There are several possibilities about the variation that you're seeing. One is that you've got a defective unit. Possible, but doubtful. The second one is that you have a defective reference salt solution. Also possible, and far more likely than a defective actual unit.
Here are a few procedural notes that should give you consistent readings:
Very small amounts of residual solution (whether salt, calibration solution, or rinse solution) can strongly affect the reading. For that reason, it's important to clean the window (with RODI and a paper towel), place a sample of RODI water on the reader, and zero it.
When ready to take a measurement, either tank water or a calibration reference solution, remove the RODI off of the sample window with a dropper/pipette. Then using a separate, dry pipette, place the solution to be measured on the window, remove it, then place a second sample on the window and read it. If you only have one pipette, you can eject the RODI from it, suck up a bit of the solution to be tested, shake the pipette to ensure that the test solution removes any residual drops of water from inside the pipette, expel the test solution, refill it, and place it on the unit to read it.
A comment about the temperature equilibration. I've personally found it important to get the sample to be tested and the unit itself to be close in temperature (within a degree Celsius or so). If the unit itself is much colder than (or warmer than) the water to be tested, you will get incorrect readings. Ditto for the RODI water that you use to calibrate it. For this reason, I place the unit (and RODI if I'm going to zero the unit) in the cabinet of my tank's sump. Within about 10 minutes, all components are quite close to tank temperature, and the unit will read accurately. A comment about this - don't leave your refractometer inside the cabinet permanently. The unit is not completely waterproof, and salt spray will eventually make its way into the unit and onto the circuits, and trash it.
