A scientific goal
Target or modality, desired outcome, constraints, existing evidence, thresholds, and human checkpoints.
FROGENT / workflow
Six stages keep the question attached to every decision.
The sequence is explicit. Feedback can move backward when evidence changes the design problem.
A campaign record
Evidence, candidates, scores, failure diagnostics, provenance, artifacts, and agent handoffs.
A reviewable package
Ranked candidates, evaluation records, synthesis plans, limitations, and the next validation decision.
Define
Record the scientific goal, modality, hard constraints, thresholds, prior evidence, and human decision points.
- In
- Scientific question and program constraints
- Out
- A task brief and an inspectable plan
Retrieve
Collect task-relevant literature, records, structures, and files with the source attached to each artifact.
- In
- Task brief and source requirements
- Out
- Evidence set with provenance
Design
Generate candidates through modality-specific engines under explicit structural and scientific constraints.
- In
- Target context and design objectives
- Out
- Candidate structures and design artifacts
Evaluate
Run task-specific evaluators, preserve component scores, and expose the diagnostics behind a ranking.
- In
- Candidates and a fixed evaluation protocol
- Out
- Scores, poses, properties, and diagnostics
Refine
Return evaluation evidence to shared context so the next proposal responds to the current record.
- In
- Ranked candidates and failure signals
- Out
- A revised candidate set with traceable changes
Deliver
Assemble ranked candidates, synthesis plans, provenance, limitations, and a consolidated computational report.
- In
- Campaign record and selected artifacts
- Out
- A decision-ready scientific package
Choose a task
Open the workflow at capability level.
Each capability page pairs a scientific question with its inputs, outputs, manuscript figure, interpretation, and evidence limit.