{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004046123,0.00005302234,0.00009308812,0.001004798,0.0002780647,0.00003084283,0.0003133437,0.00004959811,0.0000180777],"category_scores_gemma":[0.001034492,0.00004108156,0.00001401099,0.001075875,0.0007720466,0.000282032,0.00001443656,0.0001086797,5.812768e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001905951,"about_ca_system_score_gemma":0.001534914,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002214218,"about_ca_topic_score_gemma":0.0003207571,"domain_scores_codex":[0.9991593,0.00001881327,0.0002733229,0.0000839684,0.0002373145,0.0002272945],"domain_scores_gemma":[0.9988192,0.0001229247,0.0003308853,0.0001468802,0.0004873063,0.00009283893],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000003329097,0.0004279181,0.8709214,0.00004986068,0.000009509212,7.004807e-8,0.01994682,0.00007542287,0.0100717,0.01737984,0.00008243952,0.08103172],"study_design_scores_gemma":[0.0008376091,0.0002220674,0.8691344,0.0003693031,0.0001509653,0.0003798393,0.08402648,0.03160538,0.005528411,0.006933998,0.0005839731,0.0002275436],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.995706,0.0006791367,0.001581428,0.0005474802,0.0009788834,0.0003336693,0.000001204684,0.000001874772,0.0001703712],"genre_scores_gemma":[0.9940292,0.00001222929,0.005863541,0.00001879563,0.00003455508,0.0000194532,1.785315e-7,0.000003501978,0.00001854653],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08080417,"threshold_uncertainty_score":0.2844638,"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."}}