{"id":"W2087097337","doi":"10.1145/1022494.1022539","title":"Relevant empirical testing research","year":2004,"lang":"en","type":"article","venue":"ACM SIGSOFT Software Engineering Notes","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Software testing; Empirical research; Test strategy; Computer science; Software engineering; Position paper; Software; Software performance testing; Data science; Systems engineering; Engineering; Software construction; Software development; World Wide Web","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":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001322644,0.0003231102,0.000297428,0.0004658513,0.0003392611,0.0003024626,0.002342291,0.0001910678,0.000003707349],"category_scores_gemma":[0.51872,0.0003185662,0.00009311486,0.001881127,0.00008217432,0.0004452992,0.001251552,0.000834657,0.0001311067],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002358331,"about_ca_system_score_gemma":0.0002479949,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001681014,"about_ca_topic_score_gemma":0.000001902721,"domain_scores_codex":[0.9970318,0.00006156751,0.0003917247,0.0007771871,0.0007835544,0.0009541708],"domain_scores_gemma":[0.8677852,0.1286027,0.00009795606,0.002560145,0.0005926419,0.0003613456],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001307123,0.000408561,0.8555233,0.0003484164,0.0000669108,0.0007545282,0.002705346,0.009665702,0.004745583,0.007222175,0.01152802,0.1070184],"study_design_scores_gemma":[0.002924069,0.002456382,0.7017004,0.004303725,0.00005558055,0.001821211,0.00002128427,0.01086356,0.06381981,0.194362,0.01228476,0.005387253],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1523056,0.0005385529,0.7955483,0.001910319,0.000513497,0.0003104776,0.000004104388,0.04885053,0.00001864696],"genre_scores_gemma":[0.4973495,0.000003496324,0.5023221,0.0001219793,0.0001191898,0.00003731942,0.000002152007,0.00003607828,0.000008232168],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5173973,"threshold_uncertainty_score":0.9999266,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1418802265110641,"score_gpt":0.360298768827243,"score_spread":0.2184185423161789,"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."}}