For example, when userspace requests a certain TX gain, the chip does some calibration black box magic to determine the best way to distribute that gain across the internal amplifies. Nothing special here, the AD9361 does the same thing. The Lime chip will inexplicably fail the calibration but __will continue to operate__. If the PLL didn't lock, the chip sees something dumb like the loopback being down at -45dBFS so it cranks the gain on all of its internal amplifiers to __correct__ the loss. This results in one or more of the internal amplifiers saturating and you end up with horrible intermod. I was seeing -10dBC ACPR on a simple 1MHz QPSK signal. Sure, you can save off good calibrations or do it by hand, but if the simple loopback cal doesn't work and won't tell the user, the chip is irremediably broken.