The wiring diagrams in chapter 19 of the M905 manual show that pin 3 should be left open when connecting to an unbalanced input. That would indicate that the Grace output stage is dual amplifiers with no cross-coupling. If you jumpered pin 3 to ground at the output of the M905, the output amp for the cold side is driving into a short circuit, with only the build-out resistor to limit current. The specs indicate 300 Ohm balanced output impedance, so presumably driving full output through 150 Ohms to ground. Not recommended.
If you leave pin 3 from the M905 open as recommended, when using just pin 2 and pin 1 the level will be 6dB lower than when connecting from pin 2 to pin 3.
Depends on the output circuit. With your M905 then yes, the hot and cold leg will be complements, when one side increases the other side decreases. They each go above and below 0V, which is why the meter has a diode bridge, it provides a conduction path no matter which terminal is positive with respect to the other.
Have you used test tones at various levels with your DAC set to a known calibration just to see if each meter is responding accurately? If you are using the DAC in the M905 it has variable output level line-up, so you will have to decide if you are setting full scale digital to 18 dBu, 20 dBu, 24 dBu, etc. and work backwards to see what 4 dBu would be relative to 0 dBFS. You can then use digitally generated sine waves at various levels to check whether the meter is responding as you expect.
And of course verify that the meter is not electrically damaged as mentioned previously. Since 4 dBu is about 1.23V RMS, which is 1.74V peak, you could use a AA alkaline battery to put 1.5V DC through each way and confirm that the meter registers the same in each polarity. Should be simpler than messing with your variable bench supply if you have a battery holder lying around, or just tape some wires on a AA or AAA battery, you're just trying to do a quick check, it doesn't have to be durable.
