Assuming that this is a 10MHz reference, at the extreme your reference is 0.003Hz off from nominal (3e-10, or 0.3ppb error). It varies by 0.08ppb in the plot. IDK what it is that you bought (TCXO, OCXO, whatever), but that's rather impressive stability. Depending on what type of oscillator it is you can expect a temperature coefficient anywhere in the several ppm to 0.1ppb. Do you know by how much the ambient temperature (or even better, the oscillator temperature) changed over the duration of the plot? I don't work with temperature-compensated oscillators very often, but I don't see an issue here.