1 · Characterize
State the candidate relationship, source evidence, detector, frame, uncertainty, and strongest ordinary explanation.
DESIGN-ONLY INTERVENTION SANDBOX · P10
Manipulation begins with the reasons not to proceed.
This surface organizes a candidate experiment before any physical action. It supplies gates, not procedures. A promising hypothesis does not lower ordinary laboratory, institutional, legal, environmental, or ethical requirements.
2 · Predict
Write distinct predicted outcomes for the candidate relation, conventional model, null world, and known measurement artifact.
3 · Bound
Declare energy, material, biological, environmental, privacy, and dual-use limits before describing an intervention.
4 · Reverse
Show how the proposed change can be stopped, contained, restored, and independently observed.
5 · Review
Require qualified domain, safety, ethics, and method review appropriate to the actual hazard—not to the framework’s confidence.
6 · Stop
Do not advance when evidence, competence, authorization, containment, reversibility, or measurement independence is missing.
AUTOMATIC STOP HOOK
Stop conditions outrank expected utility
- Unbounded or poorly characterized harm.
- No credible containment or reversal path.
- Controls cannot distinguish the candidate relation.
- Required expertise, consent, oversight, or authority is absent.
- Results cannot be independently recorded or challenged.