{"id":"W4256577654","doi":"10.1007/978-1-4939-7131-2_100348","title":"Evaluation Approaches","year":2018,"lang":"en","type":"book-chapter","venue":"","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Calgary","funders":"","keywords":"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01299026,0.001680053,0.001072748,0.005325647,0.001559407,0.008669307,0.002676448,0.002232024,0.04523471],"category_scores_gemma":[0.02425307,0.0005663675,0.000918264,0.003503888,0.003129589,0.006236919,0.003585438,0.002618489,0.0166822],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005160514,"about_ca_system_score_gemma":0.005637066,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002225489,"about_ca_topic_score_gemma":0.002643297,"domain_scores_codex":[0.9843234,0.00674355,0.0007675614,0.001354324,0.006340555,0.0004705097],"domain_scores_gemma":[0.9890658,0.006063579,0.0003780869,0.001408127,0.002817619,0.000266844],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001364345,0.00003966242,0.0001164881,0.0004014864,0.00002064708,0.00002595465,0.0002683444,0.001122113,0.0001628476,0.7426443,0.03944848,0.2157361],"study_design_scores_gemma":[0.00001098149,0.00002218685,0.0002329965,0.0007973561,0.00002202175,0.00009941325,0.0002954704,0.002393042,0.0005795403,0.5221438,0.4733846,0.00001850199],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"methods","genre_scores_codex":[0.0008635695,0.008661931,0.2469178,0.005153025,0.001059452,0.0006074733,0.0003627115,0.0005276435,0.7358463],"genre_scores_gemma":[0.08435329,0.02026028,0.2810827,0.004998686,0.002020079,0.002402086,0.001492442,0.0008194022,0.602571],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.04523471,"threshold_uncertainty_score":0.1513252,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7348958051093804,"score_gpt":0.529804795289752,"score_spread":0.2050910098196284,"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."}}