| name | data-dyna-experiment-planner |
| description | Draft independent-café intervention hypotheses and experiment plans from Data Dyna AgentContextBundle facts, then route them through deterministic validation and merchant review. Use when generating Data Dyna experiment drafts from opportunity gaps. |
Data Dyna Experiment Planner
Use this skill only after Data Core has produced deterministic context: store profile, peer benchmark, opportunity gap, and evidence refs.
Boundary
- Generate hypotheses, not facts.
- Draft experiment plans, not direct business actions.
- Treat peer benchmarks as directional and non-causal.
- Keep every output as
agent_draft_not_core_truth until deterministic validation and merchant confirmation.
Required output shape
Every draft must include:
evidenceRefs copied from deterministic context.
uncertainty with confidence and assumptions.
- Merchant confirmation requirement.
- Guardrail metrics and stop criteria.
- Rollback support.
Forbidden
Do not directly apply menu, price, coupon, or customer-message changes.
Do not create Core facts, metric facts, benchmark facts, evidence facts, orders, or business configs.
Do not bypass the deterministic validator or merchant review.
Workflow
- Read
AgentContextBundle facts.
- Draft one intervention hypothesis.
- Draft one experiment plan.
- Validate with the deterministic validator.
- Submit only validator-acceptable drafts for merchant review.