{"id":"W4225015069","doi":"10.3758/s13428-022-01861-0","title":"Accuracy of paper-and-pencil systematic observation versus computer-aided systems","year":2022,"lang":"en","type":"article","venue":"Behavior Research Methods","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Brock University","funders":"Universidad Autónoma de Madrid; University of Auckland","keywords":"Pencil (optics); Computer science; Computer-aided; Computer graphics (images); Artificial intelligence; Programming language; Engineering","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01034605,0.0002068928,0.0006818493,0.0002156706,0.0006252563,0.00005911706,0.0006050377,0.0001180929,0.001650197],"category_scores_gemma":[0.0005231645,0.0001665048,0.0001325748,0.0008878646,0.0002788937,0.0001087593,0.0006593769,0.0008127484,0.0000402308],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000154487,"about_ca_system_score_gemma":0.00003534409,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001922704,"about_ca_topic_score_gemma":0.00000695771,"domain_scores_codex":[0.9877293,0.009330288,0.0007792623,0.000578702,0.0009989707,0.0005834381],"domain_scores_gemma":[0.9952277,0.003305432,0.0002800875,0.0007022676,0.000329689,0.0001548099],"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":[0.002848992,0.005886822,0.0340858,0.00737883,0.0006007156,0.0005316042,0.01446068,0.00004279428,0.04735267,0.03790956,0.009913368,0.8389882],"study_design_scores_gemma":[0.006679522,0.008877145,0.9250941,0.0008291426,0.0005658651,0.0002327825,0.03274781,0.0002660522,0.0003087404,0.0005383539,0.02274238,0.001118167],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9816059,0.008158593,0.001212232,0.000268807,0.00342182,0.003048975,0.00006361832,0.0001156679,0.002104317],"genre_scores_gemma":[0.977938,0.00009881136,0.01367227,0.00004869642,0.0001454807,0.006171277,0.0000264993,0.00004058298,0.001858391],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8910083,"threshold_uncertainty_score":0.9992625,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8223134969967072,"score_gpt":0.6146735580251322,"score_spread":0.207639938971575,"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."}}