Cross-validation design worksheet ================================= StatMedAnalysis | Original downloadable worksheet | 6 November 2026 Choose resampling that matches the intended generalization question. How to use this worksheet State whether predictions concern new people, new sites or future periods. Select grouping or temporal folds accordingly. Keep hyperparameter selection separate from final performance estimation, using nested resampling where appropriate. Record all preprocessing inside each training fold. Save a working copy and keep the original blank version. Use public, hypothetical or aggregate information only; do not enter patient identifiers, credentials or confidential records. Keep source evidence in your approved project workspace. Mark unresolved text fields UNKNOWN or NOT REPORTED with a reason; never substitute zero for an unknown count. Project/version __________________ Prepared by role/date __________________ Working fields 1. Prediction target population Response/evidence: _________________________________________________________ ___________________________________________________________________________ 2. Resampling unit and fold rule Response/evidence: _________________________________________________________ ___________________________________________________________________________ 3. Number of folds and repeats Response/evidence: _________________________________________________________ ___________________________________________________________________________ 4. Tuning procedure Response/evidence: _________________________________________________________ ___________________________________________________________________________ 5. Outer evaluation procedure Response/evidence: _________________________________________________________ ___________________________________________________________________________ 6. Preprocessing within folds Response/evidence: _________________________________________________________ ___________________________________________________________________________ 7. Metric aggregation and uncertainty Response/evidence: _________________________________________________________ ___________________________________________________________________________ 8. Reproducible split identifiers Response/evidence: _________________________________________________________ ___________________________________________________________________________ Hypothetical worked case A fictional study tunes a model on all folds and reports the best cross-validation score from that same search. A separate outer evaluation is needed to assess the selection process rather than only the chosen setting. Decision rule and limitations Do not call a tuned score independent validation when the same data selected the settings. Explain dependence between fold estimates and avoid treating them as independent study replicates. Decision _______________________ Remaining uncertainty ____________________ Next action ____________________ Owner / review date ______________________ Independent check and discrepancy resolution _______________________________ Official method or reporting reference checked 6 November 2026 https://scikit-learn.org/stable/model_selection.html This is an original planning aid, not an official checklist, validated instrument, clinical recommendation or certificate of expert approval. Apply the referenced guidance to the specific design and record any additional requirements.