This data is used by JrUnify-cloud

CSV rows use the stop-only contract name,latitude,longitude,okres,country. Approximate town coordinates are not valid geodata. External ISO country codes are normalized to the historical JDF identifiers by download.py.

Run a complete refresh into staging before replacing checked-in files:

python download.py /path/to/empty-staging-directory

The updater validates rows, removes spatially coincident duplicates, stages each source atomically, and reports every failed source after attempting the full inventory. See SOURCE_AUDIT.md for the DADOF audit and known retired endpoints. Pass the repository's other directory to JrUtil; the rail directory uses the separate SR70 four-column contract.

For residual coordinates, gapfill.py osm accepts one or more Overpass bounding boxes and caches the batched extracts. OSM candidates without municipality tags may match through the preloaded okres index. gapfill.py mapy is the final fallback and reads its credential only from MAPY_API_KEY. Both commands accept an input CSV with stop_id,name,municipality,region,country, write accepted stop coordinates to the geodata CSV, and put all ambiguous/missing rows in a review CSV. Mapy raw responses are never stored.

For a small residual, prefer gapfill.py osm-search: it performs one cached, rate-limited Nominatim search per stop instead of serial Overpass boxes. Nearby platforms with the same exact identity are emitted as one stop-place centroid; distant or competing clusters remain in review. Mapy can additionally write a structured --candidate-review CSV containing candidate labels and coordinates, but no raw responses or credential. Regenerated stop IDs are not used as matching configuration; exact locality suffixes and route context handle incorrect source municipalities without run-specific overrides.

gapfill.py audit GTFS_DIRECTORY MERGED_JDF_ZIP OUTPUT.csv builds that input from an existing bundle, so refreshing coordinates does not require another JrUtil conversion. Its default is --coordinate-status missing. Use --coordinate-status estimated to reconsider route-derived coordinates, or all for both. These modes read source_stop_metadata.parquet (override with --metadata) rather than relying on the rendered GTFS name suffix and therefore require PyArrow. The audit embeds deduplicated preceding/following timed anchors and ignores degenerate trips with fewer than two distinct stop places. Use --minimum-run-length 5 --include-unresolved-termini to produce a focused quality work list. Its additional columns identify the longest consecutive unresolved run, terminal involvement, affected trips and routes. These fields prioritize review; they do not weaken candidate validation.

For example, refine estimated coordinates without making Mapy a live build dependency:

uv run --with pyarrow --with requests --with lxml --with pyproj \
  --with shapely --with pyshp python gapfill.py audit \
  BUNDLE/gtfs-intermediate MERGED_JDF.zip estimated.csv \
  --coordinate-status estimated
MAPY_API_KEY=... uv run --with requests --with lxml --with pyproj \
  --with shapely --with pyshp python gapfill.py mapy \
  estimated.csv accepted.csv --review review.csv \
  --candidate-review candidates.csv \
  --gtfs BUNDLE/gtfs-intermediate --jdf MERGED_JDF.zip
python gapfill.py merge other/gapfill.csv accepted.csv merged.csv

Mapy searches only POIs, hard-restricts the country, and prefers the current estimate. Full names, expanded JDF abbreviations, and locality-qualified name variants are tried. Foreign OSM/Nominatim searches additionally try localized stop/station terminology and retain name:de, name:pl and name:sk aliases. Exact and fuzzy matches (default threshold 0.70, with a 0.10 winner margin) may be accepted, but a candidate for an estimated stop must satisfy the same 2 km plus 150 km/h scheduled-time ceiling as JrUtil. When normal name selection remains ambiguous, an estimated stop may use one clearly dominant route-supported candidate within 10 km; it requires at least 0.55 name similarity and a 0.12 combined name/distance margin. Exact locality suffixes may repair a bad JDF municipality only when that route check succeeds. Town/address/unrelated POIs and distant same-name stops remain in review. Accepted rows use the five-column stop contract; merge them into the checked supplement so ordinary feed generation remains offline and deterministic.

After enabling JrUtil's regional-adjacent route policy, build the actionable post-filter work list without another conversion or a JrUtil cache:

python residual_plan.py GTFS_DIRECTORY MERGED_JDF_ZIP AUDIT.csv WORKLIST.csv \
  --external-geodata other

The utility uses the emitted GTFS trip set for service validity, applies the same international-route distance rules to the merged JDF, and removes only exact refreshed-source matches with compatible geography and a candidate cluster under one kilometre. The resulting CSV includes source stop IDs, route distinctions, route names and the recommended next matching stage.

Stop ID registry

registry/ pins public stop IDs so that they stay the same between exports. JrUtil reads it with --stop-registry=registry (in merge-jdf, jdf-to-bundle and regional-gtfs-overlay). All three files are append-only:

• stops.csv (id,town,district,nearby_place,okres,country,lat,lon,status,note) pins the merged-JDF stop number N, published as jdf:stop:N.
  • Rows that share an id are aliases. Record a rename by adding a row, not by editing one.
  • A stop that is no longer served gets status=retired. Its row stays.
  • A merged_into:<id> note marks a duplicate; JrUtil uses the target number.
  • An id is never reused.
  • lat/lon are optional reference coordinates, not geodata. They only tell apart stops whose identity is the same. Such stops must be at least 75 m apart, the distance at which JrUtil stops merging same-named stops.
• posts.csv (stop_id,post_key,lat,lon,status,note) pins inferred post ordinals (est:<k>). An inferred post within 25 m of a registered one reuses its ordinal, so the ID survives changes in the evidence behind it.
• overlay_places.csv (source_id,group_key,place_id,status,note) pins regional-overlay stop places that match no national stop. Their default ID is a hash of the source group, and that hash changes when the stop is renamed.

Stops the registry does not know get a provisional number of at least 1000000000. With --stop-registry-candidates=FILE, each JrUtil command writes what is missing: unregistered stops and new spellings of registered ones (merge-jdf), new post ordinals (jdf-to-bundle) and unpinned places (regional-gtfs-overlay). To register them, fill a decision column with new or alias (with alias_of, unless alias_of_suggestion names a single value), then run:

python registry.py promote stop-candidates.csv
python registry.py validate

promote --accept-new treats every undecided row with reason=new as new. Running it on the first build's candidates numbers every stop from 1, which is the one planned stop-ID break.

Do not pass the repository root to fix-jdf -g. It loads *.csv recursively, and the registry files are not geodata.

Route rules

routes/ holds reviewed corrections of JDF lines. JrUtil reads them in jdf-to-bundle (--transport-mode-rules, --route-presentation-rules); the presentation rules are applied again after regional-gtfs-overlay (--route-presentation-rules), so they win over regional feed values. Every row needs a reason.

Rows select routes by licence and an optional agency_id (IČO):

• 915003: one licence
• 915001-915019: an inclusive range of equally long licences
• 915*: a licence prefix; * alone matches every licence

The most specific rule wins: a single licence, then a range (narrower first), then a prefix (longer first). At equal specificity a rule naming an agency wins. Two equally specific rules that can set the same field of one route are an error.

• transport-modes.csv (agency_id,licence,public_line,expected_mode,effective_mode,reason) replaces the JDF mode (A bus, E tram, T trolleybus, L cable car, M metro, P ferry). The rule applies only while the source still has expected_mode and, when public_line (11 or 1-19) is given, a numeric public line in it.
• presentation.csv (agency_id,licence,route_short_name,route_color,route_text_color,reason) overrides the published line marking and colours (six hex digits, no #). An empty field keeps the computed value, and each field is resolved on its own, so 915* can set colours while 915003 renames one line. Regional-overlay route matching still uses the source line marking.

Check changes with python routes.py.

Regional overlay overrides

overlay/ holds reviewed identity overrides of the regional GTFS overlay (source_namespace,source_id,target_namespace,target_id,valid_from,valid_to,review_note). The overlay policies name these files; Oběhy passes --overrides-root=overlay to regional-gtfs-overlay.

Filtered JDF rules

filtered-jdf/rules-v1.json lists the operators, integrated-system codes and line prefixes whose CIS lines the filtered national JDF feed omits because regional feeds already publish them.

Railway sources

SR70.csv

Extract from official SR70 made with sr70_download.py. The current checked snapshot was generated from Číselník SR70 od 15. srpna 2026.xlsx, effective 2026-08-15, with source SHA-256 6adac3d1ebce48ee09f6cf75032d09c43ad9a957cf938c37a59872499971c6f0.

Regenerate both rail snapshots atomically from the same downloaded workbook:

python sr70_download.py --output-dir rail "/path/to/Číselník SR70.xlsx"

SR70_Nazev20.csv

Variant of SR70.csv with names from column NÁZEV20. Intended for matching GRAPP names and for compact passenger-facing CZPTT fallback route endpoints. Official values are retained verbatim; presentation-specific cleanup such as removing the standalone terminal z or nz happens in the consumer.

Other sources

This includes anything other than railways. Buses, trams, funiculars...

TODO: Some of these sources include CIS JŘ IDs and distinguish stop posts. We should make use of them, but that requires CIS JŘ IDs in JDF.

LibereckyKraj.csv

Open data: https://dopravnimapy.kraj-lbc.cz/opendata/?id=584a7ad7-1680-4d8d-a20b-7068c371c416

JihoceskyKraj.csv

Open data: https://geoportal.kraj-jihocesky.gov.cz/gs/zastavky-verejne-dopravy/

PID.csv

Open data: http://opendata.praha.eu/dataset/zastavky-pid-jednotlive-oznacniky-geodata/resource/8c912738-d4cb-41ca-b223-a8e455cd4c80

MPVNet_PID.csv

Scraped: https://mpvnet.cz/pid/map

MPVNet_ODIS.csv

Scraped: https://mpvnet.cz/odis/map

MPVNet_Zlin.csv

Scraped: https://mpvnet.cz/zlin/map

MPVNet_IDOL.csv

Scraped: https://mpvnet.cz/idol/map

IDSJMK_Map.csv

Scraped: https://mapa.idsjmk.cz/

QRideDUK.csv

Scraped: https://tabule.portabo.cz/ (DÚK). Abbreviated Ústí n.L., names are additionally emitted as Ústí nad Labem, to match JDF MHD naming; this replaced the former MapaDUK.csv, whose stop positions were identical.

MapaIREDO.csv

Scraped: https://tabule.oredo.cz/idspublic/

MapaIDSOK.csv

Scraped: https://cestujok.cz/idspublic/

UsteckyKraj.csv

From ArcGIS: https://ags.kr-ustecky.cz/arcgis/rest/services/Doprava/zastavky/MapServer

PlzenskyKraj.csv

From ArcGIS: http://mapy.plzensky-kraj.cz/ArcGIS/rest/services/zastavky/MapServer/1

MoravskoslezskyKraj.csv

From ArcGIS (data.msk.cz): https://services8.arcgis.com/jfWD14yYevYeDEj7/arcgis/rest/services/Zast%C3%A1vky\_ve%C5%99ejn%C3%A9\_hromadn%C3%A9\_dopravy\_v\_Moravskoslezsk%C3%A9m\_kraj/FeatureServer/0

KrajVysocina.csv

From ArcGIS: http://geoportal.kr-vysocina.cz/arcgis/rest/services/Trasy\_dopravy/zastavky/MapServer

KarlovarskyKraj.csv

From ArcGIS: http://geoportal.kr-karlovarsky.cz/arcgis/rest/services/UAP/UAP\_msd/MapServer

Retained.csv

Rows that a refreshed catalogue no longer publishes but whose names the current feed still references, copied verbatim from the previous snapshot of their source. This keeps a refresh from silently dropping coverage when an upstream catalogue shrinks (MPVNet, 2026-10) or renames a stop. Drop a row once the JDF stops using that name.