Is this die fair, or is the table arguing?

What the math is doing

The test starts by assuming the die is fair. Each face should show up with the same chance, so with n rolls and k faces, each face is expected about n / k times. The chi-square statistic adds up (observed minus expected) squared, divided by expected for every face. A small total means the counts sit close to the expected values. A large total means the spread is hard to explain by chance.

The p-value turns that total into a probability. It answers: if the die were fair, how often would a result this uneven appear? This page calls a die suspicious when the p-value falls below 0.05. That level means roughly one fair die in twenty would still trip the test by luck, so one failed session is a reason to retest, not a verdict.

A short walkthrough

Say a player suspects a d6 favors sixes. They roll it 120 times and log every result. Expected count per face is 20. Suppose sixes show up 31 times and ones show up 12. The page compares each face to 20, adds up the squared gaps, and reports a low p-value. The verdict reads "Possibly loaded" and the chart highlights sixes. The player then rolls 120 fresh times in a new session. If sixes settle back near 20, the first result was probably noise. If they stay high, the die deserves a closer look.

A fair rolling protocol

• Roll from a cup or tray, not your palm, so the throw is consistent.
• Roll onto a surface with a bit of felt or cloth, not hard wood that bounces unevenly.
• Use the same hand and similar height for the whole session.
• Do not pick the die up from the same face each time.
• Log results as they land. Do not skip rolls that look odd.

Common causes of apparent bias

Not every odd pattern is a loaded die. Worn or chipped pips, unequal ink fill, rounded edges, and mold lines can change how a die lands. Dice with a slight tilt in the mold may favor certain faces. Heavy resin dice sometimes sink toward one side. Rolling habits also matter. A roll that is always thrown from the same place can favor a particular face because of the surface.

Common mistakes include stopping the test early, comparing several dice at once without adjusting for that, and logging only the rolls that felt right. Keep the full log, decide the sample size before you start, and retest before you call a die loaded.

Notes and limits

This tool checks one die at a time against equal odds for each face. It does not detect other forms of cheating, and it cannot tell why a die is uneven. Last reviewed for version 1.0 in 2026. Browser storage can be cleared by your browser settings, so copy a summary if you want a permanent record.