gtfs-publisher

A feed-agnostic GTFS publisher. Reads each operator's upstream source via a per-feed adapter package, and publishes the three artifacts any GTFS consumer needs:

1. Raw GTFS Schedule zips — the spec-compliant .gtfs.zip for each feed, content-addressed by sha256. This is the artifact external GTFS tooling (onebusaway, gtfs-to-html, transit planners, validators) reads. It is strictly the public GTFS Schedule spec — no per-feed extensions are added here.

2. GTFS-RT feeds — a Fastify HTTP server proxies live vehicle_positions, trip_updates, and service_alerts GTFS-Realtime protobuf feeds. Same upstream sources as the static side; the protocol-buffer wire format is the standard, no extensions.

3. SQLite blobs — <id>-<hash12>.sqlite3.gz, one per feed. This is what the n3ary/app PWA stores in OPFS and queries at runtime. Per-feed extensions are injected here — the cluj adapter, for example, adds networks.network_color and a _neary_config table — but the extensions never leak into the published .gtfs.zip. The zip stays spec-clean; the sqlite blob carries whatever each consumer needs.

Note

Live artifacts: https://gtfs.n3ary.com/feeds.json (R2 via custom domain gtfs.n3ary.com)

What this repo is not

• It is not a GTFS validator. It consumes whatever the adapter produces.
• It is not a feed-source registry. Operators live in feeds/<id>/config.json, with source.publisher pointing at a published adapter package (e.g. @n3ary/gtfs-adapter-<feed> from n3ary/gtfs-adapters).
• It is not a trip planner. It only publishes data.

Repository layout

.
├── packages/
│   ├── gtfs-static/        static pipeline: resolve feeds → acquire zips → make-sqlite → R2
│   ├── gtfs-rt/            Fastify GTFS-RT proxy: serves protobuf over HTTP
│   └── spec/               shared @n3ary/gtfs-spec (types, SQL DDL, ZIP readers)
├── feeds/<id>/config.json per-feed overrides (source.publisher, secrets[], timing, smoke)
├── countries.json          single source of truth for which feeds to publish
├── docs/
│   ├── architecture/       data-pipeline.md (system-level diagram)
│   └── ops/                secrets-and-deploy.md, publishing.md
└── .github/workflows/      daily.yml (publish), pr-validation.yml

Adding a new feed

1. Add feeds/<id>/config.json (copy an existing one as a template).
2. Set source.type to "adapter" + source.publisher to the adapter's npm package name.
3. Add the feed's required secrets to secrets[] (env var names).
4. Add the env vars to the GitHub Actions secrets on this repo, then add them to .github/workflows/daily.yml's env block.
5. Add the Transitous source name (if any) to countries.json's include[] so the catalog lookup finds it.

No code change in apps/gtfs-static/src/cli.ts — the orchestrator is feed-agnostic and discovers adapters by source.publisher at runtime.

Quick start

npm install
npm run pipeline   # full build → outputs/ + (optionally) R2
npm test           # vitest --run --passWithNoTests

Documentation

• docs/architecture/data-pipeline.md — system-level diagram of the static + RT pipelines, what each publishes
• docs/ops/secrets-and-deploy.md — R2 credentials, daily workflow, RT proxy deploy
• docs/standards/version-management.md — the bump-on-PR rule

Contributing

main is protected — every change goes through a PR. .github/workflows/pr-validation.yml bumps package.json#version on the PR branch and runs test + lint + pipeline smoke.

Brand

The visual identity (logo, favicon, social card, wordmarks) lives in the canonical n3ary/branding repo and is served at https://branding.n3ary.com. This repo doesn't ship any brand assets — if you need the GitHub org avatar, the repo social preview, or a logo, pull from there (see the SETUP.md there for the URL map).

Data license

Schedule data © its respective transit operators (per-feed license.attribution_text in feeds.json). Generated for public transit information purposes.

License

PolyForm Noncommercial License 1.0.0 — free for individuals, hobbyists, education, research, and charitable organizations. Any commercial use (paid products, paid services, or hosted services for revenue) needs a separate license from the author. See the LICENSE file for the full terms.