Context
People or agents need autonomy, while the system must constrain material risk, cost, disclosure, or irreversible effects.
Problem
Vague control produces two bad outcomes: cautious actors stop for permission, while aggressive actors discover limits only after crossing them. More restriction then reduces legitimate agency without improving clarity.
Forces
- Flexibility helps adapt to context.
- Predictability enables independent action.
- Controls can become obsolete.
- Enforcement without explanation damages trust.
Pattern
Express boundaries as observable rules around a deliberately useful operating space. State what is allowed, prohibited, conditional, and escalated. Enforce close to execution, explain decisions, and provide a governed path for changing the boundary.
Apply it
Begin with the outcome and material harms. Grant the widest reversible scope consistent with those harms. Provide safe defaults, budgets, rate limits, scoped tools, and explicit stop conditions. Review repeated escalation as a possible boundary-design defect.
Failure modes
- The boundary names prohibited outcomes but has no enforcement.
- Controls are so broad that routine work always escalates.
- Exceptions are granted privately and cannot be learned from.
- Actors optimize to the written rule while violating its intent.
Evidence needed
Outcomes showing increased safe throughput and reduced ad hoc permission requests without higher loss, disclosure, cost, or rollback rates.