Typically, you decide on the average and then adjust the shape accordingly.
Most professors would go for a 12 points (60%) out of 20 average and a standard deviation of around 3 points (15%). Every student below 10 points (50%) would fail the class.
After that, you rank the questions from easy to hard, according to the scores obtained for each.
Initially, you a award the same weight for each question. If the test was designed properly, this should create a Gaussian distribution.
If not, different weights within a range (e.g. 0.8 to 1.25) for the questions can be adjusted until it matches the Gaussian distribution.
I doesn't solve the problem of easy classes competing with difficult ones but it solves the problem of grade inflation.