In the datasheet for the LNK6663K, there are are two statements that I find conflicting. On page 3, the description for the source pin says:

SOURCE (S) Pin:
This pin is the power MOSFET source connection. It is also the ground reference for the BYPASS, FEEDBACK, PROGRAM and COMPENSATION pins.

This is clear to me, as the feedback voltage would be measured by the device between its Feedback and Source pins. There should not be any noise source in series with the feedback loop.

However, on pages 10-12, it is recommended to connect the bias winding return to the negative pin of the bulk capacitor. While star grounds used to be all the rage in the 1980s, I prefer to understand the actual current loops in my design. I'm adding a copy of Figure 15 here.

• In light green, I indicated the current path that closes the primary loop (bulk capacitor, primary winding, drain, source, back to bulk cap). Any voltage drop here will be added to the feedback voltage. Surely that can't be right?
• In dark green is the loop that the ground return path takes. Any voltage induced here by magnetic fields will be added to the feedback voltage. A ground plane could help in multilayer designs.
• In purple, I show the shortest path that closes the loop for the feedback voltage. Nature, by itself, will use the loop with the minimum possible area, which would be to pin 12. That's exactly where you need it.

With the roundabout trace suggested in Figure 15, the feedback loop current is forced to take an unnatural path that adds noise/offset to the signal. Could you please explain how that would be better than a direct connection to source?

Thanks!