@MORA, interesting results. One thing I noticed is that in every case I have tested, the Photon has been in a socket, sitting off the board. This air gap may be the key to the SoftAP working. With the Photon (or P1 I assume) soldered to the board, this is where problems begin it seems. This issue may be related to another topic:
Also, there may be some nuggets to take away from the Keurig P1 layout of their board in this topic:
Note the ground plane weaving around the keep-out area. I don't believe it's a 4-layer board but I may be wrong.
