Judgment Criteria
Declares the criteria the model should use to evaluate evidence, weigh trade-offs, and handle uncertainty while generating.

Problem
Without explicit criteria, the model defaults to generic, socially agreeable, or stereotype-driven evaluation; user preferences leak in as implicit evidence.
Solution
Declare the evaluation policy - evidence standards, trade-offs, uncertainty handling, preference treatment - before generation. Explicit criteria shift what the model optimizes for while selecting among plausible continuations.
Example
One worked example. The book has 2–4 per technique.
Evaluate this resume for a Customer Success Manager role using these judgment criteria: 1. Evidence standard: Treat only resume facts as evidence. Treat impressions and preferences as non-evidence. 2. Trade-offs: Optimize for risk reduction over speed. 3. Uncertainty: Separate Knowns, Unknowns, and Risks. 4. Recommendation rule: Recommend a next step only if the evidence meets the bar for the role; otherwise specify what must be validated. Resume: * 2 years in customer support at a SaaS company * Managed a small portfolio of onboarding accounts * Basic Excel and SQL Output: Knowns, Unknowns, Risks, Recommendation.
Knowns: 2 years of SaaS customer support experience; managed a small onboarding account portfolio; basic Excel and SQL. Unknowns: renewal ownership, escalation handling, success metrics delivered, stakeholder management depth, customer segment complexity. Risks: possible over-leveling if the role requires enterprise account ownership or quota-linked renewals; limited evidence of measurable customer outcomes. Recommendation: Do not make a hiring decision from this resume alone. Advance only to a structured interview that validates account ownership, stakeholder management, escalation handling, and measurable outcomes.
Techniques
Concrete ways to implement Judgment Criteria. Each technique fits a different situation.
- 01
Worldview Lens
Declare the evaluation framework (evidence, priorities, trade-offs, uncertainty handling) for direct control over how the model judges information.
- 02
Stance Cue
Declare how user preferences and beliefs should be weighted during evaluation to prevent agreement signals from masquerading as evidence.
