DR-301b · Module 1

Constraint Engineering

3 min read

Constraints are the most undervalued component of research prompts. An unconstrained prompt produces plausible-sounding output that may or may not match your actual requirements. A precisely constrained prompt produces output that meets a specific standard every time. The constraint set has four categories: scope constraints (what is in and out of bounds), quality constraints (what standards the output must meet), format constraints (how the output must be structured), and epistemic constraints (how the model should handle uncertainty).

Do This

  • Specify explicit scope boundaries — geography, time period, company size, industry vertical
  • Define quality thresholds — "every claim must cite a source" or "confidence level on each finding"
  • Set epistemic constraints — "distinguish assessed facts from inferences" or "flag data gaps explicitly"
  • Include negative constraints — what to exclude is as important as what to include

Avoid This

  • Leave scope implicit and hope the model guesses your boundaries correctly
  • Accept output without quality requirements — you get what you measure
  • Allow the model to present speculation and fact with equal confidence
  • Constrain so heavily that the model cannot produce useful analysis — constraints guide, they do not strangle