HeartSteps Micro-Randomized Trial

Source

• Dataset metadata snapshot: source.md
• Metadata JSON: metadata.json
• Canonical external source: https://github.com/klasnja/HeartStepsV1

Action-Time-Series Notes

• Dataset type: mobile-health longitudinal intervention data.
• Temporal structure: participant-day decision points and physical-activity outcomes over a multi-week study.
• Actions or interventions: micro-randomized activity suggestions versus no suggestion and related intervention decisions.
• Suitability: strong for randomized behavioral intervention dynamics; small and domain-specific.

Foundation TSFM Relevance

Agenda slot	Verdict	Evidence	Missing pieces
Control and counterfactuals	partially closes	Micro-randomized activity suggestions create randomized intervention decisions with longitudinal outcomes.	Domain is small mobile-health behavior, not broad operational telemetry or high-dimensional multivariate systems.
Benchmarks: causal/counterfactual reasoning	adjacent	Provides an action-conditioned time-series setting where treatment timing matters.	Not designed as a foundation-model benchmark and lacks rich system/context metadata.

Links Into The Wiki

• Foundation Time-Series Model Research Agenda
• Action-Conditioned Time-Series Datasets
• World Models