# mta.jamestannahill.com — Full Content Index

## Identity
- **Name:** Subway Intel
- **URL:** https://mta.jamestannahill.com/
- **Operator:** James Tannahill (https://jamestannahill.com)
- **Focus area:** Midtown Manhattan, New York City (all NYC subway lines covered)
- **Update cadence:** MTA GTFS-Realtime polled every 30 seconds; WebSocket push to browsers on each cycle

## Product summary
Subway Intel transforms MTA GTFS-Realtime protobuf feeds into a consumer dashboard with six views: network-wide line health, personal saved-station boards, geolocated nearby stations, heuristic delay intelligence, a Mapbox live train map, and point-to-point trip planning with one-transfer fallback.

Unlike the official MTA app, Subway Intel emphasizes:
- Per-line **Network Pulse** with NOMINAL / DELAYED / DISRUPTED status derived from live headways
- **Delay Intel** heuristics that surface delays from trip-update deviation before official alerts propagate
- **Track diagrams** with interpolated train positions and YOU ARE HERE markers
- **Arrival feedback loop**: users confirm train presence (HERE? Y/N); corrections adjust displayed countdowns per route
- **Smart commute** timing via `/api/commute` with multi-parent-stop expansion for complex stations (e.g. Times Sq-42 St)

## Views and hash routes

| Route | Purpose |
|---|---|
| `#/pulse` | Network Pulse — line health summary, worst offenders, expandable line-group accordion, saved-line badges, track diagram |
| `#/stations` | My Stations — user-saved stops with uptown/downtown/both direction filter, live m:ss countdowns |
| `#/nearby` | Nearby — GPS or address search, 3 nearest stations, select for arrivals + diagram; `?near=STOP_ID` deep link |
| `#/intel` | Delay Intel — heuristic signals across all live arrivals; filterable by saved lines |
| `#/map` | Live Map — Mapbox GL JS, NYC DOT track GeoJSON, live vehicle positions via WebSocket |
| `#/plan` | Plan Trip — origin/destination search, live departure options, transfer details |

## Data pipeline

```
MTA GTFS-RT feeds (protobuf, per line group) + camsys alerts JSON
        │
        ▼
FastAPI backend (Python, EC2)
  • ETag + content-hash skip for unchanged feeds
  • parse_feed → trip updates / arrivals
  • parse_vehicle_positions → live train locations
  • parse_alerts + parse_alerts_json → service alerts
  • LiveState in-memory aggregation
  • WebSocket broadcast (/ws) on every poll
  • TimescaleDB writer (historical positions + alerts)
        │
        ▼
React 19 SPA (Vite, TypeScript, Mapbox GL JS)
```

## Line groups ingested
1, 2, 3, 4, 5, 6, 7, S (shuttle), A, C, E, B, D, F, M, G, J, Z, L, N, Q, R, W, SIR (Staten Island Railway)

## API reference

Base: `https://mta.jamestannahill.com`

### Live data
- `GET /api/stations/{stop_id}/arrivals` — `{ stop_id, arrivals: [{ route_id, arrival_time, delay_sec, direction }] }`
- `GET /api/network` — `{ lines: [{ route_id, status, avg_delay_sec, headway_variance, alerts }] }`
- `GET /api/delay-intel` — `{ signals: [...] }` heuristic delay objects
- `WS /ws` — full LiveState JSON snapshot on connect and every poll

### Static / lookup
- `GET /api/stops/search?q=` — station name autocomplete
- `GET /api/stops/nearest?lat=&lon=&limit=` — haversine nearest parent stops
- `GET /api/stops/all` — all parent stations (1h cache)
- `GET /api/tracks` — track geometry GeoJSON (1h cache)
- `GET /api/subway-lines` — NYC ArcGIS physical line GeoJSON (24h cache)
- `GET /api/routes/{route_id}/stops?direction=` — ordered stop sequence

### Trip planning
- `GET /api/commute?origin=&destination=` — direct or one-transfer options with `compute_departure` timing

### Feedback
- `POST /api/feedback/arrival` — `{ stop_id, route_id, scheduled_arrival, train_present }`
- `GET /api/feedback/corrections` — per-route bias corrections applied to countdowns

### Config
- `GET /api/config` — `{ mapbox_token }` for Map view

## Heuristics

**Line health status:** Derived from average delay seconds and headway variance per route vs. nominal patterns. States: NOMINAL, DELAYED, DISRUPTED.

**Delay Intel signals:** Computed from all in-memory arrivals — headway gaps, schedule deviation, bunching detection.

**Commute departure:** `compute_departure(next_arrival, travel_sec)` recommends leave-in minutes based on walking buffer and live delay.

## Static pages (crawlable)
- `/` — SPA shell with JSON-LD WebSite + WebApplication schema
- `/docs.html` — technical documentation (this content in HTML form)
- `/llms.txt` — concise AI index
- `/llms-full.txt` — this file
- `/sitemap.xml` — URL index
- `/robots.txt` — allow all, sitemap reference
- `/manifest.json` — PWA manifest

## Social / SEO
- Open Graph + Twitter Card meta on `/`
- OG image: https://mta.jamestannahill.com/og.png (1200×630)
- Canonical: https://mta.jamestannahill.com/
- `<link rel="ai" href="/llms.txt">` in document head

## License
See repository LICENSE. MTA GTFS data subject to MTA terms of use.

## Related properties by same operator
- Weather: https://wx.jamestannahill.com/
- Personal site: https://jamestannahill.com/