moonbitTolerance

Deterministic mechanical tolerance stack-up analysis for MoonBit. The library is intended for CAD, process planning, and quality tooling that need a small, reusable calculation core rather than a full CAD system.

What is included

• Signed one-dimensional dimension chains.
• Worst-case stack-up and RSS (3-sigma) estimates.
• Deterministic Monte Carlo sampling with an explicit seed.
• Mean, interval, standard deviation, yield rate, and sensitivity data.
• A tiny native CLI demo under cmd/main.

The first release deliberately keeps the public model small. Two-dimensional projected chains, richer distribution models, JSON/CSV adapters, and GD&T constraints are planned extension points; they are not represented as unfinished API placeholders.

Quick start

moon test --target wasm-gc --deny-warn
moon run cmd/main

let chain = @moonbitTolerance.Chain::new("shaft", [
  @moonbitTolerance.Dimension::new("housing", 20.0, 0.05),
  @moonbitTolerance.Dimension::new(
    "cover", 0.2, 0.02,
    direction=@moonbitTolerance.negative_direction(),
  ),
])
let result = chain.monte_carlo(10000, seed=42U)

Engineering notes

worst_case sums absolute tolerance contributions. rss treats each tolerance as a symmetric independent contributor and reports a three-sigma interval. monte_carlo uses a fixed 32-bit linear-congruential generator; the same chain, sample count, and seed produce the same result on supported backends. These assumptions are explicit so downstream applications can replace them when their manufacturing data requires another distribution.

Development

The repository is licensed under Apache-2.0. CI checks formatting, warning-free type checking, interface generation, and tests with MoonBit’s current stable toolchain. See proposal.md for the one-page hackathon application.

Scope boundary

This package is a calculation kernel. It does not parse drawing files, claim conformance to a particular GD&T standard, or replace a calibrated measurement system.