Using the workspace

The gallery contains objects, not individual runs. Each object card shows a model, the latest run's status, and its best passing result. Drag in the viewport to rotate; click the object title or View iterations to open it.

An object page contains:

Comparisons use the selected run's independently evaluated baseline. Failed designs may have diagnostic metrics, but they cannot become the best design. A chart bar does not mean a failed design is validated. If the baseline itself fails, the UI does not calculate a percentage improvement against it.

Continue a design

Choose Continue from best or Continue from here. Edit the task, objective, target, constraints, iteration count, or budget. Starting creates a new linked run; it never overwrites its parent.

The seed is reevaluated under the new constraints. Compatible memories and tools remain available. A new task version isolates old tools and memory, and uses the new adapter's builder with the selected parameters. Incompatible parameters are rejected.

A failed design can seed another run if it produced a STEP artifact. A design that never built cannot. Select the previous run through its link or the run selector to compare histories without conflating different constraints.

Stop and resume

Stop run stops the loop and cancels an active CAD container. An already submitted model request may finish and incur cost; its response is archived before execution pauses. No subsequent iteration starts.

One run executes at a time. After restarting the server, interrupted runs are paused. Resume continues at saved checkpoints. A request whose outcome is uncertain is never blindly retried; continue from a saved design in a new run instead.

YAML downloads contain resolved run settings, including seed references, but no credentials. STEP downloads contain geometry. Git snapshots and request/evaluation records remain in the workspace for inspection.