Oops sorry
http://www.microchip.com/wwwproducts/en/en022889Just verified, I'm pulling the ~reset pin low until I start SPI communications with it. There is an errant (#2) regarding startup but the code has a 2ms sleep which should mitigate this.
I just changed the code a little bit to try and send data a couple times and if that fails I hard reset the chip using the reset pin and it appears to solve the issue but this is not elegant at all since it requires an extra couple seconds to the startup process 2 out of 10 times.
The more I play with it the more it appears to be an oscillator issue. I might try hooking up my o-scope to the reset pins and an oscillator pin to check the timing of oscillator startup. I'll also look up the caps I'm using on the oscillator which might be causing startup issues.