{"id":"W2991736603","doi":"","title":"Using Test Data to Find Misconceptions in Secondary Science.","year":2017,"lang":"en","type":"article","venue":"School science review","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Extant taxon; Mathematics education; Test (biology); Science education; Subject (documents); Scientific misconceptions; Psychology; Competition (biology); Computer science","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":["metaresearch","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008515203,0.00009030927,0.0001776401,0.0001943257,0.004036489,0.0008800331,0.004739972,0.00002454965,0.001921707],"category_scores_gemma":[0.008371466,0.0000816691,0.00002479927,0.001502298,0.002658007,0.003698665,0.0006974993,0.0001787866,0.0003321364],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000371627,"about_ca_system_score_gemma":0.01316127,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002830778,"about_ca_topic_score_gemma":0.003409118,"domain_scores_codex":[0.997596,0.00006114345,0.0002546911,0.0006012985,0.0009331191,0.0005537826],"domain_scores_gemma":[0.9977214,0.0001013508,0.0001609892,0.001283399,0.0002472462,0.0004855738],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00000274878,0.0004107524,0.5868396,0.0004765208,0.000005925121,0.00001118444,0.004617192,0.00000853986,0.01814977,0.04720929,0.02873799,0.3135305],"study_design_scores_gemma":[0.0001039162,0.00002068305,0.6550462,0.002134542,0.00001706726,0.000003361048,0.001623161,0.00006384004,0.00002190574,0.001680294,0.3389604,0.0003245655],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2429771,0.02230935,0.00008776113,0.1178522,0.003382445,0.002826995,0.0001147742,0.00007517781,0.6103742],"genre_scores_gemma":[0.9649386,0.01634958,0.01069095,0.003419816,0.0007691156,0.00003872626,0.000007351316,0.000009470796,0.003776337],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7219616,"threshold_uncertainty_score":0.9999815,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4929625508743163,"score_gpt":0.6059797759979968,"score_spread":0.1130172251236805,"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."}}