Zebrash reference labels

The corpus published with the zebrash renderer, plus its per-command text and barcode cases. The corpus is published here.

The suite renders every label below on each test run, and compares the output with Labelary's for the same input. The status states what matters for a label: the layout is the same and every barcode carries the same data. The percentage is the share of pixels that differ — rendering detail such as font rasterization — and it does not set the status.

Summary

The widest difference in the corpus is 4.71%, on Kmart.

A label with a content or blocking known issue reads known defect, whatever it measures. Every other label reads pass: the layout and the barcode data match. A cosmetic issue names a visible rendering difference without changing what the label says.

Labels

Known defects first, then the widest rendering difference.

Known issues

1 of the 46 labels carries an annotated defect, 1 defect in all. The severity comes from the label, not from the percentage:

• blocking means content a reader misses;
• content means the output differs in what it says;
• cosmetic means it differs only in how it looks.

Cosmetic defects

• Dhlparceluk, 3.83% — Glyph outlines differ across the twenty-six field blocks. Per-tile alignment removes a fifth of the measured difference, and the reverse-printed postcode panel sits a couple of dots away. Owned by docs/company/plan/phase-5/09-fonts-objects-stored-formats/README.md.