Model Cards for Model Reporting.¶
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., et al. (2019). Model Cards for Model Reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT, 220-229.
Cited by¶
16 citations across 16 artifacts.
Each citation links to the sentence it supports in the citing article.
Primes¶
- Applicability Scope
- Machine-learning model cards and datasheets declare the input-domain envelope, intended-use class, and known failure modes, flagging out-of-distribution application.
This sourceProposes model cards that declare a model's intended-use class, input-domain envelope, and known failure modes so consumers can flag out-of-distribution application.
Supported in partVerified against the publisher's abstract
The model-cards paper backs only the clause that documentation discloses intended-use context so models are kept out of unsuitable contexts; it says nothing of datasheets, inclusion/exclusion criteria, or off-label application.
“Model cards also disclose the context in which models are intended to be used, details of the performance evaluation procedures, and other relevant information.”
- Machine-learning model cards and datasheets declare the input-domain envelope, intended-use class, and known failure modes, flagging out-of-distribution application.
- Boundary Disclosure Card
- In machine learning, data cards and model cards — explicitly borrowing the nutrition-label idea — travel with datasets and models, carrying composition, intended use, and known limitations.
This sourceIntroduces the model-card schema carrying intended use, known limitations, and demographic-breakdown evaluation, explicitly motivated by disclosure analogues.
Supported in partVerified against the publisher's abstract
Mitchell et al. introduce model cards, short documents that accompany trained models and disclose intended use, evaluation context and demographic breakdowns; the source says nothing about data cards or a nutrition-label borrowing.
“Model cards are short documents accompanying trained machine learning models that provide benchmarked evaluation in a variety of conditions, such as across different cultural, demographic, or phenotypic groups (e.g., race, geographic location, sex, Fitzpatrick skin type [15]) and intersectional groups (e.g., age and race, or sex and Fitzpatrick skin type) that are relevant to the intended application domains.”
- In machine learning, data cards and model cards — explicitly borrowing the nutrition-label idea — travel with datasets and models, carrying composition, intended use, and known limitations.
- Memoing
- Machine learning and data science. Experiment-tracking logs, model cards, ablation rationale, and training-run notes carrying the analyst's in-flight reasoning alongside the trained model.
This sourceProposes model cards documenting a trained model's intended use, training context, and evaluation rationale alongside the artifact.
Supported in partVerified against the publisher's abstract
The paper proposes model cards: short documents shipped alongside trained models that record intended use contexts and evaluation procedures, which backs the 'alongside the trained model' half of the claim only.
“Model cards are short documents accompanying trained machine learning models that provide benchmarked evaluation in a variety of conditions, such as across different cultural, demographic, or phenotypic groups (e.g., race, geographic location, sex, Fitzpatrick skin type [15]) and intersectional groups (e.g., age and race, or sex and Fitzpatrick skin type) that are relevant to the intended application domains.”
- Machine learning and data science. Experiment-tracking logs, model cards, ablation rationale, and training-run notes carrying the analyst's in-flight reasoning alongside the trained model.
- Summary Substance Divergence
- The pattern recurs in judicial opinions (holding versus reasoning), legislation (bill title versus text), contracts (term sheet versus final agreement), marketing and consumer finance (claims versus fine print), university rankings (the headline number versus the methodology document), polling (topline versus crosstabs), corporate financial communication (earnings-call talking points versus disclosure filings), ESG reporting, AI model cards (intended-use section versus evaluation appendix), and encyclopedia entries (lede versus body).
This sourceEstablishes the model card's intended-use section versus its benchmarked-evaluation detail — a concrete instance of the summary-versus-substance two-surface structure in AI documentation.
Supported in partVerified against the publisher's abstract
The abstract says model cards carry both intended-use context and benchmarked evaluation detail; it never says those two surfaces diverge.
“Model cards also disclose the context in which models are intended to be used, details of the performance evaluation procedures, and other relevant information.”
- The pattern recurs in judicial opinions (holding versus reasoning), legislation (bill title versus text), contracts (term sheet versus final agreement), marketing and consumer finance (claims versus fine print), university rankings (the headline number versus the methodology document), polling (topline versus crosstabs), corporate financial communication (earnings-call talking points versus disclosure filings), ESG reporting, AI model cards (intended-use section versus evaluation appendix), and encyclopedia entries (lede versus body).
- Transferability Overclaim
- The prime's interventions apply directly: publish a model card that pre-registers the validated input range, re-evaluate on deployment data before relying on the benchmark (localize before exporting), and install an out-of-distribution monitor as a boundary instrument that fires when inputs leave the validated envelope.
This sourceProposes model cards documenting the validated operating conditions and intended use of an ML model.
Supported in partVerified against the publisher's abstract
Mitchell et al. support only the model-card element of the three-part intervention — documenting intended use and validated operating conditions — not pre-registered input ranges, deployment re-evaluation, or OOD monitors.
“Model cards also disclose the context in which models are intended to be used, details of the performance evaluation procedures, and other relevant information.”
- The prime's interventions apply directly: publish a model card that pre-registers the validated input range, re-evaluate on deployment data before relying on the benchmark (localize before exporting), and install an out-of-distribution monitor as a boundary instrument that fires when inputs leave the validated envelope.
Domain-specific¶
- Distributional Blind Spot
- The intervention family redirects from model improvement to coverage characterisation: document the training distribution explicitly in model cards, monitor the deployment distribution for divergence from the training distribution, produce distance-from-training-support scores alongside predictions, and gate or refuse predictions for inputs whose distance from training support exceeds a threshold
This sourceThe model-cards framework, which documents a model's intended use and performance-evaluation procedures (with accompanying Datasheets for its training and evaluation data).
Supported in partVerified against the work's full text
“Model cards also disclose the context in which models are intended to be used, details of the performance evaluation procedures, and other relevant information.”
- The intervention family redirects from model improvement to coverage characterisation: document the training distribution explicitly in model cards, monitor the deployment distribution for divergence from the training distribution, produce distance-from-training-support scores alongside predictions, and gate or refuse predictions for inputs whose distance from training support exceeds a threshold
- Positionality
- Model cards, introduced by Mitchell and colleagues in 2019, deploy the same mechanism on machine-learning artifacts
This sourceThe model-cards framework introduced by Mitchell and colleagues, which documents a model's evaluation across demographic groups and its intended-use context.
Supported in partVerified against the work's full text
“In this paper, we propose a framework that we call model cards, to encourage such transparent model reporting.”
- Model cards, introduced by Mitchell and colleagues in 2019, deploy the same mechanism on machine-learning artifacts
Mechanisms¶
- Distortion Model Card
- The format borrows directly from model cards in machine learning — the practice of shipping a model with its intended use, evaluation conditions, and known limitations rather than as a black box
This sourceProposes shipping models with documented intended uses, evaluation conditions and results, caveats, and limitations.
- The format borrows directly from model cards in machine learning — the practice of shipping a model with its intended use, evaluation conditions, and known limitations rather than as a black box
- Distributional-Assumption Card
- Its strength is that it converts a distribution from a private default into a public, inspectable commitment that survives a hand-off — the same move as a machine-learning model card, which reports a model's intended use
This sourceProposes model cards that disclose a model's intended uses, data and evaluation context, and limitations so downstream users can judge appropriate use.
- Its strength is that it converts a distribution from a private default into a public, inspectable commitment that survives a hand-off — the same move as a machine-learning model card, which reports a model's intended use
- Model Card Value Section
- It extends the model-card tradition of structured reporting from performance into values.
This sourceIntroduces model cards as structured, transparent documents for reporting trained-model performance.
- It extends the model-card tradition of structured reporting from performance into values.
- Model Limitations Card
- The card is the artifact that rides along and says, in plain terms, "valid here, uncertain there, do not use beyond this line" — so the output's limits are inseparable from the output.
This sourceIntroduces model cards as documents that accompany released models and disclose intended and out-of-scope uses, uncertainty, caveats, and recommendations.
- The card is the artifact that rides along and says, in plain terms, "valid here, uncertain there, do not use beyond this line" — so the output's limits are inseparable from the output.
- Model or Rule Card
- Its strength is making a method's assumptions, limits, and ownership legible before anyone relies on it — the practice popularized as model cards for transparent model reporting
This sourceIntroduces model cards for transparent reporting of assumptions, limits, ownership, and information relevant to method selection.
- Its strength is making a method's assumptions, limits, and ownership legible before anyone relies on it — the practice popularized as model cards for transparent model reporting
- Model Scope Review
- Its discipline is codified in artifacts like the Model Card, which pairs a model with its intended-use scope and disaggregated performance precisely so its claims cannot silently outrun its evidence.
This sourceIntroduces Model Cards pairing intended-use scope and limitations with disaggregated evaluation results to constrain claims to documented evidence.
- Its discipline is codified in artifacts like the Model Card, which pairs a model with its intended-use scope and disaggregated performance precisely so its claims cannot silently outrun its evidence.
- Model Specification
- This is the trap the Model Cards
This sourceIntroduces Model Cards to disclose intended uses, evaluation conditions, performance, and limitations so inappropriate deployment is easier to identify.
- This is the trap the Model Cards
- Scorecard Disclosure Template
- This is the transparency spirit of a model card
This sourceProposes model cards as standardized documentation that increases transparency about a model’s intended uses and performance characteristics.
- This is the transparency spirit of a model card
- Shortcut-Risk Model Card Section
- Its strength is persistence and portability: a known crutch, its evidence, and its residual risk stay attached to the model, so the next team inherits the caveats instead of relearning them the hard way — and stating the intended-use boundary forces someone to decide where the model is not validated.
This sourceA model card travels with a released model, informs downstream users about limitations and suitability, and records intended and out-of-scope uses.
- Its strength is persistence and portability: a known crutch, its evidence, and its residual risk stay attached to the model, so the next team inherits the caveats instead of relearning them the hard way — and stating the intended-use boundary forces someone to decide where the model is not validated.
Verification¶
Does it exist? Confirmed. This work's DOI resolves to a registered record, which fixes its identity. That is all it fixes.
Does it back the claim? Read against the text for 7 of 16 citations: 7 supported in part. Each verdict is shown under its citation below, with what in the work backs the sentence.
Was it audited? Yes. A second, independent pass read the citation against the article text and recorded a verdict.
Support is checked per citation rather than per work — the same source can be cited soundly in one article and wrongly in another. Per-citation recording began recently, so a citation with no recorded check is a gap in the record rather than evidence it went unchecked.
See how references were verified.
Links previously used in the corpus¶
Before the registry existed this work was also linked 1 other way.
Registry ID ref:9201eaeb1510 · see in the full table