SETTING GLOBAL STANDARDS FOR TRUSTED AI CREDENTIALSAI Competence Framework v1.29 · current release
AI Certification Standards
Apply for conformance
You are reading the current version of the framework, v1.29.Permanent address for this version
D6.L3.02ACTIVESTABLED6 Evaluation and assurance · L3 Proficient

Design an evaluation set representative of production conditions and resistant to overfitting.

Type Skill · introduced in version 0.1

Performance indicators

Normative. These state what would be observed in a person who holds the statement.

Samples from realistic distributions rather than curated examples
Holds back cases not used during development
Identifies where the set is unrepresentative and states the consequence
Evidence examples

Non-normative. Illustrative of evidence an awarding body might accept; not a required form.

A worked artefact produced in the course of normal duties, with the reasoning recorded at the time
Attestation by a competent supervisor against the indicators above, not against a general impression
Relationships

Assumes

Statements a candidate is taken to hold already. Never at a higher level than this one.

D6.L2.02

Produce a set of test cases covering expected, edge and failure conditions for a defined task.

Assumed by

Derived inverse. Statements that take this one as given.

X-AGT.L3.03

Evaluate an agentic system across multi-step trajectories, not single outputs.

X-RAG.L3.02

Evaluate retrieval quality separately from answer quality.

Related

Cross-domain relationships, stated in both directions and typed in the content model.

D1.L3.02

Compare candidate models against the stated requirements of a defined task.

D2.L3.02

Assess whether a source set is fit for the questions it will be asked.

Referenced by

Entries on the register of conformance claims whose coverage map cites this statement.

No entry on the register cites this statement. This block is populated from the coverage maps of register entries as they are listed.

Provenance
IntroducedVersion 0.1
Last modifiedNot modified since introduction
Statusactive · stable
Version displayedv1.29
Permanent URLaicertificationstandards.org/framework/statements/D6.L3.02
Cite this statement

AI Certification Standards (2026) AICF D6.L3.02, D6 Evaluation and assurance, version 1.29. Available at aicertificationstandards.org/framework/v1.29/statements/D6.L3.02 (accessed date).

Propose an amendment

Statements change through the published process, not by correspondence

An amendment to this statement, its indicators or its relationships is proposed through the contribution process. Every submission is answered on the record, and the reasoning for acceptance or rejection is published in the release record for the version that follows.

Propose an amendment to D6.L3.02

Writing rules and the controlled verb list that govern how this statement is worded are published in the methodology.

This page displays version 1.29 · last reviewed 30.08.2026