The AI Taxonomy.
Terms  /  Evaluation  /  Policy-Bound Evaluation
08 · Evaluation

Policy-Bound Evaluation

Evaluation scored against explicit plans, tasks, evidence, and policies.

Policy-Bound Evaluation is the more specific form of Structured Judgment, emphasizing that evaluation criteria should come from existing policies and standards — not ad hoc assessments. The team's coding standards, security policies, documentation requirements, and testing protocols become the evaluation rubric. This ensures consistency: the same code is evaluated the same way regardless of who (or what) reviews it. It also ensures alignment: the evaluation measures what the organization actually cares about.

Example
A security-focused team defines their evaluation policy: all agent-generated code must (1) sanitize user input, (2) use parameterized queries, (3) implement proper authentication checks, (4) log security-relevant events, (5) follow the principle of least privilege for data access. Every code review — human or AI — evaluates against these five policy requirements. An agent's code might be elegant and efficient but fail evaluation because it doesn't log authentication events. The policy is the standard, not subjective quality assessment.

More in Evaluation

Structured Judgment

Evaluation where an AI judge scores work against explicit plans, tasks, evidence, and policies rather than subjective impressions.

Eval Drift

Gradual misalignment between what an evaluation measures and what actually matters.

Judgment Bias

Systematic skew in AI evaluation due to unexamined assumptions, prompt framing, or training artifacts.

Rubric Rot

Decay in evaluation criteria relevance over time — the eval no longer tests what it should.

Eval Capture

When an agent optimizes for passing evaluation rather than doing the actual work.

Eval Integrity

Evaluation that remains aligned with real-world outcomes over time — resistant to Eval Drift, Rubric Rot, and Eval Capture.