Repository navigation
Releases
Data 2026.11.17: app-sized layer (lite)
An app-sized layer (lite, CC BY 4.0), built by the forge profile data/embed/app.yaml (D-362).
world-np-app-lite.tgd: 172,339 places. 12,550,872 B on disk, 7,039,780 B on the wire (zstd-19, seekable).
- Nepal in full, with outlines at 20 m:
admin_atanswers down to the local level (wards are points only, Q-035). - The world's countries, first-level units and cities of 5,000 or more, by point and without outlines; India's districts and towns.
- Names in Nepali, English, Hindi and Sanskrit (Devanagari), with search keys for each.
- Components: point, elevation, admin, names, search, timezone, tz-rules, boundaries, admin-index. No DEM.
Built from code commit 52892e6 (CI green). It passes cargo xtask check-pack --lite. Layer BLAKE3: bdf647710baddcc94c34dce1c507abaab9696b1e928c1c2b8b1baf63bd31175e.
This file is outside the signed catalog: open it from bytes or with --pack. To verify it, run shasum -a 256 -c SHA256SUMS and then tgd verify --deep world-np-app-lite.tgd.
ids-lite.tar.gz holds the ID registries of this build. Credits are in attribution-world-np-app-lite.md; the full report is in summary-world-np-app-lite.md.
v1.1.0
Reviewed names as their own tier of the names chain, Government of Nepal data admitted and
credited, and the fixes found publishing 1.0.1. The first release published from CI by
trusted publishing on every registry.
Fixed
- Dart: the build hook's helpers moved to
lib/src/hook/, as pub.dev accepts only
hook/build.dartandhook/link.dart; the pub.dev 1.0.1 is published from that commit (D-361). - Release: npm publishes the tarballs by path (
./dist/…), platform packages first.
Added
- Name origin
reviewed(11): a name a person corrected against a cited source. - The forge's reviewed-names tier:
data/countries/<cc>/reviewed-names.tsv, written by
forge reviewed-import, applied by stable source key with a guard, gate G-reviewed on
stale rows; only attribution-only sources reach the lite build (D-359). - Licence class
government: a government's own work with no open licence stated, admitted
per source by a recorded decision and credited; Nepal's Election Commission lookups and
Constitution are the first (D-360). Reviewed names credit their sources in the pack's
notices.
v1.0.1
The first release on every registry (crates.io, npm, PyPI, pub.dev, Maven Central); Swift and
Go install from its tag as before.
Changed
- The npm packages are
teigeodata,teigeodata-wasmandteigeodata-wasm-lookup(the
names on crates.io, PyPI and pub.dev), owned by the Teispace organisation; the per-platform
addons the main package installs stay scoped,@teispace/teigeodata-<platform>. Neither was
on npm before this release.
Fixed
tgd extract --componentsof a pack with several parts of one kind or with block hashes
(any section of 1 MiB or more) failed with "a section kind was given twice": it now keeps
every part and writes its own block hashes.tgd extract --bboxkeeps only the terrain tiles that touch the box, copied as stored,
instead of the whole grid: the Kathmandu Valley out of Nepal's lite pack is 1.5 MB, not
70 MB, with the same heights inside the box.
v1.0.0
The first stable release. Every public surface now has its promise and its gate
(what 1.0 promises); the pack format is 1.0 and older packs
stay readable; every public enum that may grow is #[non_exhaustive]; the 1.0 API review
renamed the last parameters that broke the naming rule (see Breaking).
Added
- Each place's population (D-351): packs carry it in COMPONENTS (component 1, a count and
whether a source stated it or a population grid modelled it);Engine::population(feature
extra, on by default) returnsPopulation { people, source }; every place result has the
nullable columnspopulationandpopulation_kindin every language, andtgd, the HTTP
server and MCP printpopulationandpopulation_kind.extract::mergekeeps every
component of COMPONENTS. Both WebAssembly packages buildextra(feature of
teigeodata-wasm); every binding's parity suite checks both columns. teigeodata-enginefeaturefuzz:fuzz::engine, every query on a pack with edited
sections whose hashes hold, run nightly by libFuzzer (fuzz/fuzz_targets/engine.rs) and on
seeded inputs bycargo test(D-331);teigeodata_format::fuzz::edit_sectionbuilds such
packs, and the edited sections' full check runs too.teigeodata_sync::fuzz::trust(featurefuzz, now withrepo): the TUF update chain and
the catalog on a repository with one role attacked, nightly as thetrusttarget (D-331).teigeodata_ffi::fuzz::abi(featurefuzz): every engine function of the C ABI with fuzzed
arguments, nightly as theffitarget;teigeodata_server::fuzz::mcp(featuresfuzzand
mcp): every MCP tool with arguments from its own input schema, and resource reads, nightly
as themcptarget (D-331).- The weekly fuzzing campaign (
fuzz-campaign.yml): every target for 3 hours with coverage
reports, and its gatecargo xtask fuzz-coverage, which fails unless every section kind's
decoder is reached (D-331). teigeodata_server::http::Configgainsheader_timeout,max_connectionsand
max_in_flight, andtgd servegains--max-connectionsand--max-in-flight(D-334).cargo xtask unsafe-audit: the inventory of everyunsafein shipped code with its argument
and the tests that reach it (docs/06-engineering/unsafe-inventory.md), and the
threat model (D-333).teigeodata_format::check::sections_where: the full check of the sections a caller selects.- CodeQL code scanning (
codeql.yml) and the OpenSSF Scorecard (scorecard.yml), weekly. - Extension columns (D-335): a column-table column with an ID from
0x8000
(teigeodata_core::table::EXTENSION) is skipped by a reader that does not know it, so a
format minor can add optional columns;Table::extensionslists them. COMPONENTS IDs are 1
to 127.teigeodata_format::fuzz::{bodies, write_again}rebuild a pack from its sections. - A result
Readerof one shape in every language (D-322): new in Node.js, WebAssembly,
Swift, Dart, Java (Kotlin), C (teigeodata_reader.h) and C++ (teigeodata_reader.hpp),
both header-only and in the C package (rows,columns,has,numbers,strings,nulls,column,
records, iteration); Go's gainsHas,ColumnandRecords, Python'shas,column,
recordsand iteration, .NET'sBlobReaderHas,Numbers,Column,Recordsand
iteration. cargo xtask docs-examples: runs every language's example program on the pinned public pack
and checks its output (D-322).- The API, HTTP and MCP reference of the site, generated: the API's from each binding's own
declaration macros bycargo xtask bindings, the servers' fromteigeodata_server::http::ENDPOINTS
(a route table a test holds to the parser) andmcp::tool_definitionsand
mcp::resource_templates(D-323). Engine::boundary(id, lod)(an outline as a GeoJSON Feature, with its level, tolerance and
vertex count),Engine::distance(from, to)between two places, andbearing_degin the
language API and every binding, with parity values checked in each (D-324).
teigeodata_engine::Outline::write_geojsonandwrite_point_geojson: the GeoJSON writer the
servers used, now one writer for the servers and the bindings.cargo xtask data-pages: the site's coverage, country and measured pages, generated from the
pinned pack,data/countries/and the gates' records, with--check(D-325, D-326).- The site's playground (
/playground): the WebAssembly engine and the pinned pack in the
browser, MapLibre drawing the pack's own outlines, every answer with its code and timing;
cargo xtask playground [--fetch]writes its files, andsite/scripts/check-playground.mjs
runs it in Chromium in the docs workflow (D-328). --recordonpackage-wasm,sizesandload, andcargo xtask phone <layer.tgd> [--record](the phone-class bench with the example's own features): each writes its table,
stamped with the day, commit, machine and load average, tosite/data/measured/(D-326).cargo xtask profile <pack>… [--suite] [--secs]: each query suite under Instruments' Time
Profiler, one suite per process, with the largest frames by self time and under our own
frames;cargo xtask scaling <pack> [--record]: one shared engine at 1, 2, 4 and 8 threads,
failing under 1.5× at four; the engine'sprofileandscalingexamples (D-338).cargo xtask accuracy <cc> <pack>: a shipped pack scored against the blind labels of
tests/golden/<cc>-places.csvand its second pass, written as the generated
data/countries/<cc>/ACCURACY.md(D-339).- Suite C, the competitive report:
tools/competitiveandcompetitive.ymlask teigeodata,
tgd serve, Nominatim, Photon, GeoNames' own points andtzf-rsthe same questions (D-340).
Changed
-
teigeodata_format::Sorted::findinterpolates inu64and usesu128only when the
product overflows:get−6% on Nepal lite, −4% on the world baseline.teigeodata_core::bitpack::Packed::parse
checks a column's block directory as independent steps that vectorise: opening a pack −14%
and −17% (D-338). -
distance_mcalls the engine'smeters_betweeninstead of a second haversine of its own
(D-324). -
cargo xtask loadnames each pack in bold and right-aligns its figures, so its tables read
as Markdown (D-326). -
cargo xtask load: a load test oftgd serveand its/mcpon real packs withoha,
gated bytools/load/budget.toml(requests per second, p99, footprint and its growth)
(D-320);cargo xtask sizes --tgdreports the binary with and without its servers. -
The Dart package runs on the web, compiled by dart2js or dart2wasm, over the WebAssembly
module, with the classes it has natively (D-307).TeiGeodata.init()loads the module once
on the web and returns at once natively; the web opens packs from their bytes. -
Engine::open_file_range(path, offset, length)(and_with) in every language but
WebAssembly: opens a pack stored whole inside another file, such as an uncompressed asset of
an Android APK (noCompress), mapped in place (D-306). -
Releases are built, attested and published by one workflow,
release.yml(D-302): every
file with build provenance through Sigstore, a CycloneDX SBOM per crate andSHA256SUMS;
gh attestation verify <file> --repo teispace/teigeodatachecks one. -
The crates
teigeodata-core,-blob,-format,-names,-sync,-engine,-api,
-serverand-cliare publishable to crates.io (D-303, D-310). -
teigeodata-server, the request core both servers andtgdshare (D-310, D-311): the
native, GeoJSON and Nominatim-compatible outputs moved there from the CLI;query, every
place, elevation and postal query with its bounds (Limits::SERVER: 256 bytes of text, 50
results, 1,000 km) and stable error codes (Code, each with its HTTP status);packs, the
pack set over files and the store, following the store's journal while it answers; and
seams, who is asking, may they and how much they used, open by default (ADR-0025). -
tgd serve(featureserve, on by default) andteigeodata_server::http(featurehttp):
every query as aGETroute under/v1/,/packsand/health, on axum; bound to
127.0.0.1, CORS off unless configured, a 4 KiB query string, a request timeout, a panic
answeredinternal; follows the store's journal every 5 s (D-312). -
tgd serveanswers Nominatim's/search,/reverseand/lookup(json,jsonv2,
geojson) with the parameters geopy and Leaflet Control Geocoder send, so an app switches
by changing its URL; fields we cannot fill are left out (D-314). -
tgd mcp(featuremcp, on by default) andteigeodata_server::mcp: the MCP server on
standard input and output, for revisions 2026-07-28 and 2025-11-25; fourteen read-only tools
(search,reverse,nearby,get_place,lookup,admin_at,hierarchy,children,
timezone,elevation,postal,boundary,distance,packs) with input and output schemas, structured
answers with their attribution, signed cursors for pages, and packs and places as resources
(D-315). -
/v1/hierarchy?id=(a place's ancestors) and/v1/children?id=&limit=(D-315). -
/v1/boundary?id=&lod=: a place's outline as GeoJSON, at the finest level of detail within
20,000 vertices unless one is asked;/v1/distance?from=&to=: metres and bearing between
places or points. In MCP, theboundarytool links the outline as the resource
teigeodata://boundary/{id}?lod=(D-317). -
/v1/coverage?lat=&lon=&radius_m=and the MCP toolcoverage, on a server over the store:
the countries at a point with the catalog's packs for it and their state, astgd locate
answers, read from the stored catalog without opening the store (D-318).
teigeodata_sync::client::catalog_atreads it so;PackState::of. -
MCP
installandinstall_status, on a se...
Data 2026.11.15 (main and lite)
Data files for Nepal, main and lite builds, built by the data-build workflow from 6b234d4 on main (run 43733977409), where every gate passed: G-det (the build stage run twice gives identical bytes), the licence and notice gate, the corrections gate, the world and shipped-baseline gates, and the golden search gate. This is a pre-release outside the signed catalog: teigeodata-sync does not fetch it, so download the files and open them with Layer::open.
What is new
- Pack format 1.0, the format frozen for v1.0.0 (D-336). The files need a reader at format 1 (see Code below); readers at
v0.3.xrefuse them. - Nepal's main build for the first time, with OpenStreetMap, Copernicus GLO-30 elevation, WorldPop population and timezone-boundary-builder on top of the lite sources. ODbL.
- Accuracy pages for both builds, measured on these exact files against a labelled sample of 222 places (
ACCURACY.md,ACCURACY-lite.md). Horizontal error P50 362 m and P90 958 m for main, 260 m and 939 m for lite. Read the pages' notes on how the labels were made: they favour a build that keeps GeoNames' point. - Each place's population (D-351): every current place with a known count carries it, marked
stated(a source's count) ormodelled(the WorldPop grid summed over an admin unit; most of Nepal's local levels). Readers return it aspopulationandpopulation_kind. Main: 1,732 places (956 stated, 776 modelled); lite: 792, all stated. - Search on the main build passes the golden gate at 451 of 521 queries first (D-347).
Files
| File | What | Disk |
|---|---|---|
np-places-main.tgd |
Nepal, main build: 102,278 places with admin hierarchy, boundaries, names in Nepali and English (official CBS names on every local level), search, timezone, postal codes, population and a 1-arcsecond DEM | 86,589,892 B |
np-places-lite.tgd |
Nepal, lite build: 95,716 places from clean-room sources only, with NASADEM elevation; no postal codes (their licence is not yet stated, Q-009) and no modelled population | 87,532,288 B |
np-baseline-main.tgd, np-baseline-lite.tgd |
Nepal's part of the shipped baseline | 69,116 B, 67,644 B |
world-baseline-main.tgd, world-baseline-lite.tgd |
The world part of the shipped baseline: countries, first-level units, cities of 15,000 or more | 4,807,940 B, 5,025,356 B |
baseline-main.tgd, baseline-lite.tgd |
The shipped baseline, one file: the world part and Nepal's | 4,939,816 B, 5,155,656 B |
np-summary-<build>.md |
What each build did: sources, merges, names, corrections, elevation | |
np-attribution-<build>.md |
The notices each build carries |
ids-main.tar.gz and ids-lite.tar.gz hold the ID registries these files were numbered with; cargo xtask ids restore puts them back, so the next build keeps every published ID. np-places-lite.tgd has the same places as the one in data-2026.11.14: tgd diff finds none added, removed, moved, renamed or retyped; the only new section is the populations (COMPONENTS, 4,020 B). The main build's IDs are new.
The world layer and points layer of data-2026.11.14.3 are not rebuilt here; they stay format 0, which format-1 readers still read.
Code
Read these files with the crates at 6b234d4 or later (earlier format-1 readers open them but do not return populations) (v1.0.0 once tagged):
teigeodata-engine = { git = "https://github.com/teispace/teigeodata", rev = "6b234d434f2201bbedd6a5527d6865c5fc58e351" }Verify
shasum -a 256 -c SHA256SUMS
tgd verify --deep np-places-main.tgd
Licence
- Main build (
*-main.*): Open Database License 1.0, as a database containing OpenStreetMap data. © OpenStreetMap contributors. The other sources and their notices are innp-attribution-main.md: GeoNames (CC BY 4.0), Wikidata (CC0), the Survey Department of Nepal and the UN RCO via OCHA HDX (CC BY-IGO 3.0), timezone-boundary-builder 2026f (ODbL) and tzdb 2026f, the General Post Office of Nepal, Copernicus DEM GLO-30 (© DLR e.V. 2010-2014 and © Airbus Defence and Space GmbH 2014-2018 provided under COPERNICUS by the European Union and ESA), WorldPop Global2 R2025A (CC BY 4.0), Microsoft Global ML Building Footprints (CDLA-Permissive-2.0), Google Open Buildings v3 (CC BY 4.0) and the Central Bureau of Statistics via Open Data Nepal. - Lite build (
*-lite.*): CC BY 4.0, from clean-room sources only, listed innp-attribution-lite.md: GeoNames, Wikidata, the Survey Department of Nepal and the UN RCO via OCHA HDX, tzdb 2026f, NASADEM (NASA JPL and LP DAAC), the two building datasets and the Central Bureau of Statistics.
Data 2026.11.14.3: ISO 3166 codes world-wide
Data files built from eb8ab5c on main. This is a pre-release outside the signed catalog: teigeodata-sync does not fetch it, so download the files and open them with Layer::open.
What is new
- ISO 3166 codes on every country and on most first-level units, world-wide (D-271). The codes cover 237 countries and 4,071 of 4,567 units. Each unit's code comes from Wikidata, or else from Natural Earth, and ships only when CLDR 49 lists it as current. India's states carry their current codes (
IN-OD,IN-UK,IN-CG,IN-TS), and the codes they replaced answer nothing.Engine::iso_codeandEngine::by_isonow answer outside Nepal. - Only the
ISO_CODESsection changed in the three files below. Every ID is kept: no place is added, removed, moved or renamed against data-2026.11.14.2. np-places-lite.tgdis the one in data-2026.11.14, byte for byte.
Files
| File | What | Disk | Change |
|---|---|---|---|
world-np-lite.tgd |
Offline layer: the world's countries, first-level units and cities of 5,000 or more, India's districts and statutory towns, and Nepal in full; Nepali, Hindi and English names | 15641408 B | +19,968 B |
baseline-lite.tgd |
The shipped baseline: countries, first-level units, cities of 15,000 or more | 5031560 B | +19,648 B |
points-en-lite.tgd |
Points and English names of the baseline's places | 1869516 B | +19,664 B |
ids-lite.tar.gz holds the ID registries the files were numbered with (no serial changed); cargo xtask ids restore puts them back. Each summary-*.md describes its file.
Code
The format did not change. Read these files with the crates at v0.2.0 or later:
teigeodata-engine = { git = "https://github.com/teispace/teigeodata", tag = "v0.3.0" }Verify
shasum -a 256 -c SHA256SUMS
Licence
Lite build, CC BY 4.0. The sources and their notices are listed in data-2026.11.14.2. This release adds one source: Unicode CLDR 49's subdivision validity list (Copyright © 1991-2026 Unicode, Inc., under the Unicode License v3), already credited there.
Data 2026.11.14.2 (lite)
Data files built from 482aba0 on main. This is a pre-release outside the signed catalog: teigeodata-sync does not fetch it, so download the files and open them with Layer::open.
What is new
- Former and everyday names find their place (D-268). Every country, region and city of
world-np-lite.tgdtakes its English aliases from Wikidata: 12,436 names on 6,398 places. A search for Poona, Simla, Tanjore, Jubbulpore, Cochin, Banaras, Calcutta, Trivandrum, Peking or Ceylon gives the place first. An alias is kept only when it adds a search key: not one of the place's names, no bracketed qualifier, not one of its names with words around it, not another place's name in its country. - Every ID is kept: no place is added, removed, moved or renamed against
data-2026.11.14. - Only
world-np-lite.tgdchanged.baseline-lite.tgd,points-en-lite.tgdandnp-places-lite.tgdare the ones in data-2026.11.14, byte for byte. - The engine on
main(afterv0.3.0) weighs a substring match of 3 or 4 characters less (D-269), so "Var" gives Varanasi before a country that holds "var" inside an alias. Withv0.3.0, that one query may rank a country higher.
Files
| File | What | Rows | Disk | Wire |
|---|---|---|---|---|
world-np-lite.tgd |
Offline layer: the world's countries, first-level units and cities of 5,000 or more with their alternate names and aliases, India's districts and statutory towns, and Nepal in full. Names in Nepali, Hindi and English, with heights, zones, zone rules, admin units, outlines and search | 172,339 | 15,621,440 B | 9,738,489 B |
ids-lite.tar.gz |
The ID registries the files were numbered with (unchanged); cargo xtask ids restore puts them back |
1,159,910 B |
summary-world-np-lite.md describes the layer.
Code
Read these files with the crates at v0.2.0 or later; the format did not change:
teigeodata-engine = { git = "https://github.com/teispace/teigeodata", tag = "v0.3.0" }Verify
shasum -a 256 -c SHA256SUMS
Licence
Lite build, CC BY 4.0. The layer's sources and their notices:
- Made with Natural Earth (https://www.naturalearthdata.com), public domain
- GeoNames (https://www.geonames.org), CC BY 4.0
- Time zones: IANA Time Zone Database 2026f (public domain)
- Wikidata (https://www.wikidata.org), CC0 1.0
- Unicode CLDR (https://cldr.unicode.org), Copyright © 1991-2026 Unicode, Inc., under the Unicode License v3
- Local Government Directory, Ministry of Panchayati Raj, Government of India (https://lgdirectory.gov.in)
- Administrative boundaries: Survey Department of Nepal and UN Resident Coordinator's Office in Nepal, via OCHA HDX (cod-ab-npl), CC BY-IGO 3.0
- Elevation: NASADEM, NASA JPL and LP DAAC
- Buildings: Microsoft Global ML Building Footprints, CDLA-Permissive-2.0 (https://cdla.dev/permissive-2-0/)
- Buildings: Google Open Buildings v3 (Sirko et al. 2021, arXiv:2107.12283), CC BY 4.0
- Central Bureau of Statistics (now National Statistics Office), Nepal, via Open Data Nepal (https://opendatanepal.com), Creative Commons Attribution
Data 2026.11.14 (lite)
Data files built from 8ccf25c on main. This is a pre-release outside the signed catalog: teigeodata-sync does not fetch it, so download the files and open them with Layer::open.
What is new
- One importance scale for countries, regions and cities (D-266). A region or country now weighs its population the way a city does, and every first- and second-level unit takes its population from Wikidata. A typed prefix finds big cities among the first results ("Var" gives Varanasi, "Cal" gives California), a city of a million people or more ranks as a city, and a place named like its country ranks just below the country ("Luxembourg" gives the country first). A city just off its region's simplified outline is placed in the nearest region of its country within 10 km.
- Search measured on three golden sets: world 60 of 60 (32 of 44 before), India 44 of 44, Nepal recall@1 0.839 → 0.848.
IDs
world-np-lite.tgdandnp-places-lite.tgdkeep every published ID: no place is added, removed, moved or renamed againstdata-2026.11.13.3.baseline-lite.tgdandpoints-en-lite.tgdare renumbered once (D-267). They had not changed sincedata-2026.11.12anddata-2026.11.13, and were numbered from an ID registry that was never released. They disagreed with the released registry and withworld-np-lite.tgdon 21,613 IDs:tgd:in/207was Delhi in the layer and Titlāgarh in the baseline. They now use the released registry's IDs, the onesworld-np-lite.tgdhas always used. If you stored IDs from these two files, look them up again by name or point.- From now on each world part holds its IDs under its own name in the registry, and a build that would retire more than 1% of its IDs stops (D-267).
Files
| File | What | Rows | Disk | Wire |
|---|---|---|---|---|
world-np-lite.tgd |
Offline layer: the world's countries, first-level units and cities of 5,000 or more with their alternate names, India's districts and statutory towns, and Nepal in full. Names in Nepali, Hindi and English, with heights, zones, zone rules, admin units, outlines and search | 172,339 | 15,199,040 B | 9,471,607 B |
baseline-lite.tgd |
The baseline layer the engine can embed: the world with each country's slice, every name | 39,395 | 5,011,912 B | 3,682,170 B |
points-en-lite.tgd |
Points, heights and English names with search; no outlines | 39,395 | 1,849,852 B | 1,212,307 B |
np-places-lite.tgd |
Nepal's full lite pack, with names by language and outline parts by level and level of detail | 95,716 | 87,528,192 B | see np-summary-lite.md |
ids-lite.tar.gz |
The ID registries (source keys to stable IDs) the files were numbered with; cargo xtask ids restore puts them back |
1,159,850 B |
Each summary-*.md describes its layer, np-summary-lite.md describes the Nepal pack and np-attribution-lite.md holds its attribution.
Code
Read these files with the crates at v0.2.0 or later; the format did not change:
teigeodata-engine = { git = "https://github.com/teispace/teigeodata", tag = "v0.3.0" }Verify
shasum -a 256 -c SHA256SUMS
Licence
Lite build, CC BY 4.0. The layer's sources and their notices:
- Made with Natural Earth (https://www.naturalearthdata.com), public domain
- GeoNames (https://www.geonames.org), CC BY 4.0
- Time zones: IANA Time Zone Database 2026f (public domain)
- Unicode CLDR (https://cldr.unicode.org), Copyright © 1991-2026 Unicode, Inc., under the Unicode License v3
- Wikidata (https://www.wikidata.org), CC0 1.0
- Local Government Directory, Ministry of Panchayati Raj, Government of India (https://lgdirectory.gov.in)
- Administrative boundaries: Survey Department of Nepal and UN Resident Coordinator's Office in Nepal, via OCHA HDX (cod-ab-npl), CC BY-IGO 3.0
- Elevation: NASADEM, NASA JPL and LP DAAC
- Buildings: Microsoft Global ML Building Footprints, CDLA-Permissive-2.0 (https://cdla.dev/permissive-2-0/)
- Buildings: Google Open Buildings v3 (Sirko et al. 2021, arXiv:2107.12283), CC BY 4.0
- Central Bureau of Statistics (now National Statistics Office), Nepal, via Open Data Nepal (https://opendatanepal.com), Creative Commons Attribution
data-2026.11.13.3
Data files built from 86aa753 on main. This is a pre-release outside the signed catalog: teigeodata-sync does not fetch it, so download the files and open them with Layer::open.
What is new
- Cities answer to their other names (D-265). Every city of 5,000 or more takes its GeoNames alternate names in the layer's languages (English, Nepali and Hindi): 8,300 names on 7,018 cities, historical and colloquial ones marked as such. A search for Bombay, Madras, Calcutta, Benares, Allahabad, Gurgaon or Trivandrum finds the city.
- India's towns carry their own-language names (D-263): 1,781 towns named in their state's official language from the Local Government Directory, in 10 languages and scripts (Hindi, Marathi, Tamil, Telugu, Bengali, and more), each checked to be wholly in that script and well formed.
- A city comes before the district of its name that holds it (D-264): "Varanasi" gives the city, then the district. India's municipal corporations rank as cities.
- Every ID already published is kept: no place is added, moved, renamed or removed against
data-2026.11.13.2. The Nepal pack changes in its search ranking only.
baseline-lite.tgd, points-en-lite.tgd and their summaries are byte for byte the ones in data-2026.11.13.2.
Files
| File | What | Rows | Disk | Wire |
|---|---|---|---|---|
world-np-lite.tgd |
Offline layer: the world's countries, first-level units and cities of 5,000 or more with their alternate names, India's districts and statutory towns, and Nepal in full. Names in Nepali, Hindi and English, with heights, zones, zone rules, admin units, outlines and search | 172,339 | 15,207,232 B | 9,476,053 B |
baseline-lite.tgd |
The baseline layer the engine can embed: the world with each country's slice, every name | 39,395 | 5,000,392 B | 3,647,834 B |
points-en-lite.tgd |
Points, heights and English names with search; no outlines | 39,395 | 1,838,332 B | 1,178,096 B |
np-places-lite.tgd |
Nepal's full lite pack, with names by language and outline parts by level and level of detail | 95,716 | 87,529,856 B | see np-summary-lite.md |
ids-lite.tar.gz |
The ID registries (source keys to stable IDs) the files were numbered with; cargo xtask ids restore puts them back |
1,107,044 B |
Each summary-*.md describes its layer, np-summary-lite.md describes the Nepal pack and np-attribution-lite.md holds its attribution.
Code
Read these files with the crates at v0.2.0 or later; the format did not change:
teigeodata-engine = { git = "https://github.com/teispace/teigeodata", tag = "v0.3.0" }Verify
shasum -a 256 -c SHA256SUMS
Licence
Lite build, CC BY 4.0. The layer's sources and their notices:
- Made with Natural Earth (https://www.naturalearthdata.com), public domain
- GeoNames (https://www.geonames.org), CC BY 4.0
- Time zones: IANA Time Zone Database 2026f (public domain)
- Unicode CLDR (https://cldr.unicode.org), Copyright © 1991-2026 Unicode, Inc., under the Unicode License v3
- Wikidata (https://www.wikidata.org), CC0 1.0
- Local Government Directory, Ministry of Panchayati Raj, Government of India (https://lgdirectory.gov.in)
- Administrative boundaries: Survey Department of Nepal and UN Resident Coordinator's Office in Nepal, via OCHA HDX (cod-ab-npl), CC BY-IGO 3.0
- Elevation: NASADEM, NASA JPL and LP DAAC
- Buildings: Microsoft Global ML Building Footprints, CDLA-Permissive-2.0 (https://cdla.dev/permissive-2-0/)
Data 2026.11.13.2 (lite)
Data files built from c4add7b on main. This is a pre-release outside the signed catalog: teigeodata-sync does not fetch it, so download the files and open them with Layer::open.
What is new
- India's districts and statutory towns join the offline layer (D-259): 762 districts under their states, and 4,128 of the 5,050 statutory towns in the Local Government Directory, each placed at a populated place of its name within its own district (else its state), never at a namesake elsewhere. 3,433 of them give an existing city its official name and parent; 695 are new places.
- Countries are named in Nepali from Unicode CLDR 49, with CLDR's short and variant forms as alternate names: 237 of 248 countries and disputed areas, up from 160 (D-257). These names carry the new name origin
standard(code 10); readers before it readOther(10). - Every ID already published is kept: the layer adds 1,457 places and moves, renames or removes none. The ID registries the files were numbered with now ship with each release as
ids-lite.tar.gz(D-261), so later builds keep every published ID.
baseline-lite.tgd, points-en-lite.tgd, np-places-lite.tgd and their summaries are byte for byte the ones in data-2026.11.13.
Files
| File | What | Rows | Disk | Wire |
|---|---|---|---|---|
world-np-lite.tgd |
Offline layer: the world's countries, first-level units and cities of 5,000 or more, India's districts and statutory towns, and Nepal in full. Names in Nepali, Hindi and English, with heights, zones, zone rules, admin units, outlines and search | 172,339 | 14,950,016 B | 9,326,040 B |
baseline-lite.tgd |
The baseline layer the engine can embed: the world with each country's slice, every name | 39,395 | 5,000,392 B | 3,647,834 B |
points-en-lite.tgd |
Points, heights and English names with search; no outlines | 39,395 | 1,838,332 B | 1,178,096 B |
np-places-lite.tgd |
Nepal's full lite pack, with names by language and outline parts by level and level of detail | 95,716 | 87,529,920 B | 82,684,681 B |
ids-lite.tar.gz |
The ID registries (source keys to stable IDs) the files were numbered with; cargo xtask ids restore puts them back |
1,107,063 B |
Each summary-*.md describes its layer, np-summary-lite.md describes the Nepal pack and np-attribution-lite.md holds its attribution.
Code
Read these files with the crates at v0.2.0 or later; the format did not change:
teigeodata-engine = { git = "https://github.com/teispace/teigeodata", tag = "v0.2.0" }Verify
shasum -a 256 -c SHA256SUMS
Licence
Lite build, CC BY 4.0. The layer's sources and their notices:
- Made with Natural Earth (https://www.naturalearthdata.com), public domain
- GeoNames (https://www.geonames.org), CC BY 4.0
- Time zones: IANA Time Zone Database 2026f (public domain)
- Unicode CLDR (https://cldr.unicode.org), Copyright © 1991-2026 Unicode, Inc., under the Unicode License v3
- Wikidata (https://www.wikidata.org), CC0 1.0
- Local Government Directory, Ministry of Panchayati Raj, Government of India (https://lgdirectory.gov.in)
- Administrative boundaries: Survey Department of Nepal and UN Resident Coordinator's Office in Nepal, via OCHA HDX (cod-ab-npl), CC BY-IGO 3.0
- Elevation: NASADEM, NASA JPL and LP DAAC
- Buildings: Microsoft Global ML Building Footprints, CDLA-Permissive-2.0 (https://cdla.dev/permissive-2-0/)
- Buildings: Google Open Buildings v3 (Sirko et al. 2021, arXiv:2107.12283), CC BY 4.0
- Central Bureau of Statistics (now National Statistics Office), Nepal, via Open Data Nepal (https://opendatanepal.com), Creative Commons Attribution