Sandbox · Distributions · Bézier
When the textbook distributions don't fit your data
Repair times with two failure modes. A long tail of rare events. A process that runs in shifts. Sometimes the data needs a richer shape than normal, gamma or Weibull can give it. A Bézier distribution bends to fit: fit one to your own data here, shape it by hand, check that it draws true, and take it into your model.
How it works
A Bézier distribution draws its cumulative curve with a handful of control points. Keep the points in order and the curve is always a valid distribution, so you can drag them anywhere and never break it. Several joined segments give each hump of a multi-modal distribution its own handles.
Suggest a fit solves for the points by constrained least squares; Refine then searches their positions (Nelder–Mead) and never makes the fit worse. The score is the Kolmogorov–Smirnov distance to your data, with a ✓ when it is inside the 95% critical value. Draw samples from the curve and checks the draws against it, and Time it compares its speed with the standard samplers. A precomputed table makes a Bézier draw about as fast as theirs.
Save the fit as a Prime table to paste into a ReliaSim® database, as a fit file to reopen here, or as a CSV of its control points. Your data is read in your browser; nothing is uploaded.
The method is Wagner and Wilson's (1993, 1996), with fitting ideas from Leiva, Medaglia and Zuluaga's computational framework (2026). Every number on this page comes from the ReliaSim engine's own code, compiled to WebAssembly.