Examples
Worked examples
- Is an instance
An evaluation card for a code-completion benchmark documenting the held-out test set, prompting template, and decoding configuration.
- Is an instance
An evaluation card for a clinical-reasoning probe describing rater calibration.
Counter-examples
Looks similar, but isn't
- Not an instance
A model card section labelled 'Evaluation' but not separately documented.
- Not an instance
A leaderboard table without supporting methodology.
Editorial commentary
An AI evaluation card is a structured documentation artefact that describes an evaluation of an AI system — the evaluation methodology, the datasets used, the metrics reported, the results, and the known limitations of the evaluation itself — kept separate from the model card that documents the system being evaluated.
How this differs from the other AI-governance documentation and process terms
An evaluation card documents a specific act of testing: what was measured, how, against what benchmark, with what caveats. It is not the same artefact as an AI conformance assessment, which is a formal, often regulatory, determination that a system meets applicable legal or standards-based requirements before market placement — a conformance assessment may draw on evaluation-card-style evidence, but it is a compliance judgment, not a documentation format. It is also distinct from a model card (which documents the system’s design, training data, and intended use) and from a ISO/IEC 42001 AI management system (which governs an organisation’s overall AI processes, not a single evaluation).
Why evaluation itself needed its own documentation format
Evaluation cards are a relatively recent (2023 onward) addition to the documentation-artefact family, reflecting the recognition that an evaluation — a dataset, a protocol, an analysis approach — is itself a reusable artefact deserving its own documentation, separate from documenting the model being evaluated. NIST’s GenAI evaluation profile work and Stanford’s Center for Research on Foundation Models (CRFM) evaluation reports are commonly cited examples of the genre, each making explicit what a given benchmark result does and does not demonstrate.
Sources
NIST AI RMF Generative AI Profile; Stanford CRFM evaluation transparency reports.
Also known as
eval card
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="AI evaluation card"
vocab-term-identifier="https://casrai.org/dictionary/term/ai-evaluation-card" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/ai-evaluation-card",
"name": "AI evaluation card",
"identifier": "https://casrai.org/dictionary/term/ai-evaluation-card",
"description": "A structured documentation artefact specifically describing an evaluation of an AI system, separate from the model card, including the evaluation methodology, datasets, metrics, results, and known limitations of the evaluation itself.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/ai-ml-research-outputs#set",
"url": "https://casrai.org/dictionary/term/ai-evaluation-card",
"alternateName": [
"eval card"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T02:22:51",
"dateModified": "2026-09-05T14:24:50",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
