The same loop is compressed with everything fixed except the ratio, and you place the ratio on a scale from 1:1 to 20:1. Isolating one parameter is how you learn what it actually does, instead of guessing from a mix of settings.
At level 1 the ratio is somewhere between 1.5:1 and 6:1 and the accepted margin is wide. Level 2 widens the range to 1.2:1 to 12:1, and from level 3 any ratio from 1:1 to 20:1 can come up while the margin keeps shrinking. You'll quickly notice that the jump from 2:1 to 4:1 is far easier to hear than from 10:1 to 20:1.
Past a certain point, the extra reduction is small: going from 10:1 to 20:1 barely changes how much a peak is pulled down compared with going from 2:1 to 4:1. The biggest audible changes happen at low ratios.
Something gentle like 2:1 to 4:1 on most sources, then adjust by ear. Training the ratio on its own makes it much easier to know when you've gone too far.