It seems that once I marked the previous discussion as "answered", nobody looks at it any more, even though I've added additional information and questions.
I am still seeing the TPA3125D2 being overly sensitive, as though the output short protection is kicking in. I need to understand exactly what conditions cause the device to declare an output short. I have the outputs connected via the LC filter indicated on the datasheet, in the BTL configuration, to a standard consumer-grade headset (ultimately this will be a large speaker, but I'm using the headset for initial testing). The headset is definitely not shorted; it works just fine. Under certain conditions that I can't necessarily define nor repeat, the unit cuts out a couple times per second for 1/4 second or so, as though sensing an output short. Higher input signal amplitude seems to help suppress this behavior, which seems precisely backwards. The output load is not ground-referenced, but the datasheet makes no mention of any such requirement, nor does it seem reasonable given the BTL configuration. What's going on inside the chip, and how could I be falsely tripping the output short detection?