Prior art — and the gap this registry fills
The ancient-world gazetteer landscape is rich; nabu-places adopts it
rather than competing with it. What follows is why a decisions layer
is still missing, and what each neighbor actually offers.
The gazetteers (adopted as data)
- Pleiades — the ancient-world
gazetteer: ~42k places, Greco-Roman core with a growing Near-Eastern
layer, daily exports, CC BY. The backbone id namespace: nearly
everything else crosswalks to it.
- Trismegistos Geo — ~65k
places grown from papyrology; the finest-grained authority for
Greco-Roman Egypt (nomes, villages, topoi), with multilingual name
forms (Greek, Demotic, Coptic). CC BY-SA.
- CIGS — the Cuneiform
Inscriptions Geographical Site Index (Uppsala): ~600 sites
purpose-built for the cuneiform corpus, each carrying Pleiades,
GeoNames, OpenStreetMap and CDLI-provenience ids — a crosswalk
hub in miniature. CC BY.
- GeoNames — the modern-features
complement (parishes, churches, tells’ modern names). CC BY.
The infrastructure neighbors
- Linked Places Format
— a JSON-LD interchange shape for gazetteer records (names,
timespans, links). An export dress for a place layer, not a home
for matching decisions.
- World Historical Gazetteer — a union
index of contributed gazetteers with reconciliation tooling. Its
reconciliation output stays inside its platform; the decisions are
not a portable, versioned artifact.
- Recogito / the Pelagios tradition — annotation tools that let a
scholar link a text’s place-mention to a gazetteer id. The linking
act is per-document annotation; corpus-scale per-string decisions
(this registry’s grain) are a different economy.
- Wikidata — CC0 crosswalk glue (Pleiades ⇄ Trismegistos ⇄
GeoNames properties), mineable for equivalences; not an
ancient-place authority in itself.
The gap
Every corpus that ships place names without machine refs forces each
consumer to redo the same matching — and the hard cases (homonyms like
Mediolanum ×6, ?-qualified proveniences, genuinely unlocatable sites,
house composite formats) are exactly where silent heuristics go wrong.
Reconciliation tools produce these judgments but don’t publish them as
small, reviewable, reusable data.
nabu-places is that missing artifact: the decisions themselves,
keyed by the source’s verbatim strings, versioned, validated, and
licensed for reuse — the
nabu-lects registry pattern
(identity-default, closed vocabulary, honest non-answers) applied to
place identity.
And one step beyond glue: when scholarship over the texts themselves —
joins across itineraries, enrichment, AI-assisted argument — establishes
a place no gazetteer registers at all, the registry mints its own
evidence-bearing record (places.yml, namespace np:) rather than
forcing the claim into someone else’s id space or losing it. Gazetteers
remain the authorities for what they cover; the native lane exists for
what they don’t yet.