{"id":"W7115270098","doi":"","title":"Abstract 2: Trained on what? A case study on the misapplication of machine learning in women’s ice hockey","year":2025,"lang":"en","type":"article","venue":"Macedonian Journal of Medical Sciences (University of Skopje)","topic":"Sports Analytics and Performance","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Ice hockey; Transparency (behavior); Analytics; Athletes; Statistical analysis","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003710818,0.00006642921,0.0002677436,0.0003948884,0.0001539699,0.00002857851,0.0005001393,0.00005043756,0.0006658443],"category_scores_gemma":[0.0001113357,0.00005481002,0.00006785969,0.0005717073,0.000225915,0.0001876502,0.00003772169,0.0003286753,0.000004886152],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008251632,"about_ca_system_score_gemma":0.00009485612,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009967906,"about_ca_topic_score_gemma":0.0003460003,"domain_scores_codex":[0.999106,0.0000229572,0.000369197,0.0001439063,0.0002195975,0.0001383592],"domain_scores_gemma":[0.9990698,0.0001565219,0.0005544731,0.0001056004,0.00003159387,0.00008203423],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009107625,0.004183766,0.5798455,0.000232707,0.0004211677,0.002759888,0.08074806,0.02102138,0.00005495968,0.08439771,0.0008060729,0.2246181],"study_design_scores_gemma":[0.005315656,0.00327348,0.6572645,0.001120154,0.00003354334,0.0001372911,0.2306529,0.08483094,0.00003649363,0.004344657,0.01262069,0.0003696762],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9878478,0.0002215765,0.0001940881,0.01029224,0.00008924255,0.00008305554,0.00000319542,0.000001825325,0.00126698],"genre_scores_gemma":[0.9992408,0.0004571491,0.00002478703,0.0001500135,0.00001391722,1.8151e-7,1.350175e-7,0.000001777996,0.0001113092],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2242484,"threshold_uncertainty_score":0.7290528,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03882844735276887,"score_gpt":0.2512243027153732,"score_spread":0.2123958553626044,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}