Forge Intelligence

Physics experiment design v1

Purpose: Design a controlled discriminating experiment with calibration, uncertainty, stop, reproducibility, and safety controls.

Updated

Input

Required fields: question, candidate_models, constraints. Additional bounded context may be supplied according to contract.json.

Output

{ "experiment": <object> } where the payload validates against schemas/physicsgenesis/experiment-card.schema.json.

Substrate reuse

Compose Forge LCDL tasks: decompose_problem, llm_multi_label, llm_boolean_gate, plan_decision_pack. Use standard operators for guarded sequencing, fallback, and bounded repair.

Failure modes

  • no discriminating prediction
  • missing calibration route
  • post-hoc analysis masquerading as plan
  • unsafe apparatus detail

Verification

  • JSON Schema
  • hypothesis discrimination
  • confounder coverage
  • safety gate

The task never self-promotes, invents receipts, or upgrades empirical status without authenticated evidence.