"same question using random numbers"
I would say mainly because random numbers will generate random responses and some of them could give with a given
formula invalid responses.
This is why the responses from the various the datasets values can be checked by the teacher.
There is no easy way to analyze that a given formula will not give a bad result even without math error like divide by zero, as the result could be a nonsense.
Otherwise if you use the questions for training and in which case there is no real problem if one of the question repeat itself you could use the following process.
Use calculated simple to create a first question in an empty category.
Save it
Edit , generate new values, save a new.
Repeat the process to obtain say 10 questions which then contain different data (given you 1000 variants).
Then export these questions and reimport them in the same category so you have a total of 20 questions.
Repeat you have 40 questions.
Then you build your quiz using 10 random questions.
The probability that a given student has 2 identical set of values is low but, I agree, not zero.
However you could experiment by creating even calculated questions using a spreadsheet and import them in moodle.
Pierre
P.S. Core question types should be as fool-proof as possible.
P.S. Look at the new numeric question types from OU.