Typical Workflow

1. Choose inputs

Browse the catalog and choose problem IDs aligned with the research question.

2. Instantiate core objects

Load problem instances and inspect constraints, state representation, prompt content, and available evaluator interfaces.

3. Execute or inspect

Generate candidate solutions, enumerate transitions, or call evaluators.

4. Capture artifacts

Record solution state, metrics, and any domain-specific outputs needed for comparison or reporting.

5. Compose the ecosystem seams

Let design-research-experiments resolve the packaged problem through design_research_problems.integration, invoke participants from design-research-agents, and preserve the problem metadata for design-research-analysis. The experiments layer owns cross-package orchestration; this package owns problem resolution and evaluation.

Dependency Caveats

The base install supports text and decision problems, generic grammar transitions, and analytic-surrogate battery paths. Install grammar for trussme-backed planar or 3D truss structural evaluation, battery for PyBaMM-backed battery-fidelity modes, and the relevant mcp or cad extra for external-tool workflows. Plan installation profiles per study protocol rather than installing every extra by default.