/16th, but the weld can run undersized by as much as 1/8 inch, so I'm back to 3/16 inch effective leg length for as much as 10% of the weld length. Can the entire weld be 3/32 inch under the specified weld size and still be accepted? Thank God, the answer is no, any weld that is less than the specified size is limited to 10% of the weld length. But what if I don't know there is a 1/8 inch root opening because the welder "wrapped" the welds.
So that's why the safety factor is so large!
Gotta go, gotta get some sleep.
Best regards - Al