Four levels. Endless possibilities. Choose a module, find its angle, and let physics do the rest.
TOWER BLUEPRINTTOP → BOTTOM
15°55°
A sloping surface redirects the die and turns a falling motion into a tumble.
THE CONDITIONSRIGID-BODY / D6
SoftLively
SlipperyGrippy
02 / CHAOS FORENSICS
Every fall has a story to tell.
Pause at a collision. Follow a sudden turn. The trajectory is measured from the simulated motion, one frame at a time.
↗ The cyan line in the tower is your recorded path.
ROLL RECORDERNO RECORDING YET
0.00 s—
Each line marks a new contact event.
03 / THE BUTTERFLY CHAMBER
Just 0.1°. A world of difference?
Identical starting conditions. One tiny change to the selected module. Compare two real journeys, side by side.
WORLD AYOUR DESIGN
—Top face
Run an experiment to reveal the path.
+0.1°
WORLD BONE ANGLE CHANGED
—Top face
A different path is possible, not guaranteed.
04 / THE FAIRNESS PARADOX
When does chaos become fair?
One roll is a moment. A hundred rolls begin a conversation. Run a batch and see what your tower really does.
Ideal D6 probability: 16.67% per face. Jams and tilted rests are reported separately.
OBSERVED DISTRIBUTION Uniform reference
16.67%
Your first batch is waiting.
A balanced sample is an observation, not a proof of fairness. Initial conditions, sample size and the physical model all matter.
05 / THE TOWER COLLECTION
A cabinet of possibilities.
Begin with a little borrowed ingenuity. Then make it entirely yours.
01ACRYLIC / BRASS
Glass Labyrinth
A clear path through four alternating ramps. The classic study in changing direction.
02BRASS / KINETIC
Brass Cascade
Bumpers, stairs and a splitter. An intricate conversation between motion and rebound.
03OAK / MINIMAL
Zen Cascade
Two softer ramps. More open space. A quieter way to study friction and free fall.
03→ 08
06 / THE MINIMALIST CHALLENGE
Less machine. More imagination.
Use three modules or fewer. Reach eight contact events. Settle flat in the tray within fifteen seconds. Can you design a little more with a little less?
A dice tower guides a die through a sequence of obstacles before it reaches a tray. Unlike a simple random-number button, this laboratory computes gravity, contact forces, friction and angular motion. Changing a ramp changes the physical journey; the displayed face is read from the settled cube orientation. Opposite faces sum to seven.
Gravity, friction and restitution
Gravity accelerates the die downward at 9.81 m/s². Friction opposes relative sliding at contact. Restitution controls how much normal relative speed returns after a collision: larger values produce livelier rebounds. A cube also exchanges translational and rotational energy as corners and edges strike a ramp. Energy is dissipated by friction and damping.
Chaos is not the same as randomness.
A deterministic model follows rules. Yet small changes in initial pose or obstacle angle can grow through successive collisions. This sensitivity is the idea behind the butterfly experiment. Repeating the same seed, design and numerical settings in the same environment should reproduce the journey. Different browsers or engine versions may differ numerically; a seed is not a guarantee of cross-platform identity.
Try a controlled experiment
Start with Glass Labyrinth and seed 42. Record a roll, replay its contacts, then run the butterfly experiment. For a distribution study, keep all settings fixed and run 100 or more trials. Record the number of settled results as well as jams. Change only one setting and repeat. A finite sample cannot prove that a tower is fair, and these ideal cubes do not model the manufacturing defects of real dice.
Model notes: D6 only; four collision modules; fixed 1/120 s physics steps; 15 s timeout; educational rigid-body approximation. Unsupported future concepts such as moving pendulums and polyhedral dice are not presented as active features.
No. A seeded generator sets the initial pose and velocity. The cube then follows the physics simulation. The face with its normal pointing upward is read only after the cube settles in the tray. Jams and tilted rests have no face result.
Why does the same seed give the same roll?
The seed, tower design and material settings define a reproducible initial experiment in the same runtime. Increase the seed for a new roll. Batch trials automatically use consecutive seeds.
Does a balanced chart prove fairness?
No. Sampling error, initial conditions and the model all matter. Use larger samples, report excluded trials, and compare designs under the same conditions. This is an educational tool, not a certified dice tester.
Where are my designs stored?
Saved designs stay in this browser’s local storage. Sharing a link includes only the design parameters and seed in the URL fragment. No account or personal details are required. If local storage or clipboard access is blocked, an explanatory message appears.