{"id":"W4390724925","doi":"10.1007/s42330-023-00296-9","title":"Formative Assessment Strategies by Monitoring Science Students’ Problem-Solving Skill Development","year":2023,"lang":"en","type":"article","venue":"Canadian Journal of Science Mathematics and Technology Education","topic":"Science Education and Pedagogy","field":"Social Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Formative assessment; Competence (human resources); Mathematics education; Psychology; Cognition; Cognitive development; Social psychology","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":["sts"],"consensus_categories":["sts"],"category_scores_codex":[0.005387307,0.00008441339,0.0001211877,0.001993225,0.003182551,0.0008019375,0.001090569,0.0000582539,0.00002347904],"category_scores_gemma":[0.0007071017,0.00007587444,0.00001268321,0.004205258,0.004133661,0.001386381,0.00005169571,0.0002053842,0.000009127941],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006786188,"about_ca_system_score_gemma":0.03815594,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006785055,"about_ca_topic_score_gemma":0.001228371,"domain_scores_codex":[0.9981986,0.00001816506,0.0003343654,0.0001751181,0.0007675039,0.0005062419],"domain_scores_gemma":[0.9980834,0.00006560908,0.0003035037,0.0001248193,0.0009474892,0.0004752507],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[4.666778e-7,0.0001609217,0.07388794,0.00004232327,0.0000125625,0.000002840789,0.5357049,0.000009893153,0.006789412,0.3001346,0.001647657,0.08160648],"study_design_scores_gemma":[0.00007183069,0.00004312353,0.01777702,0.000159891,0.00000633268,0.00001908893,0.940136,0.00004428233,0.002160185,0.03365025,0.005784071,0.00014797],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9889975,0.0001090938,0.0003468053,0.003888589,0.00122232,0.0001627296,6.583048e-7,0.00002544674,0.005246877],"genre_scores_gemma":[0.9844217,0.00007005426,0.01513614,0.00002372741,0.00005411091,0.00001133957,3.358246e-7,0.000003860564,0.0002787444],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4044311,"threshold_uncertainty_score":0.9985765,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03209360496268615,"score_gpt":0.4027715150529647,"score_spread":0.3706779100902786,"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."}}