{"id":"W7132906485","doi":"","title":"The Promise of Responsible Research Assessment","year":2025,"lang":"en","type":"dissertation","venue":"Trepo - Institutional Repository of Tampere University","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of Alberta; VSNU Vereniging van Universiteiten; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; Research England; Koninklijke Nederlandse Akademie van Wetenschappen; Newcastle University","keywords":"Cynicism; Doctoral studies; Accountability","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.006489284,0.0001774548,0.0004020743,0.0008241046,0.002075326,0.0000978689,0.001363845,0.0002390907,0.0001809379],"category_scores_gemma":[0.001943017,0.0001309072,0.0002797003,0.001265078,0.0006113485,0.0003509006,0.0001814837,0.0005545073,0.00001553714],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000565668,"about_ca_system_score_gemma":0.01493906,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003123961,"about_ca_topic_score_gemma":0.0006891484,"domain_scores_codex":[0.9932078,0.001067949,0.0009188246,0.0005113333,0.004069343,0.0002247623],"domain_scores_gemma":[0.9906064,0.002633602,0.0008746661,0.0009325795,0.004865563,0.00008716265],"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":[0.007424146,0.001497119,0.02106437,0.000683737,0.000848955,0.0002174687,0.005495317,0.004849545,0.02722349,0.8481024,0.04476355,0.03782992],"study_design_scores_gemma":[0.001748593,0.0005202106,0.2501918,0.00107194,0.0002222243,0.00002369211,0.0318761,0.002495258,0.02688626,0.01272712,0.6717308,0.0005059551],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5277931,0.0001747112,0.00004083559,0.0003938598,0.001305239,0.0004535977,0.00004476083,0.00001407425,0.4697798],"genre_scores_gemma":[0.702052,0.0001440559,0.0005785083,0.000005380147,0.00004765462,0.000004479211,0.0000857036,0.000004457466,0.2970778],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8353752,"threshold_uncertainty_score":0.9992238,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2129773060336819,"score_gpt":0.5109170586383193,"score_spread":0.2979397526046374,"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."}}