{"id":"W4389915531","doi":"10.21449/ijate.1321061","title":"A data pipeline for e-large-scale assessments: Better automation, quality assurance, and efficiency","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Mental Health Research Topics","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Northern Alberta Institute of Technology","funders":"International Atomic Energy Agency","keywords":"Workflow; Computer science; Pipeline (software); Scale (ratio); Quality assurance; Automation; Interface (matter); Quality (philosophy); Data science; Data quality; Risk analysis (engineering); Data mining; Database; Engineering; Operations management; Metric (unit)","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":[0.004148097,0.0001101951,0.0002076919,0.000515919,0.00006500458,0.0001396438,0.00073702,0.00007420102,0.000241466],"category_scores_gemma":[0.0004870832,0.0001055516,0.00004295438,0.0002906015,0.00002967514,0.0007961417,0.0001516585,0.0002743568,0.00001912217],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003943152,"about_ca_system_score_gemma":0.0007828641,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004321286,"about_ca_topic_score_gemma":0.0000860449,"domain_scores_codex":[0.9974316,0.0002816895,0.0009689483,0.0002811443,0.0007636819,0.0002728874],"domain_scores_gemma":[0.9975968,0.0007812559,0.0005489751,0.0003349976,0.0006378611,0.0001000881],"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":[0.000352211,0.003800635,0.2543087,0.0002477709,0.0002106641,0.0000205505,0.002386365,0.00008419181,0.0004799358,0.01766821,0.1849834,0.5354573],"study_design_scores_gemma":[0.002219481,0.0001615677,0.9099379,0.0001995728,0.00001532412,0.00003504232,0.004204353,0.01258717,0.00001938715,0.003582656,0.06689382,0.0001437593],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8682092,0.000267524,0.09822825,0.01792402,0.009729646,0.0008892681,0.0005611629,0.00003949528,0.004151454],"genre_scores_gemma":[0.9689685,0.0002694596,0.02622307,0.001038416,0.001173167,0.0001432106,0.0009858629,0.00002315677,0.001175129],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6556291,"threshold_uncertainty_score":0.4304268,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1654467129443049,"score_gpt":0.5848930899388985,"score_spread":0.4194463769945936,"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."}}