PM-201a · Module 2

Positive vs. Negative Constraints

3 min read

Constraint engineering uses two instruments: positive constraints (what to do) and negative constraints (what not to do). Both are necessary. Positive constraints define the scope and direction of the output. Negative constraints prevent the model from doing things that seem helpful but fall outside the required deliverable. Relying on positive constraints alone assumes the model knows what to exclude. It does not. State the exclusions explicitly.

Do This

  • Write both positive and negative constraints for every production prompt
  • Make negative constraints specific: "Do not recommend specific vendors" not "keep it general"
  • Derive negative constraints from failure modes you have observed in previous outputs
  • List constraints as numbered items — not buried in prose — so they are not missed

Avoid This

  • Do not assume positive constraints make negative constraints unnecessary
  • Do not write vague negative constraints: "Do not be too technical" tells the model nothing measurable
  • Do not write negative constraints that contradict positive ones — resolve conflicts explicitly
  • Do not skip negative constraints because you think the model will figure it out — it will not
[POSITIVE CONSTRAINTS — what to do]
1. Summarize the three primary financial risks identified in the report
2. Use a numbered list with one sentence per risk
3. Write for an audience of non-financial senior managers
4. Include the section reference where each risk is discussed

[NEGATIVE CONSTRAINTS — what not to do]
1. Do not recommend mitigation strategies — summarize only
2. Do not include risks not explicitly named in the report
3. Do not use financial jargon without plain-English definitions
4. Do not exceed 150 words total