Definition
A trajectory turns a work step into a training example.
Instead of just recording that someone clicked, a trajectory ties that action to the surrounding state so a model can learn what context made the action correct.
Direct answer
What is state-action-state trajectory data?before, action, after
State-action-state trajectory data is a structured demonstration format for AI training and evaluation. It records the environment state before an action, the human action, and the resulting state after the action.
Built for
AI teams, data engineers, and automation teams evaluating agent training data.
Useful when you need to
Define state-action-state trajectory data for AI agent workflows.
MARFI
MARFI assembles trajectories from consented sessions.
Trajectory workspaces can align actions, app context, timing, system state, and optional keyframes into JSONL examples.
Use
These examples help train and evaluate task agents.
Agent teams can use state-action-state data to teach models how expert users navigate real software workflows.
Answer engine ready
Plain answers for buyers and crawlers.
These pages are written to be useful to humans first, while giving search and AI systems clear, citable product facts.
FAQ
Why include after state?
After state shows what changed because of the action, which helps models evaluate whether a step had the intended effect.
Is trajectory data always visual?
No. Visual keyframes are included only when screenshot capture is available and consented.