Skip to main content
COMPEL Body of Knowledge
Get Certified Platform

COMPEL Glossary / evaluation-harness

Evaluation harness

The infrastructure that runs capability, regression, safety, and human-review evaluations on an LLM feature on a defined cadence.

What this means in practice

Treated as a governance artefact, not just an engineering convenience — its coverage and cadence directly determine whether the organisation can detect and act on drift or misuse.

Synonyms

LLM evaluation suite , eval harness , capability-and-safety evaluation

Related Terms Network

This diagram shows how Evaluation harness connects to other concepts in the COMPEL body of knowledge. Each linked node is a term you can explore in turn.

Related terms for Evaluation harness Network graph showing 4 related glossary terms connected to Evaluation harness. Evaluation har… Red-team (… Confabulat… Content sa… Model and …
4 related concepts connected to Evaluation harness.

The COMPEL Glossary Graph visualizes relationships between framework terminology, showing how concepts interconnect across domains, stages, and pillars. Term nodes cluster by pillar affiliation while cross-references reveal semantic dependencies — for example, how risk appetite connects to control effectiveness, model governance, and assurance requirements. This network representation helps practitioners navigate the framework vocabulary and understand that COMPEL terminology forms a coherent conceptual system rather than isolated definitions.

Accessible list of related terms
  • Red-team (for LLMs) — A structured adversarial exercise against an LLM feature using human, automated, or hybrid techniques drawn from MITRE ATLAS or OWASP LLM Top 10 to discover failure modes before attackers do.
  • Confabulation — NIST's preferred term for hallucination: an LLM generating fluent output that is unsupported by ground truth.
  • Content safety classifier — A model or rule system that detects policy-violating output categories — violence, self-harm, CSAM, targeted harassment, dangerous instructions, and similar.
  • Model and prompt registry — A versioned inventory of models, system prompts, retrieval sources, and guardrails deployed in production.

See also

  • Red-team (for LLMs) — A structured adversarial exercise against an LLM feature using human, automated, or hybrid techniques drawn from MITRE ATLAS or OWASP LLM Top 10 to discover failure modes before attackers do.
  • Confabulation — NIST's preferred term for hallucination: an LLM generating fluent output that is unsupported by ground truth.
  • Content safety classifier — A model or rule system that detects policy-violating output categories — violence, self-harm, CSAM, targeted harassment, dangerous instructions, and similar.
  • Model and prompt registry — A versioned inventory of models, system prompts, retrieval sources, and guardrails deployed in production.

Related articles in the Body of Knowledge