Data & Provenance
Where the data comes from, and what you can do with it
Syllogeum aggregates open numismatic data — so the aggregate goes back to the commons. Each release below is the rights-clean subset of the corpus: only collections whose image licence clearly allows redistribution (American Numismatic Society, Münzkabinett Berlin and a few other public-domain, CC0 and CC BY-SA sources, listed with their credits in each release's LICENSE-DATA.md). It includes full metadata, obverse/reverse images, and — uniquely — the DINOv2 and SigLIP2 embedding vectors for every imaged record, ready for similarity research without a GPU.
CC-BY-NC-SA material (Corpus Nummorum, most museum cabinets), collections with NC, ND or unclear terms, and Europeana records with unresolved per-item rights are not redistributed — they remain viewable on the site with attribution. Full audit and reuse rules below.
syllogeum-open-v1.1
| File | Size | SHA-256 |
|---|---|---|
| LICENSE-DATA.md | 3.3 KB | 04b0cc10a1a871c9… |
| README.md | 2.0 KB | 92ffebc85ca61839… |
| embeddings_dino_obv.npz | 348.7 MB | 3a76b08d5239be24… |
| embeddings_dino_rev.npz | 348.7 MB | cb11eed7ba5a0c49… |
| embeddings_siglip_obv.npz | 348.7 MB | 81d5b9c23a32d920… |
| embeddings_siglip_rev.npz | 348.7 MB | 8fe3ea2488e0d770… |
| images.tar | 8.3 GB | 6151b416d0e24856… |
| sightings.csv.gz | 4.5 MB | 13afd9eecd4a2559… |
| sightings.jsonl.gz | 4.7 MB | 13d82a2f974da33d… |
| types.jsonl.gz | 1010.1 KB | c428f96394c53840… |
| manifest.json | full checksums |
syllogeum-open-v1 was withdrawn on 2026-09-23 and replaced by syllogeum-open-v1.1, because it included images from collections whose licences do not clearly allow redistribution.
Provenance & rights
Syllogeum aggregates open numismatic data. That only works honestly if the provenance of every record is visible and the license terms of every source are respected — including the sources that said no. This section is the full accounting: the audit method, the license breakdown, the opt-out policy, and the AI disclosure statement.
Method
Every sighting carries an attribution label written by the harvester that ingested it — the canonical record of which collection it came from. The corpus was formally audited on 2026-07-01 by grouping all sightings by attribution and cross-checking against source URLs. The audit document ships in the source repository as PROVENANCE_AUDIT.md.
License buckets
| Bucket | Sightings | License / terms | Reuse |
|---|---|---|---|
| American Numismatic Society (Nomisma LOD) | 118,222 | Open / LOD | Open with attribution |
| Named museum collections (IKMK network, Ashmolean, KENOM, ~50 more) | ~82,900 | mostly CC-BY-NC-SA | Non-commercial, share-alike |
| Europeana-aggregated institutions | ~45,300 | per-item edmRights | Varies — see the open gap below |
| Corpus Nummorum (Zenodo dataset) | 42,516 | CC-BY-NC-SA | Non-commercial, share-alike |
| Nomisma RDF dumps (misc. collections) | ~15,900 | per collection | Checked collection by collection on 2026-09-23. Redistributed: Münzkabinett Berlin and Linz (public-domain images) and the Fralin Museum (public domain in the US), 609 records. The other ~15,300 records have NC, ND or unclear terms: viewable with attribution, not redistributed. |
| Wikidata / Wikimedia Commons (license-filtered) | 44 | CC / PD | Open |
| Smithsonian NNC | 8 | CC0 | Open |
Counts as of the 2026-07-01 audit against a corpus of ~305,000 sightings; buckets shift slightly as harvesting and cleanup continue. Next planned sources: Yale University Art Gallery (~30,000 coins, CC0) and the Vienna Münzkabinett (KHM).
What this means for reuse
- Research, teaching, and non-commercial use: the whole corpus is usable with attribution to the original collections.
- Commercial reuse: restricted. The CC-BY-NC-SA buckets (Corpus Nummorum + most museum collections, 40%+ of the corpus) may not be used commercially. Syllogeum itself is run on a strictly non-commercial, donation/grant basis for exactly this reason.
- Redistribution: the bulk data releases above contain only collections whose image licence clearly allows redistribution (public domain, CC0, CC BY, CC BY-SA), each credited in
LICENSE-DATA.md; NC-licensed or unclear material is linked and attributed, not re-licensed.
The opt-out policy
Some institutions restrict automated access to their collections. Syllogeum treats that as an answer, not an obstacle:
- The British Museum and the Bibliothèque nationale de France block automated harvesting. Both are disabled as harvest targets in the codebase — commented out, documented, and verifiable in the public source. Zero records from either institution's servers exist in the corpus. (Records crediting "BnF" are partner-collection credit lines inside Corpus Nummorum's own openly-published Zenodo dataset; "British Museum" matches are pre-1923 public-domain catalogue plates.)
- No paywalled, login-gated, or challenge-protected source is scraped. Sources that turned out to have unclear image rights (e.g. Sixbid, Wildwinds) were removed or never ingested.
Any collection that wants its records removed or its harvesting stopped gets that honored — one email, no argument: [email protected].
Takedown & corrections
Rights holders, institutions, and researchers can request removal or correction of any record: [email protected]. Include the sighting URI(s) (/sighting/{id}) if possible. Takedowns are processed promptly and logged; corrections to attribution are treated as data-quality fixes and applied at the source of truth.
Open gap: Europeana per-item rights
Honesty requires naming the unresolved part. "Via Europeana" identifies the aggregator; each image carries its own edmRights statement ranging from public domain to in-copyright. Europeana's metadata is CC0, but per-item image rights for the ~45,000 Europeana-sourced sightings have not yet been resolved into the database. That backfill (from the Europeana API) is on the roadmap; until it lands, treat Europeana-sourced images conservatively and check the linked source record.
License
Underlying records remain under their source terms (attribution required where the source requires it — each record names its collection in the attribution field; ANS and Fralin metadata are ODbL, Berlin and Linz catalogue text is CC BY-SA 4.0). The embedding vectors and Syllogeum's own additions (URIs, rights labels, type linkage) are released under CC0. Full text, with the per-collection credits, ships in each release as LICENSE-DATA.md.
AI disclosure
Separately: a meaningful portion of Syllogeum's code was written with AI assistance (Claude). The data pipeline's decisions — what to ingest, what to exclude, what to disclose — are human ones.
How to cite
kauzmic (2026). Syllogeum: an open archive of ancient coin imagery. https://syllogeum.org — a DOI-minted deposit of these releases (Zenodo) is planned.
API
The same data is queryable live: interactive documentation of the public read API at /docs. Please rate-limit yourself; for bulk work prefer the dumps above.