About marta.fail

marta.fail is an independent project aimed at making MARTA (Metropolitan Atlanta Rapid Transit Authority) service alerts more accessible. MARTA is the primary transit system for Fulton, Dekalb,and Clayton Counties in Metro Atlanta, Georgia.

Service alerts are currently only published on the MARTA app or website, X/Twitter, and via an email notification system. This is not ideal.

I started this project because:

• The MARTA app and website do not have a filtering system for alerts, so it is difficult to find the alerts you are looking for. The app provides alerts through a painfully slow scroll, Internet Explorer 3.0-style.
• X/Twitter? I don’t think I need to give a reason there.
• The email notification system only allows signups for large tranches of bus routes and rail lines, and signing up results in an unmanageable number of emails.

MARTA does (as of June 2025) provide some machine-readable data, in the form of a GTFS feed for bus lines and a proprietary API for rail lines. However, the GTFS feed requires constant polling (which in turn, requires maintaining a database of all bus lines and their stop IDs), and the rail API only indicates train arrival times.

So, that leaves us with the presumably written service alerts, whose most reliable machine-readable source of those is from the email notification system.

Implementation

With an automated system to receive the email alerts, we can parse them and store them in a database, using the magic of GPT/LLMs to process human-readable text into structured data, and to generate the < 300 character messages posted on both a central Bluesky account (alerts.marta.fail), and accounts for each bus line and rail line.

Next Steps

There are a lot of moving parts to this project: from the email alert system to the database to the Bluesky accounts to the website to the API. The most immediate goal is to ensure the resilience of the system, and to make sure that the alerts are posted to Bluesky in a timely manner.

Then, with the structured data in hand, the structured data can be utilized for bulk downloads or API-based access.