{"id":"W2168777020","doi":"10.1080/14926156.2012.732191","title":"Using Multiple Solution Tasks for the Evaluation of Students’ Problem-Solving Performance in Geometry","year":2012,"lang":"en","type":"article","venue":"Canadian Journal of Science Mathematics and Technology Education","topic":"Creativity in Education and Neuroscience","field":"Psychology","cited_by":35,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Creativity; Fluency; Social connectedness; Originality; Geometry; Correctness; Context (archaeology); Flexibility (engineering); Computer science; Mathematics education; Analytic geometry; Connection (principal bundle); Mathematics; Psychology; Algorithm; Social psychology; Statistics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00211142,0.000959496,0.0006796327,0.001202192,0.0003300593,0.0009500206,0.0007436306,0.0009165831,0.00230032],"category_scores_gemma":[0.01565753,0.0002803809,0.0005937455,0.0006106214,0.0003166207,0.001114466,0.0006915521,0.001091287,0.00067378],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004117062,"about_ca_system_score_gemma":0.0007219459,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001549243,"about_ca_topic_score_gemma":0.002557379,"domain_scores_codex":[0.9982507,0.0006146779,0.0002642598,0.0002827148,0.0004821475,0.000105392],"domain_scores_gemma":[0.9905846,0.005301541,0.001735954,0.0004379697,0.001099216,0.0008406303],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01055657,0.02404607,0.3067752,0.0008923695,0.0008510748,0.0004035975,0.009154564,0.01128937,0.1134371,0.001578472,0.003678344,0.5173373],"study_design_scores_gemma":[0.001739621,0.02483734,0.8747518,0.0001295371,0.000264207,0.0007202352,0.002774154,0.04308711,0.04428151,0.002523673,0.004614913,0.0002758966],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9970376,0.00004526323,0.001793006,0.00003085704,0.00002040794,0.0001327452,0.00007106083,0.00002512714,0.0008438413],"genre_scores_gemma":[0.9891434,0.0001055031,0.008384053,0.00002710695,0.00001335826,0.0004477851,0.0002994369,0.00001625138,0.00156305],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00230032,"threshold_uncertainty_score":0.01116639,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.112585942745943,"score_gpt":0.4127235965458379,"score_spread":0.3001376537998949,"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."}}