Two things everybody gets wrong about random numbers
Generating a number in a range looks like the simplest job in programming. It is the source of two bugs that ship constantly, and both of them are invisible until somebody checks.
"No repeats" is a different operation, not a filter
The obvious way to get six different numbers is to draw six and throw away any duplicates, drawing again to replace them. It works, it looks right, and it falls apart at the edges.
As the count approaches the size of the range, the odds of drawing a number you already have climb towards certain. Pulling all forty-nine numbers from one to forty-nine that way takes about two hundred and twenty draws to place the last one. And if you ask for more numbers than the range contains, it never finishes at all: the loop just runs, and the tab dies.
The right answer is a shuffle. Put the whole range in order, shuffle it, take as many as you need. It finishes in exactly that many steps whatever you ask for, and it refuses the impossible case with a message instead of a hang.
Modulo bias, the bug in almost every hand-rolled generator
Take a random 32-bit number and use the remainder after dividing by your range. It is the first thing anybody writes and it is not uniform.
Unless your range divides exactly into 4,294,967,296, the low values get one extra chance each and come up slightly more often. On a range of 1 to 6 the bias is far too small to notice. On larger ranges, or over enough draws, it is real and measurable.
The fix is to discard the leftover tail and draw again, which is what the unpredictable mode here does. Throwing perfectly good random numbers away feels wasteful and is the difference between uniform and nearly uniform.
Predictable and unpredictable are both correct answers
A seeded generator produces the same sequence every time from the same starting number. That sounds like a flaw and is usually the feature you want.
If you draw a name and somebody disputes it, a seeded draw can be reproduced in front of them. The seed is printed with the result. That turns an argument into a demonstration.
The catch is that a seeded generator is predictable by design. Anybody who works out the state knows what comes next. For a rota, a seating plan or a game, nobody cares. For a prize with real money attached, it is the whole problem, and that is what the unpredictable mode exists for.
Most tools pick one and say nothing. The honest answer is not "ours is random", it is "which kind of random do you actually need".
You can check it instead of trusting it
Every generator on the internet claims to be uniform. Almost none of them let you look.
Draw a million numbers here and the page shows how often each value came up, how flat it should be, and how much variation to expect from chance alone. That last number matters: a result with no variation at all would be more suspicious than a small amount, and without something to compare against a figure on its own tells you nothing.
Common questions
How do I know the numbers are actually uniform?
Press the million button. It draws a number a million times and shows how often each value came up, next to how flat it should be and how much wobble to expect from chance alone. Every random number site claims uniformity. This one lets you check it in about a second.
What does "allow the same number more than once" do?
With it on, every draw is independent, so duplicates happen exactly as often as chance says they should. With it off you get distinct numbers, which is a different operation entirely: it shuffles the whole range and takes the first few, rather than drawing and discarding duplicates.
Why does that difference matter?
Because drawing and discarding gets dramatically slower as the count approaches the size of the range. Pulling all 49 numbers from 1 to 49 that way needs about 220 draws to place the last one, and asking for one more than the range holds never finishes at all. A shuffle finishes in exactly as many steps as numbers you asked for, every time.
What is the unpredictable mode for?
Anything where somebody has an incentive to guess the result. The normal mode uses a seeded generator, which is fast, uniform and completely predictable if you know the seed. The unpredictable mode draws from the operating system instead, so nobody can work out what comes next, including you. The cost is that the draw cannot be reproduced afterwards.
Which one should I use?
Seeded for almost everything, because reproducibility beats unpredictability when somebody might dispute the result. A draw you can reproduce in front of the person who lost is worth more than one you cannot. Switch to unpredictable when there is money on the outcome and somebody has a reason to game it.
What is modulo bias?
The commonest bug in hand-written random number code. Taking a random 32-bit value and using the remainder after dividing by your range makes the low numbers come up slightly more often, unless the range happens to divide evenly into 2^32. This throws away the part that does not divide cleanly and draws again, which is the only way to get a genuinely even integer out of random bytes.
Is anything sent anywhere?
No. The numbers are generated in your browser, there is no account, and nothing is stored or transmitted. Close the tab and it is gone.