Tasks
Tasks
The current leaderboard uses three measured-data tasks with fixed person splits. Each model runs three agent trials per task, with two coding experiments per episode.
01 · Causal estimation
Motion-robust pulse estimation
Estimate trailing eight-second heart rate every two seconds from wrist PPG and acceleration. ECG-derived references remain provisional.
- Data
- GalaxyPPG · 23 people
- Budget
- 13 / 5 / 5 training / development / test people
TrainEvaluateSelect
02 · Forecasting
Core-temperature forecasting
Forecast measured rectal temperature at +5, +10 and +20 minutes from causal sensor history and demographics. Error is reported in °C.
- Data
- CORE cycling · 24 people
- Budget
- 14 / 5 / 5 training / development / test people
TrainEvaluateSelect
03 · Unseen-person prediction
Load-carriage metabolic cost
Predict gross metabolic power in watts for level walking at 4 km/h with 0, 5, 10 or 15 kg load. All 189 available measurements are preserved.
- Data
- EVOBREATH · 48 people
- Budget
- 28 / 10 / 10 training / development / test people
TrainEvaluateSelect