{"id":"W4417453887","doi":"10.1108/jsocm-11-2024-0262","title":"Scottish reusable coffee cups: a multi-intervention CBSM benchmark analysis","year":2025,"lang":"en","type":"article","venue":"Journal of Social Marketing","topic":"Global trade, sustainability, and social impact","field":"Business, Management and Accounting","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"British Science Association","keywords":"Benchmark (surveying); Scope (computer science); Psychological intervention; Key (lock); Set (abstract data type); Intervention (counseling)","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004533893,0.0001955567,0.0005723797,0.0006611234,0.0005859477,0.0006123255,0.0003607582,0.0001469271,0.0002547619],"category_scores_gemma":[0.003300831,0.0001927654,0.0009600172,0.002125916,0.0000936249,0.00114934,0.0001543242,0.0003532417,0.000006383902],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003190926,"about_ca_system_score_gemma":0.00008147285,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007352246,"about_ca_topic_score_gemma":0.0002256136,"domain_scores_codex":[0.9979311,0.0001743363,0.0008706144,0.0002000613,0.0004157572,0.0004081873],"domain_scores_gemma":[0.9977786,0.0001808228,0.001008018,0.0001194746,0.0008871691,0.00002592526],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009762812,0.001200436,0.9272256,0.001645737,0.00359181,0.00009568464,0.001444698,0.0001728967,0.0001551674,0.005014298,0.02904025,0.02943708],"study_design_scores_gemma":[0.001955318,0.00002190708,0.9516927,0.0003705143,0.002794124,0.000001565693,0.02176473,0.001054159,0.000007140294,0.004506437,0.0154996,0.000331822],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9528714,0.0004554483,0.001388029,0.002122897,0.001128569,0.0002279605,0.000004629467,0.00005191329,0.04174912],"genre_scores_gemma":[0.9969454,0.00001993483,0.0002939801,0.0005228099,0.001398928,0.000002158435,0.000007584807,0.00001347375,0.0007957334],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04407396,"threshold_uncertainty_score":0.7860745,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01405556417846751,"score_gpt":0.2894176184173998,"score_spread":0.2753620542389323,"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."}}