{"id":"W3021197821","doi":"10.3390/ani10050781","title":"Does Thirty-Minute Standardised Training Improve the Inter-Observer Reliability of the Horse Grimace Scale (HGS)? A Case Study","year":2020,"lang":"en","type":"article","venue":"Animals","topic":"Veterinary Equine Medical Research","field":"Veterinary","cited_by":33,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"National Centre for the Replacement, Refinement and Reduction of Animals in Research","keywords":"Reliability (semiconductor); Psychology; Horse; Physical therapy; Physical medicine and rehabilitation; Medicine","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":[0.002121127,0.0002736582,0.0004924277,0.00002873159,0.0002704521,0.00007196621,0.0009946587,0.00009266201,0.0006080993],"category_scores_gemma":[0.002638616,0.0001164197,0.000289698,0.0004300518,0.0004951938,0.0001744078,0.001369589,0.000731584,0.00003454037],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000844411,"about_ca_system_score_gemma":0.0002059492,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001209572,"about_ca_topic_score_gemma":0.0003416614,"domain_scores_codex":[0.9962545,0.001086843,0.0006468007,0.0006282857,0.0008674286,0.0005161528],"domain_scores_gemma":[0.9974545,0.000813054,0.0001928144,0.001068367,0.0002137596,0.0002574792],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"qualitative","study_design_scores_codex":[0.01903375,0.003037424,0.13833,0.001710257,0.0009872605,0.02738976,0.357663,0.00001768006,0.3927377,0.0001087241,0.01003209,0.04895231],"study_design_scores_gemma":[0.01652981,0.1104254,0.274901,0.0008170175,0.001103763,0.005971458,0.5095196,0.009071714,0.01577212,0.001549065,0.05154679,0.002792168],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9907116,0.00004807604,0.0000562965,0.006478049,0.00024347,0.001625816,0.0001454861,0.00006346462,0.0006277687],"genre_scores_gemma":[0.9988434,0.000004418589,0.0001333978,0.0004000793,0.0002119831,0.0001111173,0.000001208211,0.00004257276,0.000251857],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3769656,"threshold_uncertainty_score":0.6658261,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1375896401226467,"score_gpt":0.3855516134574291,"score_spread":0.2479619733347824,"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."}}