KX-07 · Modeling
Bayesian Active Experiment Planner
Rank candidate experiments by expected information gain under real constraints.
Model-InferredPilot-Validation Required
Produces a next-best-experiment queue that maximises discrimination between surviving hypotheses per unit of cost, time, and facility access.
Institute module developed within the KRYOS v6 framework ecosystem. Established scientific methods retain their conventional attribution.
When to use
When the experiment backlog exceeds available beam time, budget, or personnel.
When not to use
When hypotheses are not yet falsifiable or priors cannot be defended.
Required inputs
- Ranked hypotheses
- Cost and time estimates
- Facility availability
Optional inputs
- Historic yield data
- Risk tolerance statement
Assumptions and constraints
- Priors are stated explicitly and their sensitivity is reported.
Process steps
- 01Elicit and document priors with rationale.
- 02Estimate expected information gain per candidate experiment.
- 03Apply cost, time, and access constraints.
- 04Produce a ranked queue with rationale per position.
- 05Report ranking sensitivity to prior choice.
Human review gates
- Sponsor and PI approve the queue before resource commitment.
Primary output — Next-best-experiment queue with information-gain rationale.
- queue_position, experiment_id, expected_information_gain
- cost, duration, facility_requirement
- prior_sensitivity, rationale
Failure signals
- Queue order flips under small, defensible prior changes.
Quarantine triggers
- Queue used to justify a claim rather than a plan.
Mini prompt
Rank candidate experiments by expected information gain given [CONSTRAINTS]. State priors and report ranking sensitivity to prior choice. Return a queue with rationale per position.