physics-lint

A linter for trained neural PDE surrogates. Runs in CI, catches physics violations on every model PR, and produces actionable SARIF reports for GitHub code scanning.

Getting started

• Installation
  • Python version
  • Optional dependencies
  • Pinning a version
  • Virtualenv
• Quick start
  • Lint a model dump
  • Lint via an adapter
  • CI-style SARIF output
  • Configuration
• Loading models
  • Adapter mode
  • Dump mode
  • Which mode is right?

Reference

• Rule Catalog
  • PH-BC
  • PH-CON
  • PH-NUM
  • PH-POS
  • PH-RES
  • PH-SYM
  • PH-VAR
• Case studies
  • Case Study 01 — LagrangeBench
  • Case Study 02 — PhysicsNeMo MGN
  • Summary
• GitHub Action
  • Convention note
  • Minimal example
  • Matrix example (multi-model)
  • Inputs
  • Version pinning
  • Permissions
  • Advanced usage
• SARIF schema reference
  • What is covered by semver
  • Result schema
  • Field reference
  • Status values
  • Categories
  • Example: SARIF in CI

Project

• Stability policy
• Security notes
• FAQ
• Contributing

Quick start

pip install physics-lint
physics-lint check my_model.py --format text

For CI integration, see the GitHub Action.

For the full rule catalog, see Rules.

For the published case studies showing physics-lint detecting real issues on published checkpoints, see Case studies.

Source

physics-lint is open-source at github.com/tyy0811/physics-lint.