# Syllogeum open-data release v1.1 (2026-09-23)

The redistributable subset of the Syllogeum ancient-coin archive:
118883 sightings from 6 collections, 21512 type cards, 237699 images,
plus DINOv2/SigLIP2 embedding vectors for every imaged record.

## Inclusion rule

A record is included only if its collection's image licence clearly allows
redistribution (public domain, CC0, CC BY or CC BY-SA). The collections, their
licences and the credit each one requires are listed in `LICENSE-DATA.md`.
15308 records from 34 other collections that the site files as
open linked data are left out, because their terms are non-commercial,
no-derivatives, or unclear.

This release replaces `syllogeum-open-v1` (2026-07-09), which was withdrawn
because it included images from collections whose licences do not clearly
allow redistribution.

## Files

| File | Contents |
| ---- | -------- |
| `sightings.jsonl.gz` / `sightings.csv.gz` | one record per sighting: labels, attribution, rights, physical data, stable URI, joined type fields |
| `types.jsonl.gz` | referenced type cards (OCRE / CRRO / PELLA — ANS open corpora) |
| `embeddings_dino_obv.npz` … `embeddings_siglip_rev.npz` | `ids` (int64) + `vectors` (float32, n x 768), L2-normalized, aligned by position |
| `images.tar` | obverse/reverse JPEGs referenced by `obv_image` / `rev_image` |
| `manifest.json` | counts + SHA-256 of every artifact |

## Loading embeddings

```python
import numpy as np
z = np.load("embeddings_dino_obv.npz")
ids, vecs = z["ids"], z["vectors"]   # cosine similarity = vecs @ vecs.T
```

## Citing

kauzmic (2026). Syllogeum: an open archive of ancient
coin imagery. https://syllogeum.org — cite individual records by
their stable URI plus the collection in their `attribution` field.

## Provenance, rights, AI disclosure

https://syllogeum.org/data#provenance — embeddings are machine-produced
image fingerprints (DINOv2 / SigLIP2); similarity between them is a machine
suggestion, never a verified attribution.