Conformal

Conformal is a free static code analyzer for the MATLAB language. It tracks shapes across functions and files, resolves symbolic dimensions, and reports the exact line where a mismatch originates. Size errors are found deterministically using path-sensitive analysis and relational loop bounds. Conformal does not require a MATLAB license.

Try it in your browser Install for VS Code Download CLI GitHub

your code

function result = process(data, weights)
    scaled = data .* weights;
    result = scaled' * scaled;
end

A = rand(10, 3);
w = rand(4, 1);  % wrong size
R = process(A, w);

conformal output

W_ELEMENTWISE_MISMATCH line 2:
  data .* weights:
  matrix[10 x 3] vs matrix[4 x 1]
  (in process, called from line 8)

Shapes follow into the function
data dimensions are 10x3 and weights dimensions are 4x1
Line 2 is the location of the bug

Try it on real problems

The playground runs the Conformal analyzer on a single file in your own browser without needing to install it. Paste in code of your own, or try one of the examples there. Examples were chosen to cover a broad range of applications to show how Conformal can work for any discipline that uses MATLAB. Change variable values in place to see how Conformal reacts to new shapes and dimensions. Note: the playground runs entirely in your browser, so your code is never uploaded and the page cannot read or write files on your computer.

Open the playground

process.m

Conformal showing inline diagnostics in VS Code

Diagnostics appear as you type. Hover any variable to see its shape.

15,085 real-world files tested

0 crashes

615 test cases

<250ms per file

650+ builtins recognized

What it catches

Matrix sizes that do not line up in a multiply or an element-wise operation
Indexing past the end of an array
Mistakes that only show up when one function hands data to another, including across files
Size errors hiding inside structs, cell arrays, and class objects
Code that MATLAB Coder will refuse to compile when the option is enabled

Where to run it

Where	What you get
VS Code	Squiggles as you type, and variable sizes on hover
MATLAB command window	Checks the file open in the editor, without leaving the IDE
Terminal	One file, or a whole folder in a single pass
A build system	A machine-readable report, so a bad merge fails the build

Using Conformal from the MATLAB IDE

Conformal ships as a standalone CLI binary. To use it from MATLAB's command window:

Download the CLI
Get the conformal binary for your OS from GitHub Releases and put it on your system PATH. It's a single native executable with no dependencies.

Add conformal_check.m to the MATLAB path
Download conformal_check.m and save it anywhere on your MATLAB path. This is a small wrapper that calls the CLI binary and prints results in the command window.

Analyze
>> conformal_check to analyze the file that is currently open in the editor.
>> conformal_check('ekf_update.m') to analyze a specific file.

For a team: put both files on a shared drive and add it in each individual's startup.m.

Where it's been tested

15,085 .m files from 34 public projects, including chebfun (3,435 files), PlatEMO (2,416), MATPOWER (979), NASA's MUSCAT spacecraft toolkit, the KU CReSIS polar radar toolbox, LADAC aircraft dynamics, eeglab, and ecg-kit.

aerospace optimization numerical methods biomedical power systems geophysics computer vision signal processing CFD controls radar robotics rocketry

How the analysis works

Shapes are inferred over a small lattice and carried across function calls, files, nested functions, and closures. This allows mismatches deep in a call chain to trace back to where they were introduced. Struct fields, cell contents, classdef properties, and the nargin/nargout/varargin forms are all tracked. Dimensions can stay symbolic throughout the abstract interpretation, so zeros(n+1, 2*m) is held as matrix[(n+1) x (2*m)] and the algebraic expression survives the rest of the program instead of collapsing to unknown. Loop index bounds are tracked relationally, which is why for i=1:n-1; A(i+1) raises no false alarm. The analysis is path-sensitive, so shapes that only differ on a branch that can't run stay silent.

CI and toolchain flags

--format sarif produces SARIF 2.1.0 with SHA-256 file hashes for traceability. --batch dir/ sweeps a directory in one process. .conformal.json locks per-repo settings so a team can run the same checks. --coder flags constructs that MATLAB Coder can't compile. --fixpoint switches to widening-based iteration to analyze loops beyond a single-pass.