{"id":"W4387429753","doi":"10.1177/02655322231202947","title":"Our validity looks like justice. Does yours?","year":2023,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":27,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Lethbridge","funders":"","keywords":"Economic Justice; Oppression; Psychology; Licensure; Set (abstract data type); Social psychology; Mathematics education; Pedagogy; Law; Political science; 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":[0.0009659495,0.00009823933,0.0001186746,0.00006772931,0.0006002703,0.0001531346,0.0002785468,0.000071526,0.0001168319],"category_scores_gemma":[0.001352648,0.00007979859,0.00004718095,0.0006186633,0.00004147402,0.0001366406,0.0001072283,0.0001505593,0.000480638],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004373112,"about_ca_system_score_gemma":0.00007037711,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002054527,"about_ca_topic_score_gemma":0.0009748512,"domain_scores_codex":[0.9986535,0.0001098266,0.0001318718,0.0002061942,0.0004596371,0.0004389512],"domain_scores_gemma":[0.9992421,0.000374879,0.00007408189,0.0001468281,0.00006604948,0.0000960645],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.000008392523,0.00009093931,0.7409391,0.0001066006,0.00004430742,0.0004566314,0.1979117,0.00002210043,0.005274525,0.002637563,0.02542272,0.02708534],"study_design_scores_gemma":[0.000614701,0.00006505467,0.2255498,0.0001364411,0.0002020808,0.000002362614,0.7611564,0.0006111523,0.0002490338,0.001368878,0.009372399,0.0006716757],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8966624,0.00004253489,0.00002493187,0.0008389241,0.001366069,0.0001577929,0.000005546059,0.000987027,0.09991485],"genre_scores_gemma":[0.9799092,0.000009063267,0.001238313,0.0001896213,0.001612206,0.00001018265,0.000008395496,0.00001531933,0.01700765],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5632446,"threshold_uncertainty_score":0.617779,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1345071057179136,"score_gpt":0.4132933080995925,"score_spread":0.2787862023816789,"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."}}