{"id":"W2288920388","doi":"10.18806/tesl.v32i0.1218","title":"Walk a Mile in My Shoes: Stakeholder Accounts of Testing Experience with a Computer-Administered Test","year":2016,"lang":"en","type":"article","venue":"TESL Canada Journal","topic":"Discourse Analysis in Language Studies","field":"Arts and Humanities","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Test (biology); Psychology; Context (archaeology); Language assessment; Stakeholder; Applied psychology; Social psychology; Pedagogy; Public relations; Political science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"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":[0.000162582,0.0001607431,0.0002919543,0.0001136488,0.0002110603,0.0001078795,0.0002616089,0.00001465122,0.01044199],"category_scores_gemma":[0.0001730386,0.00009186804,0.00003264931,0.00008658949,0.0002895451,0.0002619904,0.00005021695,0.0001190642,0.000001768111],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001718096,"about_ca_system_score_gemma":0.0005789028,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.05762835,"about_ca_topic_score_gemma":0.6378977,"domain_scores_codex":[0.9986169,0.00002900416,0.0003998207,0.0001668003,0.0004694578,0.0003179762],"domain_scores_gemma":[0.9988782,0.0003183135,0.000265694,0.0001803188,0.000273948,0.00008354615],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003679049,0.001749094,0.2366318,0.0003965756,0.001585911,0.006965145,0.4124456,0.0002546593,0.004015983,0.006996216,0.1939818,0.1346093],"study_design_scores_gemma":[0.01294136,0.004786535,0.07714409,0.0102669,0.0006048757,0.001572848,0.5256914,0.001034369,0.002425401,0.0009510201,0.3584328,0.004148436],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9837105,0.001984232,0.00007768779,0.0005782615,0.000281709,0.0001125616,0.000107448,0.00001640224,0.0131312],"genre_scores_gemma":[0.9960684,0.000006470163,0.0003722273,0.001678893,0.0003575605,0.000007648809,6.468841e-7,0.00001582691,0.001492366],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5802693,"threshold_uncertainty_score":0.9904626,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06509561069371143,"score_gpt":0.2435075109749999,"score_spread":0.1784119002812884,"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."}}