{"id":"W3092500786","doi":"10.47670/wuwijar201821lh","title":"Direct assessment of second language writing: Holistic and analytic scoring","year":2018,"lang":"en","type":"article","venue":"Westcliff International Journal of Applied Research","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Wycliffe College","funders":"","keywords":"Rubric; Writing assessment; Language assessment; Task (project management); Test (biology); Mathematics education; Standards-based assessment; Reliability (semiconductor); Peer assessment; Psychology; Educational assessment; Computer science","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":[0.004118748,0.00007336249,0.0001995392,0.0004582709,0.0002119314,0.0001792971,0.0005831615,0.00005229736,0.0007771702],"category_scores_gemma":[0.0002320004,0.00006502123,0.00005703363,0.0002906217,0.0006044782,0.0001516358,0.0001628571,0.0003189319,0.00001018089],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001753167,"about_ca_system_score_gemma":0.0003271256,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001345342,"about_ca_topic_score_gemma":0.0002706816,"domain_scores_codex":[0.9972514,0.0001296074,0.0003962625,0.0001419844,0.001824962,0.0002558148],"domain_scores_gemma":[0.9978742,0.0005815964,0.000240779,0.00008946581,0.001085529,0.0001284456],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008392234,0.001080908,0.3764555,0.0001414287,0.001739985,0.0002943853,0.06999467,0.00002715177,0.1291042,0.3705464,0.008854318,0.04092182],"study_design_scores_gemma":[0.003641723,0.0007986943,0.854688,0.000519008,0.0000945393,0.00002918697,0.1006754,0.0004099937,0.008513853,0.006755574,0.02340904,0.0004649973],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7533541,0.00006071447,0.00007115144,0.0003409313,0.0003143726,0.00009369109,0.000004478331,0.000005431968,0.2457552],"genre_scores_gemma":[0.9965631,0.0001365099,0.001040558,0.00003389312,0.00146438,0.00000288311,0.000001410298,0.000008569677,0.0007487255],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4782325,"threshold_uncertainty_score":0.8509468,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09942307377488162,"score_gpt":0.5055932593544762,"score_spread":0.4061701855795946,"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."}}