Industries / Government, education and community / Government / Public transit
Passenger information
This is the work of telling riders where their bus or train is, when it will come, and what has changed.
4
Milestones achieved, from 1992 to 2011.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1992
Arrival times at the stop
In May 1992 the trade weekly Commercial Motor reported that London Buses would start a system in September on route 18, from Sudbury in west London to King's Cross, that told people waiting at stops when the next bus would arrive. It cost £700,000, and London Buses hoped new riders would pay for it within three years. The system, Countdown, later covered the whole city: a Transport for London paper describes predicted arrivals for every one of more than 19,000 bus stops on 700 routes, by internet and phone, plus 2,500 signs at the roadside.
Credit
London Buses; Transport for London
2005
Timetables become open data
In July 2005 Bibiana McHugh of TriMet, the transit agency for Portland, Oregon, began working with Chris Harrelson of Google Maps on a simple file format for timetables, and Google's transit trip planner went live with TriMet's schedules in December 2005. The format, now the General Transit Feed Specification (GTFS), lets any app read any agency's routes, stops and times. An Open Data Institute case study counts Google Transit in 14 cities by December 2006 and GTFS in use at more than 700 agencies by 2013.
Credit
Bibiana McHugh and TriMet; Chris Harrelson and Google
2008
Arrival times on the phone
OneBusAway, built by University of Washington graduate students Brian Ferris and Kari Watkins, went live in Seattle after tests with riders in summer 2008. It gave predicted arrivals from King County Metro's bus tracking data by website, an automated phone line, text message and phone apps, and served more than 50,000 riders a week by 2011. Watkins' study at Seattle bus stops found that riders using real-time information waited almost 2 minutes less than riders using timetables, and did not overestimate their wait, while timetable users did.
Credit
Brian Ferris, Kari Watkins and Alan Borning, University of Washington
2011
One format for live data
In 2011 GTFS Realtime added live vehicle positions, predicted arrival times and service alerts to the timetable format, so apps could read any agency's live data the same way. Its development began with six agencies, and OneBusAway wrote the first converters from older tracking systems. In 2019 the nonprofit MobilityData was set up to maintain both formats with their users.
Credit
Google and the first six agencies; OneBusAway; MobilityData
Watch for
Nine in ten predictions right
CDTA, the transit agency for Albany, New York, says the real-time system it launched with the app maker Moovit achieves "a 90% arrival prediction accuracy for fixed routes", using vehicle positions, current traffic and past travel patterns. It does not say how accuracy is defined.
Achieved whenCDTA publishes its definition, for example the share of predictions within one minute of the actual arrival at each look-ahead time, and someone outside reproduces the figure from its public real-time data.
Claimed by
CDTA and Moovit, April 2026
Open
01
Alerts apps can act on
A tool that takes a notice of a service change as staff write it ("Route 12 detoured off Main St between 5th and 9th until Friday"), together with the agency's GTFS timetable, and fills in the structured parts of a GTFS Realtime service alert: the affected routes and stops by their IDs, the effect (no service, detour, reduced service) and the start and end times. Test it on an archive of past alerts whose structured fields have been checked by hand.
The briefCredit · unclaimed02
Apps checked at the stop
A field test of the departure times that trip-planning apps show riders. At 30 stops in three cities, record at set times what each of four apps shows for the next departures and whether it marks each as live or scheduled, then log when each vehicle actually left, from the agency's public vehicle positions or by watching.
The briefCredit · unclaimed
Open 01 · the brief
Alerts apps can act on
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A tool that takes a notice of a service change as staff write it ("Route 12 detoured off Main St between 5th and 9th until Friday"), together with the agency's GTFS timetable, and fills in the structured parts of a GTFS Realtime service alert: the affected routes and stops by their IDs, the effect (no service, detour, reduced service) and the start and end times. Test it on an archive of past alerts whose structured fields have been checked by hand.
Why it is still open
MobilityData's guidelines warn that an alert tied to the wrong stops or given the wrong effect leads trip planners to give "incorrect journey suggestions". Published work on alert text does something else: a 2025 demonstrator turns structured alerts into plain-language summaries and translations, and a 2025 paper predicts how long incidents last from New York MTA alert text. No public scored test of filling in the structured fields was found.
Built with what exists
Agencies' public GTFS timetables and GTFS Realtime alert feeds, saved every few minutes; a model that picks from a fixed list of answers for the effect field (see the Note "A thousand questions, one pass"); fuzzy matching of street and stop names to stop IDs; Python.
First step
Start saving one large agency's GTFS Realtime alert feed every 15 minutes, and label the first 50 alerts by hand.
Done looks like
At least 500 archived alerts from at least five agencies, checked by hand, and a published score for routes, stops, effect and time window, each reported separately, with the code and the labeled set public.
Open 02 · the brief
Apps checked at the stop
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A field test of the departure times that trip-planning apps show riders. At 30 stops in three cities, record at set times what each of four apps shows for the next departures and whether it marks each as live or scheduled, then log when each vehicle actually left, from the agency's public vehicle positions or by watching.
Why it is still open
Comparisons of the apps rest on complaints. A June 2026 analysis of one- to three-star app store reviews by Unstar, a site that analyzes app reviews, finds Google Maps "the most likely to show a scheduled time as if it were real", and no controlled test of the apps against actual departures was found. Open tools that score an agency's own predictions, such as accuracy.rt.gtfs.zone, test the data feed, not what each app shows.
Built with what exists
Phones with screen recording; the agencies' public GTFS Realtime vehicle positions to log actual departures; Python for matching and scoring.
First step
Pick one stop on a frequent route, record what four apps show every 10 minutes for two hours, and match each departure to the vehicle positions feed.
Done looks like
At least 1,000 recorded departures in three cities, with a published table for each app of the median error in minutes, the share of departures more than two minutes off, and the share of scheduled times shown as live.