Same fitness, eight real IRONMAN 70.3 courses. The model sees each course's climbing, so the hilly ones cost more on the bike. Tap one to load it above.
Before trusting it with anyone else's race, I made it predict mine. Every past race leg was predicted using only the training logged before that day, then compared with what actually happened. Filled dots landed inside the 80% range.
Strava's activities.csv or a Garmin CSV. Every swim, ride and run you've logged.
Reads every activity title and picks out the races, so race-day effort is labelled without you tagging years of log by hand.
Distance, climbing per km, how long ago, the last 7 and 42 days of training load, and whether it was a race.
An open tabular foundation model learns from your sessions in seconds and returns a full spread of likely speeds, not one number.
Each leg is drawn from its spread, transitions added, and the share under your goal is your chance.
Short and plain, from the computed numbers only. Any number it makes up gets the note thrown out.
Heart rate, GPS and every session you've logged are health and location data. Race Card reads your Strava or Garmin export and never sends it anywhere.
TabPFN v2 is an open tabular foundation model that runs on a laptop CPU in seconds. Gemma 4 writes the note through Ollama, offline.
pip install git+https://github.com/iamrobertmoore/race-card
racecard your-strava-export.zip --name You --goal 5:00