Dice Roller

In-Depth Guide

A dice roller simulates fair rolls of one or many dice of any polyhedral type — d4, d6, d8, d10, d12, d20, d100, and non-standard shapes up to d1000 — using the browser's crypto.getRandomValues, a cryptographically secure random-number generator. A single d6 produces each face with exactly 1/6 probability per the standard uniform distribution. Expected value for an n-sided die is (n+1)/2: 3.5 for d6, 10.5 for d20. For multi-dice rolls (3d6, 8d10), the sum distribution converges toward a normal distribution by the central limit theorem; for 3d6 the mean is 10.5, standard deviation ≈ 2.96. FastTool's dice roller supports standard tabletop notation (4d6, 2d20+5, 1d100), runs entirely in the browser, and produces results indistinguishable from physical dice within the limits of any statistical test.

Why This Matters

Physical dice are heavy, noisy, prone to getting lost, and often genuinely unfair — mass-produced plastic dice have measurable face biases from injection-moulding shrinkage around the pips. A software dice roller is auditable (a CSPRNG is provably uniform), portable (runs on any device), and convenient for large rolls where summing 8d10 by hand is slow. Tabletop RPG players, board-game rule prototypers, educators teaching probability, and researchers running Monte Carlo simulations all benefit from a reliable, fast, free roller. Privacy matters too: a web-based roller should not log your campaign's critical hits to a server.

Real-World Case Studies

• D&D remote session. A Dungeons & Dragons dungeon master running an online campaign uses the roller to resolve combat initiative (20 players × 1d20), damage rolls, and saving throws live on screen-share. The roller's deterministic display and history let all players see each result, eliminating the I rolled a 20 trust me problem that plagues remote play with physical dice off-camera.
• Probability classroom demo. A high-school maths teacher rolls 10,000 virtual d6 in a single batch to demonstrate the law of large numbers. The observed face distribution converges to within 0.5 percentage points of the theoretical 16.67% per face, and the sample mean approaches 3.5 as students watch. Physical dice could never produce this demonstration in a class period.
• Monte Carlo estimator. A statistician prototyping a simulation uses the roller to generate 100,000 d20 rolls for an estimate of P(rolling ≥ 15 on at least one of three attempts). The crypto-grade RNG is sufficient for illustration; she later re-runs in Python with the Mersenne Twister for production, but the browser-based draft helps her sanity-check the expected ~58% result.

Technical Deep Dive

crypto.getRandomValues(new Uint32Array(n)) fills an array with cryptographically secure 32-bit random integers, each uniformly distributed in [0, 2³²−1]. To roll an n-sided die, the classic method value % n introduces a tiny bias when n doesn't divide 2³² evenly (the so-called modulo bias). The rejection-sampling method — draw, reject values ≥ floor(2³² / n) × n, retry — produces a truly uniform distribution. For a d6, the rejection region is only about 4 in 2³² ≈ 10⁻⁹ rejection probability per draw, so the overhead is negligible. For exotic dice (d100, d1000), the same approach scales at O(1) amortised cost per roll. Sum distributions for multi-dice rolls are not uniform: 2d6 has a triangular distribution peaking at 7 (probability 6/36 = 16.67%), and 3d6 approximates a bell curve with mean 10.5 and standard deviation ≈ 2.96 — often mis-intuited as flat by new RPG players. By the central limit theorem, the sum of n independent uniform dice converges to a normal distribution as n grows, reaching visually-bell-shaped by about 4d6. For cryptographic applications (shuffling card games, secret sharing, random passphrase generation), crypto.getRandomValues is appropriate; Math.random() is not, because its underlying xorshift128+ algorithm is predictable given enough outputs.