Model card template — Google MLSE pattern + EU AI Act Article 13 mapping

# Model Card — {{MODEL_NAME}}
**Version:** {{VERSION}} · **Date:** {{DATE}} · **Provider:** {{COMPANY}}

## 1. Model details
- **Developed by:** {{COMPANY}}
- **Model type:** {{MODEL_TYPE}} (e.g., transformer, CNN, RNN, GBT)
- **Architecture family:** {{ARCHITECTURE_FAMILY}}
- **Specific architecture:** {{SPECIFIC_ARCHITECTURE}} (e.g., BERT-base, ResNet50, LightGBM)
- **Parameters:** {{PARAMETER_COUNT}}
- **License:** {{LICENSE}}
- **Released:** {{RELEASE_DATE}}
- **Citation:** {{CITATION}} (if academic)

## 2. Intended use
- **Primary use case:** {{PRIMARY_USE}}
- **Primary intended users:** {{INTENDED_USERS}}
- **Out-of-scope use cases:** {{OUT_OF_SCOPE}}

## 3. Factors
- **Relevant factors:** {{FACTORS}} (e.g., demographic groups, geographic region, time period)
- **Evaluation factors:** {{EVAL_FACTORS}}

## 4. Metrics
- **Model performance metrics:** {{METRICS}} (accuracy, F1, AUC, BLEU, etc.)
- **Decision thresholds:** {{THRESHOLD_AND_WHY}}
- **Variation approaches:** {{HOW_VARIATION_HANDLED}}

## 5. Evaluation data
- **Datasets used:** {{EVAL_DATASETS}}
- **Motivation for evaluation choices:** {{EVAL_MOTIVATION}}
- **Preprocessing:** {{EVAL_PREPROCESSING}}

## 6. Training data
- **Datasets used:** {{TRAINING_DATASETS}}
- **Data collection period:** {{COLLECTION_START}} to {{COLLECTION_END}}
- **Source / curation:** {{SOURCE_AND_CURATION}}
- **Data card link:** {{DATA_CARD_URL}}

## 7. Quantitative analyses
| Slice | Accuracy | Demographic parity | Robustness |
|---|---|---|---|
| Overall | {{ACCURACY}} | {{DP}} | {{ROBUSTNESS}} |
| {{SLICE_1}} | {{ACCURACY_1}} | {{DP_1}} | {{ROB_1}} |
| {{SLICE_2}} | {{ACCURACY_2}} | {{DP_2}} | {{ROB_2}} |

## 8. Ethical considerations
- **Sensitive use cases:** {{SENSITIVE_USE_CASES}}
- **Risks:** {{RISKS}}
- **Mitigations applied:** {{MITIGATIONS}}
- **Residual risks:** {{RESIDUAL_RISKS}}

## 9. Caveats + recommendations
- {{CAVEAT_1}}
- {{CAVEAT_2}}
- {{RECOMMENDATION_1}}

## 10. EU AI Act Article 13 mapping
Article 13 requires that high-risk AI systems are "designed and developed in such a way to ensure that their operation is sufficiently transparent" — the model card is a primary transparency artifact.

| Article 13 requirement | Where in this card |
|---|---|
| Identity + contact of provider | §1 (Provider) |
| System characteristics, capabilities, performance | §2-7 |
| Foreseeable misuse + risks | §8 (ethical considerations) + §9 (caveats) |
| Performance + accuracy metrics | §7 |
| Specifications for input data | §6 (training data) |
| Computational + hardware resources | (in CycloneDX AIBOM environmentalConsiderations) |
| Instructions for use | (separate user guide; reference here) |

## 11. Annex III high-risk classification
- **Annex III category (or out-of-scope):** {{ANNEX_III_CATEGORY}}
- **Classification rationale:** {{RATIONALE}}
- **Conformity assessment route:** Annex {{VI_OR_VII}}
- **CE marking affixed:** {{TRUE_OR_FALSE}}
- **EU AI Database registration:** {{REGISTRATION_ID_OR_NA}}

## 12. Update history
| Date | Version | Changes |
|---|---|---|
| {{DATE_1}} | {{V_1}} | {{CHANGES_1}} |
| {{DATE_2}} | {{V_2}} | {{CHANGES_2}} |