Stepping back a bit on this, there are really 4 things I think that can cause this issue:
- 8484A sensor
- 11708A attenuator
- E4418B reference output
- E4418B meter itself
Out of all of these, the most likely is that the 8484A sensor itself is actually broken - I have generally seen that unless you pay a couple of hundred dollars for the sensor and buy from someone who says it is working and offers a refund then you're going to be buying a dead sensor.
Now, that said, the manual for the 8481A does provide a very basic way for you to determine if it is the meter or the sensor - This will at least help you get in the rough area for where things might be wrong.
When you measured the reference output, as long as it was somewhere near the 50MHz 0dBm then the calibration process should work. Measuring the 11708A might be harder given what you have but if you can measure S11 then you should have the equipment available to confirm that the attenuator is 20dB near 50MHz, again as long as it is somewhere near that then the process should work even if the actual numbers are highly uncertain.
TonyG