mrtdown-data

Station, service, and layout data for Singapore's rail network, plus records of disruptions, maintenance, and infrastructure issues.

Use the tools here to inspect records, validate changes, process evidence, and build data exports for applications.

Get started

Use the Node.js version in .nvmrc, then install dependencies and validate the data:

npm ci
npm run build:packages
npm run data:validate

The CLI can list records, show an individual record, or inspect an issue's state:

node packages/cli/dist/index.js --data-dir data list station
node packages/cli/dist/index.js --data-dir data show station <id>
node packages/cli/dist/index.js --data-dir data issue state <id>

See the CLI command reference for creation commands and other tools.

Explore the data

The dataset lives in data/. External GTFS files can be inspected locally.

Data	Contents
station/	Stations, including exits and optional platform information.
line/, service/, operator/, town/, landmark/	The rail network and nearby places.
issue/YYYY/MM/<issue_id>/	Issue metadata (issue.json), source evidence (evidence.ndjson), and append-only impact events (impact.ndjson).
schematic-map/system/	Map rules and design constraints; see the map plan.
rights/source-registry.json	Source rights and attribution rules.
External GTFS files	Schedule, realtime, and departure inspection; arrival generation and publication require mapping and licensing review.

Issue types are disruption, maintenance, and infra; ongoing issues use open-ended periods.

Validate every data change with npm run data:validate. For station layout updates, follow the exit and platform source policy. These commands import LTA exits and report possible platform labels inferred from reviewed observations:

npm run data:import:lta-station-exits -- /path/to/exits.geojson
npm run data:platforms:infer -- --write /tmp/platform-inference.json

Build and publish exports

npm run pages:build

This writes pages-dist/ with browsable JSON records, a manifest, and ZIP and TAR archives. It validates the exported data and redacts source text and rendered content when the source registry disallows public export. Generated test data appears under fixtures/ in the same bundle.

For relevant changes, CI builds branch and pull request previews as temporary artifacts. The Pages deployment workflow publishes from main, then asks mrtdown-site to import the new archive. The site serves the runtime API. See the workflow for the required site URL and token secrets.

Process evidence

The triage tools use AI models to turn source evidence into issue records. Webhook ingestion requires OPENAI_API_KEY and makes paid model calls. Set MESSAGE to a JSON payload matching the shared webhook schemas, then run:

npm run ingest:webhook

The triage guide covers crowd reports, targeted repairs, and model setup. fixtures/ingest/crowd-report.json is a sample payload.

Inspect past ingestion outcomes without model calls or changes to data/:

npm run triage:regressions -- --list

Model evaluations (npm run test:eval) and regression --replay make paid API calls and require OPENAI_API_KEY. Regression replay uses a temporary copy of the recorded data revision. These checks are separate from deterministic tests.

Work on the tools

The packages separate data definitions, file access, and evidence processing:

Package	Purpose and dependencies
core	Schemas, time helpers, and state helpers; independent of the other packages.
ingest-contracts	Webhook payload schemas; independent of the other packages.
fs	Read, write, validate, and export files; depends on core.
triage	Process evidence and replay outcomes; depends on core, fs, and ingest-contracts.
cli	Provide command-line access to the tools; depends on core, fs, and triage.

Run the repository checks and deterministic tests before committing:

npm run check  # Lint, package boundaries, and documentation links
npm test       # Deterministic tests; generates fixtures first

npm run lint runs lint checks alone. Package-specific build and test commands are listed in package.json. Tests and examples use generated data in fixtures/generated/data; recorded ingestion cases live in fixtures/triage-regressions.

Keep generated output out of commits and separate bulk data changes from code changes where practical. See AGENTS.md for contribution rules and the plans index for active work and migration history.

Licenses and sources

Material	License or terms
MRTDown-authored data, curation, and export metadata	CC-BY-4.0.
Code, scripts, tooling, and documentation	MIT License.
LTA data in station exits	Singapore Open Data Licence v1.0.
Third-party evidence	Upstream rights and terms apply; MRTDown does not relicense these materials.

Observed platform facts and recorded inferences follow the station layout source policy. The source registry records recurring source classes and attribution requirements.