What the standard dice set is for
The polyhedral set offered here, d4, d6, d8, d10, d12, d20 and d100, comes from tabletop roleplaying games. Each die is a uniform distribution over its faces: a d20 gives every value from 1 to 20 with equal probability of one in twenty, and a d6 gives each face a one in six chance.
The d20 dominates because it is the resolution die in the most widely played systems: roll, add a modifier, compare against a target number. Its flat distribution means every outcome is equally likely, so a natural 1 and a natural 20 each come up five percent of the time. Damage rolls use the smaller dice, and the d100, usually read as two d10s, covers percentage tables.
Notation follows the same convention: 3d6 means roll three six-sided dice and add them, and the tool writes its history in that form so a result reads as 3d6 [4, 2, 6] = 12. The individual values are shown as well as the total, because in many systems each die matters separately.
Why rolling several dice is not the same as rolling one bigger one
A single d20 is flat: every value is equally likely. Three d6 summed is not. The total ranges from 3 to 18, but there is only one combination that gives 3 and twenty-seven that give 10, so the middle of the range is far more common than the ends. The probability of rolling 3 on 3d6 is one in 216, while 10 comes up about one time in eight.
This is the central limit theorem showing up in a very small sample: add more dice and the distribution of the sum bunches harder around the mean. Game designers use it deliberately. A flat d20 makes outcomes swingy and dramatic, while a 3d6 system makes characters behave close to their average most of the time and makes extreme results genuinely rare.
The same reasoning explains why 2d10 and 1d20 feel so different despite covering similar ranges. On 2d10 the value 11 is ten times more likely than 2, so a threshold set at 11 is a very different proposition from the same threshold on a flat die.
Where the random numbers come from
Each die is rolled with the browser's Math.random, scaled to the number of faces and rounded down, which gives a uniform result across the faces. For games, puzzles and any casual decision this is entirely adequate: the distribution is even and consecutive rolls are not correlated in any way a player could exploit.
It is not a cryptographically secure source. Math.random is a pseudo-random generator whose internal state could in principle be inferred by an observer who collected enough output, and browser vendors document that it must not be used for security purposes. That matters for key generation and for anything with money at stake, and not at all for deciding who goes first.
If you need verifiable fairness, a digital roller of any kind is the wrong instrument, because the person running the page controls the code. Physical dice rolled in view of everyone, or a commit-and-reveal scheme where each participant commits to a hidden value before any are revealed, are the standard answers when trust is actually in question.
Reading the history and thinking about streaks
The tool keeps the last twenty rolls with their notation, individual values and totals, which is useful for checking a sequence after the fact and for keeping track during a turn that involves several rolls.
That history is also a good place to observe how misleading intuition about randomness is. Runs happen: four 20s in a row on a d20 has a probability of about one in 160,000, which sounds impossible until you consider how many d20 rolls are made worldwide every evening. In any long sequence, streaks that look impossible are expected.
The gambler's fallacy is the corresponding mistake in the other direction. A die has no memory, so a d6 that has come up 6 five times running is exactly as likely to roll a 6 on the next throw as it was on the first. Nothing is due, and no sequence of past results changes the next one.