Canvas

The Canvas is the default view of a Discovery Run's Experiments page: a map that grows from your Objective into the Experiments the agent tries and the Versions they produce. Zoom, fit the map to the view, and show or hide the legend from the controls on the canvas. The same Experiments are also available as a Kanban board, a List, and an Archive of dismissed ones. See Experiments.

Nodes

• Objective: what the run is trying to improve. It anchors the map.
• Experiment (EXP-…): an approach or hypothesis, with a confidence score from the reviewing models. Confidence is a model judgement, not a statistical measure.
• Version (v1, v2, …): a code variant an Experiment produced, compared with the Baseline. The run's Best version is marked.

Click a node to open its details: the hypothesis, the Versions it produced and, once it has finished, its conclusion.

Reviewing and steering drafts

The agent proposes draft Experiments, and the recommended ones are highlighted. For each draft you can Queue it so the agent works on it, or Dismiss it. Add your own with New experiment.

In Manual approval mode the run pauses until you queue at least one draft and click Resume. In Automatic mode the agent queues and runs Experiments itself.

From an Experiment's actions you can also Edit it, move it Back to draft, Restore a dismissed one, or Generate similar to ask for variations on it. Select several Experiments and choose Combine to ask the agent for a new Experiment that combines them.