Record Atlas
3.50M volumes

3,502,257 volumes in the Atlas

Local
36 public files · no key, no rate limit, no account

Use the data

Everything this map draws is a plain JSON file on a public URL, and the map is only the first client. Take any of it, for anything, without asking.

New to this? Connect to the Atlas says in plain words which file answers which question, with a command that fetches each one.

The licences — there is more than one, and the difference matters

Read the licence field of the file you are taking. Every file derived from somebody else's work declares whose and on what terms, and the table below prints what each one actually says rather than one blanket claim over all of them.

  • ODbL 1.0, © OpenStreetMap contributors — the archives and the cemeteries harvested from OSM, and the 60,883 Irish townlands, which come from Townlands.ie. Attribution required, and share-alike: a derived database has to be published under ODbL too.
  • CC0 1.0 — the sets harvested from Wikidata, because Wikidata is CC0, and the surname corpus, which declares it.
  • No licence stated — collections.json, FamilySearch's public catalogue metadata: collection titles, years, and record and image counts. FamilySearch states no licence for it here and none is asserted. Titles and counts are facts about its catalogue, not this project's work, and they are published so a researcher can see what exists. If you reuse them at scale, read FamilySearch's own terms; we have not cleared that for you.
  • No licence stated — the holdings list of HR-HDA-883, the register collection of the Croatian State Archives, taken from the published inventory in Angelika Milić, «Zbirka matičnih knjiga u Hrvatskom državnom arhivu», Arhivski vjesnik 60 (2017), 251–311, hrcak.srce.hr/194831. The article is openly readable on HRČAK and grants no licence — not CC, not public domain, nothing. Openly readable is not openly licensed, so none is asserted here and none should be inferred. What is carried is the factual content of an inventory of a state holding — parish, confession, register type, years — and not the author's text, arrangement or commentary. If you reuse it, cite Milić 2017 and satisfy yourself independently; do not rely on us having cleared it, because we have not.
  • No licence stated — the register rows of HR-HDA-1448 (the Croatian State Archives' collection of microfilms of parish registers) for the two parishes a family archive has looked up, in hda1448-holdings.json, taken from the archive's finding aid at arhiv.hr. The finding aid states no licence and none is asserted. What is carried is the factual content (parish, kind of register, years, holder, roll) and not any text. It is two parishes, not the collection; a parish missing from it has not been looked up.
  • CC BY-SA 4.0, attribution AND share-alike — the English Wiktionary, for what a surname is said to be. Three things are taken and no prose: the language sections a name is tagged a surname under, the from= of Wiktionary's own surname template, and the word a name derives from where the template names one. The Etymology and Statistics sections are somebody's writing and are left alone — a template parameter is a fact in a field, and that is the line. Every row carries the page and the revision it was read at, because Wiktionary changes under you. Share-alike: a database derived from this part has to carry the same licence.
  • CC BY 4.0, attribution required — GeoNames, which is the gazetteer and therefore the reason any old place name resolves at all; and Statbel, for the Belgian surnames at commune grain.
  • CC BY 3.0 IT — Istat, for the Italian comuni.
  • Open Government Licence v3.0 — the National Records of Scotland, for the civil parishes.
  • Trove API Terms of Use — the 2,014 Australian newspaper titles, from the National Library of Australia. Two conditions, and the second is unusual. Attribution must name the contributing source AND Trove, with a link back, which is why every newspaper row carries the National Library as its holder and links to its own Trove page. And the metadata may only be CACHED FOR THIRTY DAYS — so this atlas re-harvests fortnightly and scripts/build.py fails the build once the file is older than thirty days. A static site that bakes its data in would otherwise drift out of the terms in silence.
  • Public domain — Natural Earth, for the country outlines.
  • SIL Open Font License 1.1 — Caveat (© 2014 The Caveat Project Authors), the handwriting in the story tour. Served from this site, with its licence at /fonts/Caveat-OFL.txt.
  • Terms not declared — the Estonian parish and county vocabularies from the Rahvusarhiiv. Said plainly rather than assumed either way.
  • FamilySearch catalogue metadata — not an open licence. This one matters most and this page had it wrong. Every volume title and waypoint id in a place file comes from FamilySearch's catalogue: «Defunciones 1980-1989» is their cataloguing, not our writing. The fact that a collection covers a place is not theirs to own and is what this atlas is for; the titles are theirs. Take the shape and the coverage freely, and do not treat the strings as relicensable.
  • Open data — Open Archieven's index of more than two hundred Dutch and Belgian archives: which places each archive holds records for, and how many. Their own rule is that the HOLDING ARCHIVE is named as the source, so every row here is titled with the archive that holds it and that name travels to the place page. These are index entries — a name, a date and a shelf reference — and never an image; whether a scan exists is that archive's business and this says nothing about it.
  • No licence stated — the National Archives of Ireland's census coverage: which counties the 1901 and 1911 censuses cover and how many records each holds, read from its public catalogue API. Those are facts about what an archive holds rather than the records themselves, and the archive publishes no terms for them. They ride inside the place files as coverage claims. The census itself is free to search and view on the archive's own site, and permission to publish an image from it is a separate thing you ask them for.
  • No licence stated — the Greenland National Museum and Archives' church-ledger pages: which parishes and year ranges are online, and which volumes are held offline, read from the archive's public pages (robots.txt respected, one request a second). Only the facts of coverage are taken; the Issuu scans of the ledgers are the archive's and nothing here fetches or reproduces them.
  • This atlas's own compilation — the index, the place files, the record regions, the provider spine, the flag timelines and the 1,454 hand-drawn flags that go with them. That is our research and it carries no licence file yet, because the repository has none to point at. Declaring one is the owner's to do and not a build script's; until it is done, take it and cite it.

What no licence here covers: the records themselves. This atlas publishes pointers — that a collection exists, what it covers, where to look. Each provider sets its own terms for the images and indexes at the other end of those links, and those terms are theirs.

How to fetch it
# the first paint only — the twelve biggest places per 5-degree cell
curl -s https://recordatlas.org/index.json | jq '.places | length'

# every place: the first paint, plus every viewport cell, plus the quiet half
curl -s https://recordatlas.org/index.json > i.json
jq -r '.tail.cells | keys[]' i.json   | xargs -I{} curl -s https://recordatlas.org/pi/{}.json   | jq '.places[]' | jq -s 'length'

# one place, in full
curl -s https://recordatlas.org/p/gracisce.json

# which archive held this ground, and when
curl -s https://recordatlas.org/regions.json | jq '.regions[0]'

No key, no rate limit, no account, readable from any origin — these are static files on GitHub Pages, which is also why there is no query endpoint. Fetch the index and filter it yourself; the first paint is 309 KB, and every place is that plus the 170 cells under /pi/.

The context contract

/context.json answers «what held this ground in this year?» for the sites that quote the atlas beside a dated event. It is versioned: the contract.version field follows semantic versioning, and anything a consumer could notice — a renamed field, a changed id, a different lookup rule — raises it. Version 1.13.0.

  • Ids are permanent. A region era is <region>--era-<from> (calabria--era-1815); a country era is <page slug>--<era id> (croatia-flags--era-1991), the id the flag layer already uses. Each carries a url that lands on the era.
  • An era covers from ≤ year < to. The year an era ends is the year the next begins, so a bare year equal to a boundary belongs to the newer era. A consumer holding only a year should say «in that year» or stay silent.
  • Look in regions first. Fall back to countries only when no region covers the place. Countries are keyed by flag page slug, not ISO code: Great Britain is four pages, and a modern country cannot say what held a place whose country has since moved.
  • Stay silent where quality says so. gaps are intervals a page leaves empty; overlaps are two eras at once, concurrent where the page really lists two polities. evidence is attested only where a primary or reference source, or two independent compiled ones, support the era; unconfirmed means the sources were read and do not confirm the era's main claim; unreviewed and unlabelled mean the sources could not be read or nobody has checked. Cite none of the three as established.
  • basis says where a region's dates came from. seeded:defranceschi-sovereignty means the bands were taken from the Defranceschi archive's own research, so the atlas cannot be cited as independent confirmation of them. researched:YYYY-MM-DD means the bands were read from pages fetched on that date, and the provenance file keeps the URLs for each era. not-recorded means the file does not say.
  • A decade is not a year. An era written «1816 – 1820s» carries to: 1820 and approx: true, the first year of the decade. A century («1500s») still carries no number at all.
  • null is not false. contested and approx on a region era are null where nobody has assessed them; false means someone looked and the sources give a clean date, true comes with a why. Forty-one regions are assessed so far (the sixteen written in October 2026 and twenty-five older ones); each assessed era carries a why and, from the older batch, an assessmentSource. Treat null as uncertain.
  • Margins say how uncertain. approx concerns only an era's start and end years. fromMargin and toMargin give the uncertainty on each in whole years (0 = the sources agree on one year, null = unknowable), and contestedScope says what a contested flag is about: bounds (when the change happened), regime (who held the ground in the middle) or part (only some of the ring). A note beside an event is honest when the event year is further than the margin from each boundary and the scope is not regime or part. marginSourced false means the margin rests on dates the assessor knew but did not find in a fetched page.
  • Pilot: rulers and languages. A region with hasRulers true carries rulers: every reign a source gives, with from and to as the source states them (a day or a year), numeric fromYear and toYear for lookup, and a Wikidata id where one could be confirmed, fromMargin and toMargin (0 where the source gives a day, 1 where only a year or a boundary known to differ in fact) and a stable polityId. Reigns overlap where the source's dates do, and a reign is the legal one, not who held the ground in fact. language rows record only what a fetched page states, with what it does not cover; languages: null means the sources read do not say. So far only Calabria.
  • Rings are in the file. Each region carries its ring as [lat, lon] vertices and a bbox as [south, west, north, east]. The rings are approximate by design, so a place within about fifteen kilometres of an edge may belong to the next region.

Not in this version, and not to be inferred: a language of administration, a ruler per year, sources on region eras, and events (wars, famines, epidemics). Ten country eras with no id on their page are left out because they cannot be cited.

curl -s https://recordatlas.org/context.json   | jq '.regions[] | select(.id=="calabria") | .eras[] | select(.from<=1860 and (.to==null or .to>1860))'
The files
PathSizeWhat it isLicence
/index.json
{ countries, access, rank, kinds, kindLabels, tail: { dir, deg, places, cells }, quiet: { file, places }, places: [{ i, n, y, x, c, v, a, q, s: [from, to] }] }
309 KB The first paint: the 5-degree cell grid's twelve biggest places in each cell, with only what a marker needs — id, name, latitude, longitude, folded search text, access class, collection and volume counts, and the year span its books cover. 2,606 rows, NOT every place: `tail` says where the other 44,623 live and `quiet` the unwalked ground. If you want them all, read this plus every pi/ cell plus index-quiet.json. this atlas — none declared
/pi/<lat>_<lon>.json
{ places: [ … as above … ] }
170 files The places index.json leaves out, in cells of 5 degrees keyed on the floor of their south-west corner. The map fetches the cells its viewport covers and then the rest in the background, so searching never depends on where the reader has panned. 44,623 rows across 170 cells, same shape as index.json's. this atlas — none declared
/index-quiet.json
{ places: [ … as above … ] }
1,023 KB The unwalked ground — places with nothing catalogued yet. Fetched after first paint so the map draws before it arrives. 10,125 rows, same shape. this atlas — none declared
/p/<place-id>.json
{ id, name, lat, lon, country, names[], collections[], volumes: [{ t, from, to, url, in }] }
57,354 files Everything about one place that a marker does not need: every collection covering it, every volume with its year span and deep link, its other names, which archive holds it, the nearest parishes and the nearest cemeteries. The `i` field of an index row is the filename. NOTE: the volume TITLES here are FamilySearch's catalogue metadata, not this atlas's writing — see the licence column and the note above. this atlas — none declared
/regions.json
{ regions: [{ id, from, to, polity, filedUnder[], note, geo }] }
2.0 MB The jurisdiction layer — 994 record regions, each an era with the polity that held that ground and the archives its paper went to. This is the file nobody else publishes. this atlas — none declared
/context.json
{ contract: { name, version }, rule, regions: [{ id, label, alt[], eras: [{ id, from, to, present, regime, regimeLabel, polity, filedUnder[], basis, contested, approx, url }] }], countries: [{ slug, cc, eras: [{ id, from, to, present, approx, dates, polity, evidence, concurrent, url }] }], quality: { gaps[], overlaps[] } }
3.0 MB The versioned contract for «what held this ground in this year?». Every region era and every country era has a permanent id, an evidence class and a link; `quality` lists the gaps and overlaps where a lookup should stay silent. Written for sites that snapshot it. Documented in full under «The context contract» below. This atlas's own compilation. Cite it as «per the Record Atlas» with a link to the era's url. See /data/ for the licence column.
/history.json
{ contract, chapters: [{ id, title, from, to, view, layers, blurb }], density: { decade: count }, moves: [{ s, v, d, f, t, ot, how, a, b, seat?, fy, open?, present?, e, p }] }
996 KB The movements behind the History page: every conquest, occupation and migration the flag pages record, with an origin and a destination point so it can be drawn as an arrow, the year taken from the page's own words where it gives one, the thirteen chapters, and the sourced voyages and expeditions (each with dated waypoints, a leader with a Wikidata id, a sponsor, what was reached and at least two sources). `how` says how each arrow was drawn: between two countries, from the seat of an empire, from a named people's country, or a stub with no second place. this atlas — none declared
/history-ground.json
{ shapes: [{ n, q, s, e, p: [ring[[lon, lat]]] }] }
2.2 MB Who held the ground, year by year, for the History page: about 3,100 dated polity shapes from 600 CE, coarse enough for a world map and wrong for a parish. A year with no shape for a place has no data there.
from OpenHistoricalMap sovereign boundaries, simplified to 0.3 degrees
CC0 1.0 Universal (public domain)
/leaders.json
{ countries: { <ISO 3166-1 alpha-2>: [[name, wikidataId, fromYear, toYear|null, startDay, endDay|null]] } }
— Heads of state by country from 1000 CE, each with a Wikidata id and the day the term starts and ends. COVERAGE IS UNEVEN: Wikidata names few holders before the 1800s and none for some countries in some years, and a country-year with no row means Wikidata names none. Caretakers under thirty days and terms with no end that are followed by another holder are dropped.
from Wikidata: position-held and head-of-state statements, harvested by scripts/harvest-leaders.py
CC0 1.0 Universal (public domain)
/flag-eras.json
{ eras: [{ slug, id, dates, from?, to?, approx?, present?, polity, tags[] }] }
683 KB Every era card on every flag page, extracted once: the author's own dates, the years that could be read from them, the polity, and the tags. The source of context.json's country eras; read context.json instead if you want ids, evidence and gaps.
from site/src/flags-pages/*.html — this project's own pages
This atlas's own compilation
/flag-map.json
{ at: { cc: [lat, lon] }, f: { cc: [[from, to, eraId]] }, slug: { pageSlug: cc } }
57 KB The flag that flew over each country, for the map's flag layer. KEYED ON THE MODERN COUNTRY (ISO 3166-1 alpha-2), so it is wrong for a place whose country has moved: Vodnjan in 1925 was Italian, not the Kingdom of Serbs, Croats and Slovenes. Great Britain merges four timelines under GB. For anything but drawing a flag on a country shape, use context.json.
from data/flag-eras.json and data/era-flags.json, both extracted from the flags atlas pages
this atlas — none declared
/providers.json
{ providers: [{ id, name, kind, access, countries[], url, checked }] }
2.5 MB 2348 providers with their coverage, access class, and the date each was last checked to be alive. this atlas — none declared
/c/<cc>.json
{ cc, name, box: [minLat, minLon, maxLat, maxLon], archives[], … }
237 files One file per country: its bounding box, its archives, and what this atlas holds there. Keyed by ISO 3166-1 alpha-2, lowercased. this atlas — none declared
/s/<3-char>.json
{ n, variants: [{ n, how }], countries: [{ cc, n, how }], tl?: { cc: [decade] }, ctx?: { lang, kind, after, wiki, q }, wk?: { langs, from, etymon, cite }, places?, pg? }
11,360 files The surname register, sharded on the first three folded characters of a name. Which countries attest a name, how many people each counted, and which spellings are linked to which — plus, where they exist, what Wikidata and Wiktionary say the name IS and the decades it was being born in. Drop `wk` and the rest is CC0.
from data/surnames.json and surname-context.json (Wikidata, CC0); the `wk` field from the English Wiktionary (CC BY-SA 4.0, share-alike)
CC0-1.0 — EXCEPT the `wk` field, which is CC BY-SA 4.0
/world-archives.json
{ archives: [{ name, lat, lon, cc, url, kind }] }
49 KB The archives layer — institutions at their street address, with what it takes to get an answer out of them.
from Wikidata — national archives (Q2122214), national libraries (Q22806) and research institutes (Q31855), by country
CC0-1.0
/archives-wikidata.json
{ archives: [{ q, name, lat, lon, cc, site }] }
1.3 MB The wider, unchecked set harvested from Wikidata. Drawn as its own layer because it is not all genealogical and none of it has been verified by hand.
from Wikidata via QLever, P31/P279* Q166118
CC0 1.0 (Wikidata)
/surnames.json.gz
{ name, licence, built, note, surnames: [{ n, countries[], variants[] }] }
21.6 MB The whole surname corpus in one gzipped file, including the 946,575 names that do not earn a page. There is a CSV beside it.
from national statistical registers — see the file's own note
CC0-1.0
/archives-osm.json
{ archives: [{ name, lat, lon, cc, tags }] }
96 KB Archives OpenStreetMap knows about. 572 of 726 are more than 500 m from anything else on this map, so OSM holds buildings the other sets do not. Unchecked. THIS FILE IS SHARE-ALIKE — see the licence column.
from OpenStreetMap, amenity=archive, via the Overpass API.
ODbL 1.0 — © OpenStreetMap contributors
/cemeteries-osm.json
{ cemeteries: [{ name, lat, lon, cc }] }
1.4 MB Named cemeteries, clipped to this atlas's record regions. Unnamed ones are left out: a shape on a map is not a place a researcher can ask about. Share-alike, as above.
from OpenStreetMap — landuse=cemetery and amenity=grave_yard, via Overpass, clipped to this map's record regions.
ODbL 1.0 — © OpenStreetMap contributors
/g/<3-char>.json
{ "<folded name>": [{ n, y, x, cc, alt[] }] }
18,712 files The gazetteer: the settlements the search box can find, sharded on the first three folded characters. This is what makes «Gallignana» resolve to Gračišće. COVERAGE IS NOT UNIFORM and the file says so itself: for the 50 countries in its `full` field it holds every populated place GeoNames knows, down to the hamlet of thirty people; for the rest it holds only places of five thousand and up, so a village there may genuinely not be in it. Shards are keyed on ASCII, so Cyrillic names are searchable in the data but not yet from this box. ATTRIBUTION REQUIRED — see the licence column.
from GeoNames cities5000 as a worldwide floor plus per-country dumps, CC BY 4.0 — https://www.geonames.org/
CC BY 4.0 (GeoNames)
/a/<3-char>.json
{ "<folded name>": [{ n, y, x, cc, level }] }
2,848 files Administrative divisions no collection names — findable in the search box, never drawn as a dot, because a district is not a parish.
from GeoNames admin1CodesASCII, admin2Codes and cities5000, CC BY 4.0 — https://www.geonames.org/
CC BY 4.0 (GeoNames)
/rs/<3-char>.json
{ "<folded name>": [{ n, y, x, r }] }
20,198 files Settlements inside a mapped record region. Findable, not drawn.
from GeoNames cities500, CC BY 4.0, clipped to this map's record regions.
CC BY 4.0 (GeoNames)
/bs/<3-char>.json
{ "<folded surname>": [[[communeIndex, count]], total, people] }
4,721 files Belgian surnames at COMMUNE grain — 510,811 rows, and nothing else in this atlas has that grain; every other register answers only «how many in the country». Statbel suppresses anything under five people, so the smallest number published anywhere here is 5. ATTRIBUTION REQUIRED.
from Family names by municipality, 1 January 2026 (Statbel, TA_POP_LST_2026)
CC BY 4.0 (Statbel)
/t/<3-char>.json
{ "<folded name>": [{ n, y, x, county, parish }] }
1,619 files 60,883 Irish townlands — the unit an Irish record actually names. Findable, not drawn: «Glebe» is the name of 213 of them. Share-alike.
from Townlands.ie — county, barony, civil parish and townland boundaries built from OpenStreetMap and Logainm
ODbL 1.0
/it/<3-char>.json
{ "<folded name>": [{ n, y, x, prov }] }
1,153 files 7,894 Italian comuni, 121 of them carrying an official second name.
from Istat, Codici statistici delle unità amministrative territoriali
CC BY 3.0 IT
/sc/<3-char>.json
{ "<folded name>": [{ n, y, x }] }
388 files Scottish civil parishes, from the National Records of Scotland.
from National Records of Scotland — Civil Parishes (WFS)
Open Government Licence v3.0
/ee/<3-char>.json
{ "<folded name>": [{ n, y, x }] }
88 files 113 Estonian parishes and 11 historic counties, names only.
from Rahvusarhiiv — the county and parish vocabularies of the manor and rural-municipality registers
Terms not declared (Rahvusarhiiv)
/cemeteries-wikidata.json
{ features: [{ q, n, y, x, r }] }
3.2 MB Named cemeteries from Wikidata. Drawn as a layer, and listed on a place page as the nearest few — unvisited, and not a claim that a burial register survives.
from Wikidata, instance of (or subclass of) Q39614, within the bounding boxes of this map's record regions. CC0.
CC0 1.0 (Wikidata)
/churches-wikidata.json
{ features: [{ q, n, y, x, r }] }
10.0 MB 108,054 churches. A church is a parish and a parish is where a register was kept, so these are the «nearest parishes» on a place page, with the distance shown.
from Wikidata, instance of (or subclass of) Q16970, within the bounding boxes of this map's record regions. CC0.
CC0 1.0 (Wikidata)
/libraries-wikidata.json
{ features: [{ q, n, y, x, r }] }
985 KB Libraries, as a layer. Unchecked, and not all of them hold anything genealogical.
from Wikidata, instance of (or subclass of) Q7075, within the bounding boxes of this map's record regions. CC0.
CC0 1.0 (Wikidata)
/country-points.json
{ points: [{ cc, n, y, x, p, v, c, s, w, fs }] }
23 KB One point per country, so that clicking anywhere on earth answers with something rather than nothing. this atlas — none declared
/place-names.json
{ licence, note, statuses, kinds, places: [{ id, name, country, names: [{ name, lang, kind, from?, to?, status, sources: [{ title, url }] }] }], counts }
130 KB What a place was called, and when: Salisbury 1890–1982 is Harare. Each name is a dated claim with the page that dates it, or a name given without dates (a lead: search it, never date a record by it). A register is indexed under the clerk's name in his year, so a search for today's name misses it. CC0
/directory.json
{ source, licence, licenceNote, note, checked, counts, rows: [{ id, repo, name, url, countries[], tags[], access, from, to, what, checked, gated }] }
456 KB 969 collections people and archives contributed, one row per named collection at a named repository: what it holds, the years, what it costs to look, and a link. The `checked` date means only that the LINK answered on that day: nobody has opened the volumes behind it.
from Contributed by David Defranceski, September 2026
CC0 1.0 Universal (public domain)
/collections.json
{ source, harvested, note, licence, licenceNote, collections: [{ cc, title, from, to, records, images, kind, url, countries[], filedUnder[], crosses[], places[] }] }
1.3 MB 3,504 FamilySearch record collections: title, years, how many records and images. A collection is a COVERAGE CLAIM over an area, not a volume standing in a town. Country-wide ones (the IGI, a whole country's church books) live only here; the ones that reach a place are also in that place's /p/ file. FamilySearch's catalogue metadata, and NO LICENCE IS STATED for it: see the licence list below.
from FamilySearch historical record collections, public catalogue endpoint
No licence stated
/sitemap-index.xml
sitemap index → six sitemaps
— Every page on this site, split by type — places, archives, countries, regions, surnames. this atlas — none declared
/feed.xml
Atom 1.0
— An Atom feed of what changed, one entry per day, generated from the git history. this atlas — none declared
What the numbers are today

57,354 places, 3,502,257 record volumes, 2348 providers, 994 jurisdiction regions. Rebuilt whenever anything lands, so these files move under you — there is no version number, because adding a field and calling it v1 forever would be a promise this project cannot keep. Fields are added; existing fields are not quietly repurposed.

Subscribe to the feed to be told when they change, or read what changed. If you build something with this, say so — knowing what the data is used for is the only way to know which parts are worth deepening.