Leaving 95.5%, which is about the battery to AC spec (at 65% load)
Maybe SolArk measures voltage and current feeding inverter, rather than having sensors to measure AC power? That would not include 48V battery to HV DC losses.
Do you have a clamp ammerter? That could measure AC RMS current (cheap ones less likely to be true RMS.) AC VA / Vrms = Irms. Watts may differ. But compare clamp measured Irms; if waveform isn't sine wave, has shorter higher spikes, that would have more I^2R loss in FETs and inductors compared to a nice "real" resistive load. Such a meter reading migh explain your results.