Examples
Worked examples
- Is an instance
An algorithm card for XGBoost describing its inductive biases, suitability for tabular data, and known limitations on high-dimensional sparse data.
- Is an instance
An algorithm card for k-means describing the convexity assumption.
Counter-examples
Looks similar, but isn't
- Not an instance
A model card for a specific trained XGBoost classifier.
- Not an instance
A textbook chapter on the algorithm.
Editorial commentary
An algorithm card documents an algorithmic method or method family — a particular gradient-boosting estimator, a clustering algorithm, a specific attention mechanism — independent of any single trained instance of it. It covers inductive biases, underlying assumptions, computational complexity, known failure modes, and the classes of problem the method is and is not suited to.
How this differs from a model card
The distinction is instance versus class. A model card describes one specific trained model — a particular set of learned weights, trained on particular data, with measured performance on particular benchmarks. An algorithm card describes the method before any training happens at all: it would be the same document whether the method was ultimately trained on one dataset or a thousand. A research team choosing between two candidate methods for a task reads algorithm cards; a team evaluating whether a specific already-trained model is fit for their use reads model cards.
Relationship to use cards and system cards
A use card documents a specific deployment of a specific trained model; a system card documents the full system a deployment sits inside. Algorithm cards sit at the opposite end of that spectrum — the most abstract, least deployment-specific layer of the documentation family.
Why it matters
Algorithm cards are most useful during method selection, well before a specific model exists to document: a methods section citing “we selected algorithm X because of property Y, per its algorithm card” is a stronger, more falsifiable claim than an undocumented choice.
References
- See also: Model card, Use card, AI evaluation card.
Also known as
method card
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Algorithm card"
vocab-term-identifier="https://casrai.org/dictionary/term/algorithm-card" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/algorithm-card",
"name": "Algorithm card",
"identifier": "https://casrai.org/dictionary/term/algorithm-card",
"description": "A documentation artefact describing the algorithmic method or family (e.g., a particular gradient-boosting estimator, a clustering algorithm) independent of any particular trained instance, including inductive biases, assumptions, complexity, and intended use cases.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/ai-ml-research-outputs#set",
"url": "https://casrai.org/dictionary/term/algorithm-card",
"alternateName": [
"method card"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T02:22:50",
"dateModified": "2026-09-05T14:24:32",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
