SNCB × BMC GTFS-RT comparison

A GitHub Pages report with side-by-side feed coverage, trip matching and expandable arrival/departure delays at each reported stop. The report is built from two concurrently fetched snapshots, with no live browser requests to the upstream feeds.

Feeds

• SNCB: https://sncb-opendata.hafas.de/gtfs/realtime/d22ad6759ee25bg84ddb6c818g4dc4de_TC
• BMC: https://api-management-discovery-production.azure-api.net/api/gtfs/feed/nmbssncb/rt/trip-update?format=protobuf

The originally supplied BMC URL ending in /trip-update.pbf returns 404. The replacement opendata-discovery-gtfs-realtime.api.production.belgianmobility.io hostname did not resolve during verification. The working BMC endpoint uses /trip-update?format=protobuf; without the format option it returns JSON. Both feeds were successfully decoded during development.

Both static archives are downloaded on every workflow run, alongside the realtime feeds:

• SNCB static GTFS: https://sncb-opendata.hafas.de/gtfs/static/c21ac6758dd25af84cca5b707f3cb3de, overridden with SNCB_STATIC_GTFS_URL.
• BMC static GTFS: https://gtfs.flatturtle.cloud/sncb-nmbs/_latest/sncb-nmbs-gtfs.zip, overridden with STATIC_GTFS_URL.

Each archive has its own protobuf trip index, ID audit and expandable trip information.

GitHub Pages setup

1. Push these files to the repository's main branch.
2. In Settings → Pages → Build and deployment, set Source to GitHub Actions.
3. Run Build and deploy feed comparison from the Actions tab, or let the push trigger it.

The workflow in .github/workflows/pages.yml installs the pinned protobuf dependencies, tests the comparison logic, fetches both feeds, builds _site/, uploads the Pages artifact and deploys it through the github-pages environment. It does not commit generated data to the repository. The deployed URL appears in the workflow's deployment output; for this repository it is normally https://irail.github.io/gtfs-rt-comparison/.

Refresh schedule

schedule:
  - cron: '0 7-9,17-20 * * *'
    timezone: Europe/Brussels

Runs every day at 07:00, 08:00, 09:00, 17:00, 18:00, 19:00 and 20:00 Brussels time, including daylight-saving changes. Pushes to main and manual dispatch also build and deploy immediately. To collect only on weekdays, replace the final * with 1-5.

GitHub supports timezone-aware schedules. Scheduled runs may be delayed or dropped during high load, and public-repository schedules are disabled after 60 days without repository activity. The page displays capture and publication timestamps, so it remains clear which snapshot is shown.

Optional endpoint overrides / authentication

Defaults work without credentials at the time of verification. Set these in Settings → Secrets and variables → Actions if the providers change:

Feed	Repository URL variable	Header-name variable	Header-value secret
SNCB	SNCB_FEED_URL	SNCB_AUTH_HEADER	SNCB_AUTH_VALUE
BMC	BMC_FEED_URL	BMC_AUTH_HEADER	BMC_AUTH_VALUE

Both header fields must be set to send authentication. Credentials and overridden URLs are not included in the public report. Use the provider's actual header name, such as Ocp-Apim-Subscription-Key, if required.

Comparison rules

• Match exact (trip_id, start_date, start_time) after removing the observed BMC gt:nmbssncb: trip namespace. Preserve original identifiers in the details. Missing descriptor fields remain empty; no heuristic matching is performed.
• Stop details remove only the observed gs:nmbssncb: namespace, retain platform suffixes, and align repeated stops by occurrence because SNCB does not supply stop sequences. SNCB's stop order takes precedence; extra BMC stops follow.
• A delayed trip is a non-cancelled trip with at least one explicit positive trip-level, arrival or departure delay. The displayed trip delay is the maximum of all usable explicit values in the snapshot, including past stops. It is not necessarily the current delay.
• Zero means an explicit zero delay; missing fields mean unknown. Absolute event times alone do not establish a delay without the static schedule. No inferred delay propagation is performed.
• Cancelled service instances and NO_DATA/SKIPPED stop updates are excluded from delay counts. Deleted feed entities are ignored. Duplicate trip instances retain the newest trip update (first wins if timestamps tie).
• Both raw stop-update counts and counts with usable event data are shown, because BMC includes NO_DATA stops absent from SNCB.
• The agreement summary compares maximum explicit delay values across all matched, non-cancelled trips with values in both feeds. Equal maxima do not establish equal stop-level data.
• The detailed table includes every trip delayed in either feed, including trips absent from the other feed. It is sorted by the larger delay, descending. Positive Δ means BMC reports a larger maximum.
• Feeds older than five minutes at collection are flagged. A missing timestamp leaves freshness unknown. Different publication times may explain differences. Each run replaces the previous snapshot; this is not a historical archive.
• Failed feeds have unknown statistics and comparisons, rather than zero counts. The report still deploys the healthy feed's data and a visible error, and emits a GitHub Actions warning. Differential feeds are rejected because they cannot be compared as independent full snapshots.

See the GTFS realtime specification for field semantics.

Static trip lookups

The report checks all realtime trips against both static trips.txt databases. Each source has its own summary of IDs found, IDs absent, active service dates and start-time differences. The delayed-trip table has In SNCB static GTFS? and In BMC static GTFS? columns, displaying Yes, No or Unknown independently. Missing IDs appear in red. Each Yes has a Trip info button for that source's train number, headsign, route, calendar and scheduled stops.

Only the known gt:nmbssncb: namespace is added or removed for lookup. All other parts of IDs remain intact. Service activity and start-time checks remain separate from ID presence. An unavailable static source leaves only its own matches unknown and does not prevent the other lookup or realtime report from publishing.

Train website links prefer the SNCB static record and fall back to the BMC record. Links use an advertised numeric train number and the realtime service date, and show the websites' latest information rather than captured delays. An ID match does not establish complete planner applicability: service dates, start times and stop compatibility also matter.

The saved source comparison records why checking both databases matters.

The workflow publishes:

• sncb-static-trips.pb and static-trips.pb: independent SNCB and BMC protobuf maps of every original static trip ID to its trip metadata, plus route records. Its schema is schemas/static_trips.proto. It is a trip/route index, not a complete routing database; schedules and service-date checks for current matching realtime trips are in report.json.
• sncb-static-audit.json and static-audit.json: each archive’s SHA-256, source, feed version, download metadata, all-trip summary counts and evidence for every SNCB trip missing from BMC.
• report.json: realtime hashes/timestamps, both static audits and matching trip/calendar/schedule information used by the page.

To query the protobuf index locally:

from pathlib import Path
from scripts.static_trips_pb2 import TripIndex

index = TripIndex.FromString(Path('_site/sncb-static-trips.pb').read_bytes())
trip_id = 'YOUR_TRIP_ID'
if trip_id in index.trips:
    trip = index.trips[trip_id]
    print(trip.trip_short_name, trip.trip_headsign, trip.service_id)

Local development

python3 -m venv .venv
.venv/bin/python -m pip install -r requirements.txt
.venv/bin/python -m unittest discover -s tests -v
.venv/bin/python scripts/build_report.py
.venv/bin/python -m http.server 8000 --directory _site

Open http://localhost:8000. To verify with saved protobuf snapshots:

.venv/bin/python scripts/build_report.py --sncb-file /tmp/sncb.pbf --bmc-file /tmp/bmc.pbf --static-file /tmp/static.zip --sncb-static-file /tmp/sncb-static.zip

Reports built with saved files display a local verification banner. Production builds fetch live snapshots. report.json contains summary metadata and the original trip/stop details for delayed trips. The site also includes the static index, schema and audit, plus .nojekyll, HTML, CSS and JavaScript. All asset paths are relative so the site works under a GitHub Pages repository subpath.