In another discussion I was trying to explain what is meant by "does not compare the final grades".
The whole point behind all these dimensions and vectors is actually pretty simple. If you are using the accumulative grading strategy with two aspects, every filled assessment form can be represented by a tuple of numbers - that is the 2-dimensional vector. The "normalized" means that the method does not really care if the aspect is graded using a three point scale or a number from 0 to 20. It normalizes both assessed aspects into two number from 0 to 100%.
For example, let us say you have the first aspect assessed using a scale "poor, average, good" and the second aspect using a numerical value from 0 to 20. Let us say Alice fills your form with the values "average" for the first aspect and 15 for the second one. For now, let us presume both aspects have the same weight set to 1.
The grading evaluation plugin "sees" such filled assessment form as two numbers (i.e. as two-dimensional vector):
It does not really care if the first of these numbers were obtained via a scale or a number.
So the "dimension" is a general concept describing items at any assessment form.
The sentence with "distance of two assessments" is not that clear in the docs, I just realized. The grading evaluation method firstly decides which of all peer assessments of a single submission should be considered as a referential ("best"). It then calculates, how different each of assessments is from this referential one and calls that difference as a "distance". The bigger distance is, the lower grading grade is given.