Random.Range... biased?

Questions & Answers
1

Hi, I’m currently using this to generate a random capital letter between A to Z, as follows:

char letter = (char) Random.Range(65, 91); 
//Where 'A' is 65 in ASCII
//and 'Z' is 90 in ASCII
//And, reading the documentation, 91 should be ignored, since the (int, int) version of Random.Range ignores MAX

However, for the past ~80 times I’ve run this, I almost always NEVER get an A or Z (or anywhere close to them). Yes, As, Bs, Cs and Xs, Ys and Zs appear, but rarely, VERY rarely, within the first 30 seconds; More often than not its anything in the center, I, T, N, M, J and such, with the letters closest to A and Z appearing only much later down the line.

The sequence does not repeat itself so I’m pretty sure I’m using Random.Range correctly. Besides, IIRC, Random.Range seeds itself on creation and doesn’t need to be explicitly seeded with system time.

Am I just lucky enough to always hit a sequence where min/max hardly appears (I know that’s very possible for a Random, but extreme cases should not be so common!), or is Random.Range biased away from min and max? If so, any alternative that isn’t?

2

Well, I have done a lot of simulation tests with the Random function.
I generated a random number between 1 and 10, keeping the count for each number.

After a couple of billion loops of simulation, all the numbers have about the same count.
This means that the border values are not generated less.

It’s completely normal that some values will not appear for a while, it’s really “random”.
Random.Range is not biased as far as I know.

3

Thanks for the answer.

I understand an RNG is random enough if, over a long period, the sequence has a long enough loop time, and each number in the sequence appears about the same number of times over said period.

I’m just not sure if it’s really random over the short term, given the luck of the draw I’ve had (note: I don’t need to keep an RNG running for a long period). I should be getting edge cases right off the bat or near it a bit more often, but I have no reliable way to test for that other than “run it another X times for ~30 seconds each, and count the results”.