Documentation

Getting started

Datasets are distributed as signed links rather than public files. Request a copy from the dataset page, and we send a link that stays valid for seven days. No account, no API key.

# the email contains one link per release
curl -O "https://dl.deeplinedata.xyz/<signed-token>/urbangrid-eu-2026.10.zip"
curl -O "https://dl.deeplinedata.xyz/<signed-token>/SHA256SUMS"
sha256sum -c SHA256SUMS --ignore-missing

Archives unpack into a directory named after the release, so two versions can sit side by side without colliding. A typical layout:

urbangrid-eu-2026.10/
  buildings.geojson
  roads.geojson
  ATTRIBUTION.txt
  schema.json
  README.md

Formats

Tabular sets ship as UTF-8 CSV with a header row. Geometry is GeoJSON in WGS84, longitudes first. Transit feeds follow GTFS, text corpora use JSONL with one document per line, and transcripts are TSV keyed by utterance id.

Format	Used for	Note
CSV	tabular	UTF-8, header row, comma delimiter
GeoJSON	geospatial	WGS84, coordinates as [lon, lat]
GTFS	transit feeds	standard .txt tables inside a zip
JSONL	text corpora	one document per line, UTF-8
TSV	transcripts	tab-separated, keyed by utterance id
JSON	reference tables	single object, no trailing commas

Licensing

Most datasets are CC BY 4.0. Geospatial derivatives are ODbL 1.0 because the source data requires share-alike. The retail sample is CC BY-NC 4.0. The exact licence is printed on the dataset page and repeated in ATTRIBUTION.txt inside the archive.

Versioning

Releases use yyyy.mm. A file inside a release never changes after publication. If a correction is needed, a new monthly release is cut and the previous one stays available for 90 days. Schema changes get a major bump in the release name.

Checksums

Every signed link comes with a matching SHA256SUMS file. Verify before ingesting; partial downloads are the most common source of corrupt data and the least obvious to debug later.

Mirrors

Signed links resolve to one of two mirrors and rotate weekly. If a transfer stalls, request a fresh link — the next one points at the other mirror. Transfers are rate-limited per token, which keeps large pulls stable for everyone.

Changelog

Release	Change
2026.10	urbangrid-eu: added 6 cities, fixed 4 road segments with broken geometry.
2026.10	airquality-hourly: 12 new stations, station metadata moved to a side table.
2026.09	weather-reanalysis-sub: extended to 2025, grid unchanged.
2026.08	openlex-multilingual: 84k new documents, language labels re-checked.

Status

All datasets are currently available. Scheduled maintenance happens on the first Tuesday of the month, 02:00–03:00 UTC; downloads started before the window are not interrupted.