Quality
The SVGZO Quality Score
Nine measurable components, weighted to 100. The badges below are the definitions a line is measured against; each is awarded to a dataset only once the check has been run over the whole line. A composite score is published per dataset once the formula is frozen.
- SVG Validated
- Live Text
- Duplicate Checked
- Metadata Complete
- Provenance Documented
- Licence Verified
Components
| Component | Weight | Measure |
|---|---|---|
| SVG validity | 15 | Parses; viewBox and dimensions present; renders in resvg without error |
| Rendering consistency | 10 | resvg versus librsvg pixel agreement above threshold |
| Text editability | 15 | Share of files where all visible copy is live text |
| Geometry integrity | 10 | No raster images; path count within band; no empty groups |
| Metadata completeness | 15 | Share of schema fields populated |
| Duplicate rate | 15 | Perceptual-hash near-duplicates and structural exact duplicates below stated thresholds |
| Taxonomy quality | 10 | Every asset mapped to sector, subject and format |
| Provenance | 5 | Per-asset source and licence present |
| Licence clarity | 5 | Tier licence attached; notice bundle complete |
Badge definitions
- SVG Validated: every file in the line parses and renders in resvg. A line is badged only once the whole line has been run, not a sample of it.
- Live Text: at least 99% of files carry all visible copy as text nodes.
- Duplicate Checked: exact and near-duplicate rates measured and published.
- Metadata Complete: at least 95% of record fields populated across the dataset.
- Provenance Documented: per-asset manifest present.
- Licence Verified: tier text attached and notice bundle complete.
Tooling
resvg for deterministic rendering, librsvg for cross-checks, svglint and OXVG for structural lint, a canonicalised-DOM SHA-256 for exact duplicates, and perceptual hashing over renders for near-duplicates. The ingest code that produced the previews on this site is the same code that will score the full catalogue.