If you look at Connor Winfield's OCXO application note, they have a schematic for a DAC driven VCOCXO. That schematic shows an Analog Devices AD5680 DAC followed by a low-pass RC filter and a buffer amplifier. I sort of always took it on faith that the RC filter was a noise reduction strategy, but never really understood the source of that noise.

Turns out that the AD5680 is an 18 bit DAC, but the bottom two bits are interpolated. In other words, it's a 16 bit DAC in truth, but it has the capability of oscillating rapidly between the value N and N+4. It does so with a variable duty cycle depending on the bottom two bits of the DAC value. If the bottom two bits are both zero, then it doesn't oscillate at all - it simply outputs the top 16 bits. If the bottom two bits are 0 and 1, then it oscillates between N and N+4 with a 25% duty cycle. If the bottom two bits are 1 and 0, then it uses 50%. 1 and 1 results in 75%, and 0 and 0 go back to not oscillating at all. The result of all that is that if you average over a long enough time period, you get the extra resolution of those bottom two bits. It's a bit like how you can use PWM and a low-pass filter as a DAC on Arduino.

So that's what the low-pass filter on Connor Winfield's circuit is doing: it's averaging the interpolated DAC output. Since my version uses the AD5061, which is a non-interpolated 16 bit DAC, its output should be stable, so the LPF probably isn't doing anything. But, well, if there is any noise output, then maybe the RC filter will squelch it a little.