Studies, cases, results and comparison

Discover and prepare

Open the study manager or ask the AI to list definitions, equipment and cases. Discover actual settings and qualification before submission; Study reference lists the registered kinds. Readiness diagnoses whether inputs are complete without inventing missing values. Named cases retain study configuration separately from the network version; edits create a new case version.

Submit exact inputs

Select the immutable network version, exact case/version, study kind and bounded settings. Submit a single job or explicit batch. Jobs run outside the GUI/backend process. Monitor persisted status; refresh after restart. Cancel requests cancellation, not evidence of clean shutdown. A failed/unsupported/nonconvergent result stays explicit. Do not replace it with a zero quantity.

Available study families

Power flow screens balanced AC operating conditions. Scenarios and time snapshots select declared operating points. Selected outages test explicit branch contingencies; source-less islands can be non-passing outcomes. Dispatch/optimization and corrective AC are admitted formulations, not certification of all optimization controls. Short circuit and protection have source/manufacturer and profile limitations. Loadability and P–V continuation depend on declared stress/control assumptions. Sensitivity is bounded nonlinear AC load perturbation, not a universal analytic derivative. RAW33 exchange validates its admitted format. See each generated kind reference and tool schema for exact input fields.

Inspect and compare results

Open the retained run, verify input/model/configuration/solver hashes, timestamps, convergence, warnings and applicability. Results tables and diagrams must bind the same exact run. Units, tolerances and control/slack assumptions must match before comparing. Preserve disagreement; never average it away. Cross-version runs can be historical or incompatible, not current acceptance. Filter/page large result sets rather than loading unbounded profiles.

Export and acceptance

Export only to an explicitly chosen destination. Handover import requires consent and does not transfer approvals. Solver receipts and independent checks support a human disposition; neither AI text nor a green job status is engineering evidence by itself. See Evidence.

Start without an AI chat

Select Studies on the left workspace rail. You do not need to send a prompt or pay for inference. With no selected candidate, a visible Start a study window explains the prerequisite instead of reporting a hidden chat error. Prepare study from current network explicitly retains a detached copy; opening Studies alone creates nothing. An empty network must be opened first. During active AI work, stop or finish that task before preparing a mutable candidate study.

The Study Manager is bound to a specific candidate and backend/account. Opening Studies again reuses that bound window instead of accumulating duplicate dialogs. Switching to a different candidate requires treating any unsaved setup first. Closing a manager stops its polling and invalidates late callbacks; cancellation of an actual running study job is a separate explicit action.

A complete case lifecycle

Select the study kind and review its definition, admitted settings and required equipment. Configure exact inputs; readiness can identify missing data but cannot repair it. Save setup with a meaningful label. Cases are immutable versions of configuration and source binding, not a second electrical model. Select a saved case and Run case to use that version. New setup starts editable configuration; editing an existing case creates another version rather than changing prior evidence. Archive retains history, while duplicate creates a separate case identity.

Runs lists retained result receipts; opening one is inspection, not a fresh solve. Compare runs only when quantities, units, methods and source profiles are compatible. Replay exact run creates a new execution using retained inputs, not the old result. Queue lists persisted jobs; a cancel request is not proof the worker has already stopped. Refresh and earlier-run paging recover retained state without resubmission. Batch run submits the explicitly selected cases, not every visible row.

Working with results and failures

Verify convergence and warnings first, then inspect exact tables, network overlays, tolerances and input/configuration/solver hashes. A stale result can be valid historical evidence while not describing the current editor. A disconnected island or unsupported profile remains an explicit non-passing outcome, not a zero voltage or zero loss. If a request fails, retain the case and inspect diagnostics before an explicit retry; never assume a retry is identical after inputs changed. Export/import handover requires actual path consent and does not transfer approval. See Study reference, AI tools, Evidence and the populated example.