Typical Workflow#

Use the control topology to decide who owns each step and the runtime flow to trace the artifacts that connect those steps.

Select Problems inputs. Choose packaged tasks, prompts, grammars, or evaluators through design_research.problems.
Select Agents inputs. Choose participants or workflows through design_research.agents. Problems and Agents are peer inputs to the study.
Define and run the study. Use design_research.experiments for hypotheses, conditions, replication, execution control, and artifact export.
Analyze the exported artifacts. Use design_research.analysis to validate and transform the event tables, then compute or visualize results.
Refine the protocol. Feed the evidence into the next Experiments study definition without changing ownership of Problems, Agents, or Analysis.

The runtime handoff is therefore:

Problems + Agents -> Experiments artifact set -> Analysis -> protocol refinement

Start with Canonical Artifact Flow for the smallest deterministic all-package handoff. Continue to Prompt Strategy Comparison Walkthrough only when a live model-backed Agents workflow is useful. Use the owning component documentation for package-specific APIs and optional dependencies.