Back to the explorer

Methodology

How the data is built

MusiGraph queries open, linked sources in real time. Auto mode is the default: Wikidata is the core (identity, base data and identifiers) with automatic failover to QLever, and the result is enriched with Wikipedia, DBpedia, MusicBrainz and Discogs. Every piece of data keeps its origin and is labelled with a source badge.

Data flow in Auto mode

The Wikidata core is essential and answers first; enrichment sources are queried in parallel, never block the profile, and each one reports its status in the final merge.

MusiGraph data flow in Auto modeThe UI calls /api/sparql, which validates and caches the request; the Auto orchestrator queries Wikidata with failover to QLever and, in parallel, Wikipedia, DBpedia, MusicBrainz and Discogs; the result is merged with provenance and feeds the profile tabs.UIUIsearch · profileAPIPOST /api/sparqlvalidation · cache · dispatchAUTOAutoorchestrator · provenanceCOREWikidatacore · failover: QLeverRESTEnrichment sourceswikipedia · dbpedia es · musicbrainz · discogsMERGEMerge + provenanceP436 key ↔ release-groupUIProfile tabsOverview · Discography · Influences · Collaborations

Auto mode rules

The orchestrator combines sources field by field, deduplicates rows and keeps the origin of every piece of data. These are the rules it applies.

Field precedence rules in Auto mode.
FieldPrecedence order
NameWikidata label in the UI language; if missing, a multilingual fallback label.
BiographyWikipedia in the UI language → DBpedia ES (Spanish UI only) → short Wikidata description.
ImageWikidata P18 image, served as a thumbnail; no further fallbacks today.
DiscographyMusicBrainz → Wikidata → Discogs, backfilling empty fields (date, label, genre and type).
GenresUnion of Wikidata and curated MusicBrainz genres, deduplicated; folksonomy tags are discarded.
InfluencesWikidata P737 and P1066 + real MusicBrainz relations (teacher, tribute), with their relation type.
CollaborationsWikidata works with multiple performers (P175) + MusicBrainz collaboration relations.
SuggestionsGenre similarity is labelled as a suggestion; it is never presented as a verified influence.

Deduplication

Discographies are deduplicated with two keys: the Wikidata P436 identifier ↔ MusicBrainz release-group as the strong key and, failing that, normalised title (no accents, lowercase) plus year.

Statuses and provenance

Provenance badges show each source's outcome: ok, empty, error, timeout or skipped. Enrichment sources are disposable: if one fails, the profile still renders and only its status is flagged.

Data sources

What each source contributes, through which endpoint and under which license.

Wikidata

Contributes
Artist identity (QID), per-language labels, dates, country, genres, instruments, partial discography, influences (P737/P1066), collaborations and external identifiers (MusicBrainz, Discogs, Spotify, Last.fm, AllMusic).
Endpoint
query.wikidata.org/sparql (Blazegraph, WDQS v1)

QLever

Contributes
The same Wikidata database served by QLever. It kicks in as automatic failover when the core fails or times out, and is also available as an explicit source in the selector.
Endpoint
qlever.dev/api/wikidata (community trial endpoint)

Wikipedia

Contributes
Biography in the UI language, from the artist's linked article (sitelink). Shown with attribution and a link to the original article.
Endpoint
{lang}.wikipedia.org/api/rest_v1/page/summary (REST)

DBpedia

Contributes
Spanish abstract as a biography fallback. Also available as an explicit provider for search and traits.
Endpoint
es.dbpedia.org/sparql and dbpedia.org/sparql (Virtuoso)

MusicBrainz

Contributes
Discography (release groups: albums, singles and EPs) with date and type; curated genres; real artist relations: member of band, collaboration, teacher and tribute.
Endpoint
musicbrainz.org/ws/2 (REST)

Discogs

Contributes
Releases (up to 100 per artist) with year, label and format; artist identity search when there is no Discogs ID.
Endpoint
api.discogs.com (REST, personal token)

Limits and resilience

Public quotas for every API and how MusiGraph mitigates them.

Public limits for each source and the mitigations applied in MusiGraph.
SourcePublic limitMitigation in MusiGraph
Wikidata (WDQS)≈60 s of processing per minute and 5 parallel queries per IP.1 h server cache and 10 min client cache; 12 s timeout with failover to QLever; 2–3 queries per view.
QLeverCommunity trial endpoint, no service guarantee.Only used as failover or explicit source; 8 s timeout.
DBpedia (Virtuoso)Can return 200 with partial results.Light queries and batched enrichment; aggregates are never trusted; 6 s timeout.
Wikipedia (REST)Public read API, no strict quota.6 s timeout and 6 h cache per language and title.
MusicBrainz1 request per second.Queue with 1.1 s spacing and a dedicated User-Agent; 24 h cache; never blocks the profile.
Discogs≈60 requests per minute authenticated.Server-only token; 24 h cache; the source disables itself without a token.

WDQS v2 migration

Wikidata's v2 endpoints (query-next.wikidata.org) open to the community in November 2026 and the current ones (Blazegraph) will start degrading in February 2027. QLever already works as failover and as an explicit selector option, and every query is written to run on both engines: no label service on QLever (OPTIONAL/COALESCE blocks) and strict xsd:dateTime dates. The endpoint is switchable through server configuration, with no code changes, so the cutover and rollback happen when the v2 endpoints go public.