Dice Calculation Methodology
The calculators answer the rule you enter, using the assumptions shown beside each tool. This page explains how exact results, random samples and statistical estimates differ.
Exact outcome counts and mixed dice
For independent fair dice, every ordered face combination has the same probability. Each die’s face distribution is added to the running total distribution by convolution. The resulting integer counts are summed for the requested event and divided by the total outcome count. BigInt arithmetic preserves integer counts; reduced fractions avoid premature rounding. Constants shift totals without changing weights.
For different dice sizes, multiply their face counts to obtain the sample space. A D8 and D6 have 48 ordered outcomes, not 36. Comparisons of independent rolls count pairs where the first total is higher, tied or lower.
Rounding, very small probabilities and spread
Displayed probabilities normally round to two percentage decimal places. A positive chance smaller than 0.01% is labelled <0.01%; a chance short of certainty by less than 0.01% is labelled >99.99%. Exact zero and one remain 0% and 100%. Fractions retain the exact distinction. Decimal means are rounded for display; standard deviations require an approximate square root.
Exported distribution CSV contains exact fractions plus numerical percentage values. Use the fraction columns when checking very large outcome counts or reproducing the calculation precisely.
Rules, limits and numerical estimates
Supported models have visible size limits to bound computation, rather than pretending every game rule is accepted. Independent success counts use a binomial model. Keep and reroll rules count finite weighted outcomes; infinite maximum-face explosions use an exact finite lower tail and its complement.
The fairness check is different: it uses an approximate chi-square p-value from observed counts. Independence and adequate expected counts matter. It cannot prove a die’s physical balance. Healing caps and damage floors apply to individual outcomes before averaging.
Random rolls and browser privacy
The roller samples with the browser’s cryptographic random generator and rejection sampling. This avoids biased remainder mapping of random integers to faces. It does not prove any physical die or external gambling service fair. Inputs and results run in the browser, and shared input links use URL fragments rather than tracking parameters. No calculator account or upload is required.
Reference checks you can reproduce
| Tool / rule | Inputs | Expected result |
|---|---|---|
| Probability | 2d6, exactly 7 | 1/6 |
| Range probability | 3d6, 9–12 inclusive | 13/27 |
| Mixed dice probability | 1d8 + 1d6 + 2, at least 10 | 9/16 |
| Average | 1d8 + 1d4 − 1 | 6 |
| Advantage | Two D20s, keep highest one, at least 15 | 51% |
| Repeated rolls | At least one six in four D6 rolls | 671/1296 |
| Conditional | 2d6 sum 7, given at least one 6 | 2/11 |
| Matching faces | Five D6s all equal in one roll | 1/1296 |
Automated regression tests check these cases, brute-force small dice spaces against the counting algorithms, validation boundaries and numerical reference values. Browser tests also submit the tools and check interaction behaviour. These checks support the stated models; they do not certify every rule in a game or prove field performance.
Mathematical references and corrections
- OpenStax: probability terminology and sample spaces
- OpenStax: expected value and standard deviation
- OpenStax: binomial distribution
- NIST: chi-square goodness-of-fit test
These sources explain the mathematical principles; they are not endorsements of this website. To report a discrepancy, send the tool name, inputs, expected rule and result. No author credentials or independent certification are claimed.
