Let's assume the source impedance of the Si5351 is 80 ohms,
and the Si5351 can internally generate a 3v pk to pk square wave, so 1.5vrms.
The current through a 50 ohm load is: i=1.5/(80+50)=0.01154 amps rms.
That's a bit more than the AN619 value of 8ma, but in the ballpark.
I would guess that evaluating the source impedance using a lighter load
will show the source impedance dropping toward 50 ohms.
By varying the load resistor and measuring the voltage across it, we may
come up with consistent numbers for the source impedance and the voltage source.
I would do this first at fairly low frequencies, then see if it remains true at
the frequency of interest where things may get considerably more complicated.
Power into the 50 ohm load is: i*i*r = 0.01154*0.01154*50 = 0.006659 watts, or 6.659mw
In dBm, that would be: 10*log10(6.659) = 8.23
With a 6dB attenuator, the 7dBm mixer is seeing 8.23-6 = 2.23 dBm
That can work fine, though might be have noticeable distortion with very strong signals.
There may be variation between Si5351's.
If some truly do limit current to 8ma, we get:
mw=0.008*0.008/50= 0.0032 watts, 10*log10(3.2)=5.05dbm, mixer gets 5.05-6 = -0.5dBm
The uBitx v5 and v6 have an extra 33 ohm resistor between the Si5351 and the 6dB attenuator,
That would raise the effective Si5351 source impedance from 80 to 103 ohms, and give around 9ma.
(Or perhaps it raises the typical source impedance from 50 to 83 ohms?)
Since the Si5351 is not being asked to work so hard, it will distort less.
The fact that Farhan has found we need an attenuator between the Si5351 and the mixer
suggests to me that the source impedance of the Si5351 is not truly 50 ohms.
Without an attenuator, mixer products would come out of the mixer toward the Si5351,
and be reflected back into the mixer to mix again, creating birdies in the receiver.
I assume this was evaluated with 50 ohm coax between the Si5351 and the mixer,
or with extremely short wires.
Note that I have lots of "may", "could", "think" and "perhaps" in the above text.
Gordon is correct, this sort of armchair engineering only goes so far.
I really should do my own measurements of the Si5351 source impedance.
But I'd bet that Farhan has arrived at a good solution on the uBitx.
It keeps the circuit simple, avoiding a buffer amp between the Si5351 and the attenuator.
And does what it must to quiet down the unwanted mixer products.