{"id":"W4391037889","doi":"10.25035/pad.2024.01.003","title":"Calculator Provision as an Accommodation for the Canadian Forces Aptitude Test (CFAT)","year":2024,"lang":"en","type":"article","venue":"Personnel Assessment and Decisions","topic":"Human Resource Development and Performance Evaluation","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Calculator; Test (biology); Aptitude; Psychology; Reliability (semiconductor); Accommodation; Mathematics education; Computer science; Developmental psychology","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0008681586,0.0001533755,0.0001190249,0.0002358217,0.001265065,0.0005540417,0.0001909216,0.0001124694,0.001267794],"category_scores_gemma":[0.0000916168,0.0001016909,0.00006377195,0.0001832266,0.00005050386,0.0002946339,0.00002932697,0.0001710036,0.0001716096],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001440275,"about_ca_system_score_gemma":0.0003498256,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005684047,"about_ca_topic_score_gemma":0.04908124,"domain_scores_codex":[0.9987928,0.00004261184,0.0002209807,0.0003594773,0.0002892388,0.0002948382],"domain_scores_gemma":[0.9985642,0.0008725811,0.00003837291,0.0002562875,0.0000867677,0.0001817643],"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":[0.0001312844,0.0002946432,0.08375657,0.00005752303,0.0003540243,0.00002667118,0.06812945,0.000226293,0.0003527199,0.1172494,0.04565674,0.6837646],"study_design_scores_gemma":[0.001149941,0.0006158198,0.4053463,0.0001436838,0.0002012344,0.0000338327,0.01720656,0.3246382,0.00000998823,0.005269916,0.2448585,0.0005260435],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9804565,0.0006358679,0.006896874,0.002557121,0.000944701,0.001160753,0.00004882519,0.0000989293,0.00720046],"genre_scores_gemma":[0.9946799,0.0000411222,0.001521753,0.0002776395,0.0003332579,0.000333925,0.0001855101,0.00002418694,0.002602718],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6832386,"threshold_uncertainty_score":0.9996452,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09007763641916103,"score_gpt":0.4371027858237127,"score_spread":0.3470251494045516,"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."}}