sofer/constrain

Constrain

Purpose

Make implicit judgement explicit before work is handed to an agent, person, or automated process.

Process

1. Clarify the task, executor, audience, and success criteria.
2. Surface failure modes:
   • If someone optimised only for the stated goal, what collateral damage could they cause?
   • If an agent followed the request literally, what reasonable bad outcome could result?
   • What context does the user have that the executor does not?
   • Where might two good priorities conflict?
3. Convert the answers into constraints.
4. Validate the constraints with the user before execution when the work is high-risk.

Output

## Must
[Hard requirements. If these fail, the work fails.]

## Must not
[Specific prohibitions, each tied to a plausible failure mode.]

## Prefer
[Decision rules for tradeoffs.]

## Escalate when
[Specific stop-and-ask conditions.]

Rules

• Every must not should prevent a real plausible failure mode.
• Prefer items are tradeoff rules, not aspirations.
• Escalation triggers must be observable. Avoid vague triggers like "when unsure".
• Keep the constraint set as small as possible while covering the important risks.