The ShirtMath garment measurement dataset

What is in the dataset?

The dataset holds flat garment measurements for blank T-shirt styles, copied from each manufacturer's own published specification sheet. The first release covers eight styles in sizes S to XL. Eight points of measure are modelled per size: chest width, body length from high point shoulder, shoulder width, sleeve length, sleeve opening, bicep, bottom hem width and neck opening.

Not every style publishes all eight. Where a manufacturer does not publish a point of measure, the field stays blank rather than filled in. A blank cell is information; a guessed cell is noise.

What do the provenance labels mean?

Every figure carries one of three labels, and the label travels with the number wherever it is shown.

1. Published specification — copied from the manufacturer's spec sheet, with the date it was retrieved. The strongest label.
2. Measured — a garment laid flat and measured with a tape. Strong for that garment, but one garment is one sample.
3. Estimated — derived from neighbouring sizes or a related style. The weakest label, and flagged so you can exclude it.

Published figures also drift. A brand can revise a spec sheet without announcing it, so the retrieval date is part of the record.

How is it versioned and licensed?

Each release carries a date and a changelog entry. Figures are never edited in place: a correction becomes a new release, with the change named below.

The numbers themselves are the brands' own published specifications, and ShirtMath claims no ownership of them. What ShirtMath publishes is the compilation: normalised points of measure, provenance labelling, and a structure that makes styles comparable. That compilation is offered for reuse with attribution. Check each manufacturer's own terms before republishing their spec sheets wholesale.

How should I cite it?

Copy this, changing the release date if you are citing a later version.

Where do I start?

Changelog