Jobs and activityΒΆ

Follow durable jobs in the workspace and use activity records to return to work started through Studio, REST or MCP. Keep the job identity when moving between clients.

Queue messages explain why a run has not started. During execution, inspect stage, elapsed time and available progress or convergence data. Waiting, running, awaiting review and preparing a viewer are different states.

A browser disconnection does not cancel work. Request cancellation explicitly and wait for the confirmed outcome. A failed or unconverged run should retain diagnostics; review those before deciding whether to submit a changed request.

Open saved viewers and histories from finished jobs. If view preparation fails, recover that operation separately from the source compute. Use job recovery for exact states and convergence to assess the numerical result.