nabu-places

The schema

Three YAML files; plain data, no code.

names.yml

<source-slug>:
  "<verbatim place-name string>": <decision row>

• Section keys are consuming-catalog source slugs (cdli, oracc).
• Name keys are the source’s place-name strings byte-exact as they occur in that source’s data — composites, ?-suffixes, diacritics and all. Never normalized: the string IS the join key.
• Decision rows, one of:

{refs: [cigs:GIR, pleiades:912855], status: matched}
{refs: [...], status: matched, certainty: low, note: "upstream ?-qualified; ..."}
{status: unlocatable, note: "site unidentified; RGTC 3 s.v."}
{status: region, note: "regional bin, not a place"}
{status: ghost}
{status: rejected, note: "the Umbrian homonym, not this corpus's referent"}
{alias_of: "Roma", certainty: low}

Rules (validator-enforced):

• matched requires ≥1 ref; every other status forbids refs.
• unlocatable and rejected require a note.
• alias_of must target an existing key in the same section; alias rows carry only alias_of/certainty/note. One hop, no chains.
• certainty ∈ high (default) · low.
• Duplicate section keys and duplicate name keys are refused (YAML last-wins silently loses rows otherwise).

places.yml — the native minting lane

The registry is not only glue: where scholarship (text analysis, enrichment, AI-assisted argument) establishes a place no gazetteer registers, mint a record here and cite it from names.yml as np:<ID>.

KARKAR_SU:
  name: Karkar (the Umma-province waterway station)
  lat: 31.9
  lon: 45.7
  evidence: >-
    Identified from the join of CDLI texts P123456/P234567 (the barge
    itineraries name it between Umma and KI.AN); argument in <ref>.
  refs: [cigs:JOK]   # related records — context, never identity

Rules (validator-enforced): ids match [A-Z][A-Z0-9_]{1,23} (the CIGS mnemonic style); name and evidence are required — a minted place is a reviewable scholarly claim; lat/lon come together or not at all (absent is honest, never 0,0); refs cite gazetteer records as context only — if a gazetteer actually holds the place, match it in names.yml instead of minting; every np: ref in names.yml must have a record here.

namespaces.yml

pleiades:
  name: Pleiades — ancient-world gazetteer
  id_shape: '\d+'
  uri: https://pleiades.stoa.org/places/{id}

Every ref’s namespace: must be declared and its id must match id_shape anchored whole. Current namespaces: pleiades, tm (Trismegistos Geo), cigs (site mnemonics, e.g. GIR — no per-place URL), geonames, and np — this registry’s own minted records (places.yml).

Doctrine, restated as constraints

1. Identity-default: an unlisted (source, string) pair is unmatched. Consumers must render that as unmatched, never guess.
2. No fuzzy matching: rows are placed by judgment (or generated from a gazetteer’s own authority — e.g. CIGS’s legacy_name IS the CDLI composite — and reviewed); the registry never encodes a similarity algorithm.
3. Namespaces are parallel: a row may cite several, asserting the string denotes each; cross-namespace equivalence lives in the gazetteers’ crosswalk data, not here.
4. Provenance on anything non-obvious: note: says why, so every row is reviewable years later.

This site is open source. Improve this page.