Paramus — Experiment Campaign Planner

Overview

Turn a research goal into a runnable closed-loop campaign: choose the design strategy, define the DMTA loop, and set stop criteria. The value is the campaign-level strategy, not any single design call.

Guidance vs execution: designs and optimization come from tested Paramus tools; execution dispatches to OPERATE / lab robots. Do not hand-generate run matrices in ad-hoc code.

When to Use

• Setting up an autonomous / closed-loop experimental campaign
• "How should I structure experiments to reach this goal within budget?" Do not use for: one-off analysis of an existing dataset (use the analysis workflow).

Workflow

Goal + factors + ranges + budget + constraints
  ↓ 1. Formalize objective & constraints  → measurable target + limits
  ↓ 2. Select design/strategy             → screening → optimization → active learning
  ↓ 3. Generate initial design            → Paramus DOE tool
  ↓ 4. Define DMTA loop                    → make → test → analyze → decide
  ↓ 5. Set stop criteria                   → convergence / budget / confidence
  ↓ 6. Emit campaign plan                  → runnable plan + provenance

Procedure

1. Discover tools via search/get_schema ("design of experiments", "optimization", "campaign").
2. Formalize the objective into a measurable target with explicit constraints.
3. Choose the strategy (screening vs RSM vs active learning) and justify it.
4. Generate the initial design; echo factor coding and randomization.
5. Define the DMTA loop and how results feed the next iteration.
6. Set stop criteria and emit the campaign plan with provenance.

Domain Judgment

• Match strategy to budget: don't propose RSM when only a screening budget exists.
• Always define stop criteria up front — an open-ended loop is a runaway cost.
• Separate simulate-in-the-loop steps from wet-lab steps; state which is which.

Fallbacks

• Planning/DOE tool unavailable → deliver a manual design + loop description, clearly scoped.
• Endpoint unreachable → stop and report; do not fabricate a run matrix in local code.