Studies and predictions¶
The central API coordinates parameter studies and PhysicsAI requests with the same ownership, artifacts and usage attribution as other engineering operations.
Parameter studies¶
Use plan_geometry_study to validate explicit template variants and obtain a plan. Planning does not launch the variants. submit_study submits selected geometry, mesh or simulation cases with a combined estimate, budget check and concurrency cap. Discover its current exact schema before submission.
Keep a stable identity for each child request and record which geometry, mesh, physics and output definitions changed. Use consistent convergence and acceptance criteria when comparing designs. A rejected or unqualified candidate is not evidence that it performs worse physically.
Continuous and discrete parameters have different semantics. Keep integer counts integral, use published choice values, respect constraints and revalidate each topology. Do not interpolate a discrete design choice as though it were a length.
PhysicsAI¶
list_models returns model releases and their availability. Use the selected immutable release and its supported input schema with validate_prediction, then predict_physics / POST /v1/predictions. A disabled or incompatible model cannot be replaced silently by a different one.
A model prediction, reduced-order estimate and Fluxcore solution are different types of evidence. Preserve the label, model/release identity, applicability range and input design provenance. Use a simulation when the requested decision requires solver evidence or the design lies outside the prediction model’s supported range.
Cold-plate helpers¶
The design-source and brief tools retain cited requirements from documents. plan_cold_plate creates bounded candidate plans; prepare_cold_plate_simulation prepares thermal setup from actual mesh regions and the brief. Inspect the returned plan before submission. compare_cold_plate_candidates compares saved qualified summaries; it does not manufacture missing CFD evidence.