Well I've learned some interesting things and while I suspect what I discovered would have been instantly apparent to someone using a Lulu, I'll post it here in case it's useful to someone down the road.
I've been buying generic "TPA3116" and/or "TPA3118" (the only difference is whether the thermal pad is on the top or bottom side) from the lowest priced eBay sellers, assuming (and we know what that means) that they were more-less the same. Which they may be, but that begs the real question as all of them I've got do not in fact have either 3116 or 3118 ICs on them, even though that's how they are marked. They cannot, because the real TI parts are in 32 pin packages and the ones I've got are in 28 pin packages! However, they may be genuine TI parts, just different ones - namely TPA3111d1, which is a 10 watt amplifier. TI offers a family of parts built on the same platform in various power levels, and for those like me who are interested in low power QRP transmitters, the good news is they all can take up to 26 volt supplies, so the same modulator trick can be done. I should note that one board appears to be DOA and one bridge output on another is shorted to ground, so maybe the protection circuitry isn't quite as good as claimed.
I was planning to implement a change to cause faults to automatically reset by connecting the FAULT and SD (shutdown) pins together, but that's when I realized the pinout was wrong. Both parts have the same functionality but the pins are different and on the TPA3111 it's pin 1 and 2, and they're already connected on the PCB, and the datasheet clearly says: "If automatic recovery from the short-circuit protection latch is desired, connect the FAULT pin directly to the SD pin. This allows the FAULT pin function to automatically drive the SD pin low that clears the short-circuit protection latch." So why then am I getting latched faults? No idea as of yet, but it is definitely related to connecting input devices with power applied.
The input impedance relates to frequency response and gain as described in the datasheet, which also states: "The impedance at the inputs must be limited to an RC time constant of 1 ms or less if possible. This is to allow the input DC-blocking capacitors to become completely charged during the 14-ms power-up time." Also: "The input capacitors used should be a type with low leakage, like quality electrolytic, tantalum or ceramic polarized type is used the positive connection should face the input pins which are biased to 3 Vdc."
While I don't understand why it is not automatically resetting, the fault that occurs when connecting an input is probably as I guessed, a DC Offset fault: "To avoid nuisance faults due to the DC detect circuit, hold the SD pin low at power-up until the signals at the inputs are stable. Also, take care to match the impedance seen at the positive and
negative inputs to avoid nuisance DC detect faults." The real 3116/8 parts have a separate MUTE input but that pin is omitted on these misbranded TPA3111 parts, and the MUTE terminal is actually connected to pins 1 and 2, which are SD/FAULT. So it may be possible to connect a capacitor that will act like a power-on reset circuit to hold the SD input low at power-up.
I have one of the Wondom 100 watt amplifiers on the way, which I assume will be a genuine TPA3116 for use with a Lulu-type transmitter and it will be interesting so see how it compares.
Bob