{"id":"W4362683301","doi":"10.3138/cjpe.18.002","title":"The Language of Evaluation Theory: Insights Gained from an Empirical Study of Evaluation Theory and Practice","year":2003,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Terminology; Ambiguity; Confusion; Epistemology; Field (mathematics); Vernacular; Empirical research; Psychology; Linguistics; Sociology; Computer science; Philosophy","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1587768,0.0004352005,0.001194326,0.006175108,0.007364613,0.0189218,0.002082191,0.00275929,0.003974317],"category_scores_gemma":[0.3476903,0.0007378128,0.0005331867,0.008225615,0.03621741,0.03065108,0.008654378,0.007362838,0.0003907696],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01721682,"about_ca_system_score_gemma":0.01489051,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00785162,"about_ca_topic_score_gemma":0.007219051,"domain_scores_codex":[0.8133613,0.1583256,0.005066535,0.00291291,0.0175133,0.002820197],"domain_scores_gemma":[0.3213283,0.6346717,0.01296293,0.006425544,0.02253646,0.002075078],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001555991,0.0006448482,0.0292428,0.001004223,0.00003916336,0.0004739524,0.5975813,0.00107521,0.0004099686,0.3054845,0.00175449,0.06213391],"study_design_scores_gemma":[0.0000989047,0.0003751578,0.02578959,0.004432948,0.00006002015,0.0004179041,0.7691315,0.009867581,0.001181767,0.1688925,0.01965353,0.00009867726],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7939139,0.004731739,0.07001135,0.03143054,0.0001714787,0.0006976783,0.00008038101,0.00005373539,0.09890927],"genre_scores_gemma":[0.9942773,0.0005219213,0.004178576,0.0004070611,0.00001548594,0.0001836177,0.00001381976,0.00001954743,0.0003825601],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1587768,"threshold_uncertainty_score":0.8397022,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2871302531925219,"score_gpt":0.573348207363572,"score_spread":0.28621795417105,"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."}}