SchemaScope — schemascope.org

A cross-domain catalog of ontologies, vocabularies, schemas and data standards. It is harvested from the major registries, extended by discovering ontologies that no registry lists, and evaluated automatically for FAIRness, ontology quality, maintenance and adoption. Every record keeps its provenance. The site is fully static.

Quick start

pip install -r requirements.txt     # rdflib, for harvesting, discovery and pitfall scans
npm run harvest                     # registries → data/catalog.json (+ data/sources.json)
npm run discover                    # dereference prefix.cc, w3id.org, Zenodo, GitHub, data/candidates.txt
npm run refresh                     # quality checks + evaluation + build + commit (add -- --push to publish)
npm run quality:all                 # ROBOT, pitfalls and imports for every ontology; resumable (QUALITY_JOBS=2 for two at a time)
npm run rescore                     # recompute scores from stored facts after a rubric change (no network)
npm run validate:rubric             # weight sensitivity, OBO Dashboard agreement, expert ratings → data/validation/report.json
npm run serve                       # preview at http://localhost:3300

Requirements: Node ≥ 18, Python 3 with rdflib, and Docker. Docker runs OOPS! and ROBOT locally, so Java isn't needed. Set GITHUB_TOKEN (or log in with gh) before evaluating; it raises GitHub's API limit and enables the GitHub discovery channel.

Where the catalog comes from

Source	What is taken	How
Bioregistry	semantic-artefact prefixes, metadata, cross-registry mappings	registry.json export
OBO Foundry	ontologies, domains, status, licences	registry JSON-LD
OLS (EMBL-EBI), TIB Terminology Service	every loaded ontology, counts, imports	OLS v2 API
LOV	~900 semantic-web vocabularies, tags, reuse counts, version files	full RDF dump
NeuroSchema	curated neuroscience entries	neuroschema.org/catalog.json
Discovery	ontologies no registry lists	scripts/discover.py dereferences every prefix.cc and w3id.org namespace, Zenodo ontology files, GitHub ontology repositories and data/candidates.txt (e.g. ontologies found in papers). A candidate is accepted only if its artefact declares an ontology with a title and ≥ 5 terms.

Records are merged across sources by prefix, cross-registry mappings, ontology or namespace IRI, and normalised name on the same host. Curated entries and overrides in data/curated.json take precedence over registry values.

Provenance.

• Per field: each catalog record lists every registry that supplied it, with URL and retrieval time, and which registry supplied each field.
• Per download: data/sources.json records each registry download (URL, SHA-256, size, licence), and data/sources/*.gz archives the raw files, so python3 scripts/harvest.py --offline reproduces the catalog exactly.
• Per evaluation: each evaluation run is archived in data/runs/.

Domains

There are 15 domains, from foundational ontologies to privacy & law (see data/categories.json). Each record has a primary domain and optional secondary domains, which drive the cross-domain map on the home page. Domains come from OBO Foundry domains, LOV tags, TIB classifications and keyword rules.

Scoring

Score = max(0, 100 · (F + A + I + R + Q + M + U) / possible − P). FAIR is 4 × 12 points. Q (ontology consistency & quality, 20) applies to ontologies only; for other resources possible is 80 and the result is scaled. Q checks SchemaScope couldn't run are left out of possible too (rubric v4, "not assessed"), and such grades are marked with *. M is maintenance (16), U is adoption (16), and P is penalties for inactivity, inconsistency and critical pitfalls. The rules are in scripts/rubric.mjs, and the site's /evaluation/ page shows the formula and a worked example.

Contributing

Use the site's /submit/ form, which opens a GitHub issue that a bot turns into a pull request. You can also add a URL to data/candidates.txt, or edit data/curated.json, and open a pull request. CI validates every PR and posts a scoring preview. Evaluation results are produced only by the maintainer's local npm run refresh, so they never conflict.

Deploy

Vercel builds the site with node scripts/build.mjs into dist/ (see vercel.json). Set the Application Preset to Other and leave the install command empty.

License

The SchemaScope code and site are licensed under the Apache License 2.0. Catalogued ontologies and schemas belong to their maintainers and keep their own licences, which are shown on each record. Registry data keeps the licence recorded in data/sources.json.