{"meta":{"query_hash":"586ff0674e7a","filters":{"venue":"Evaluation"},"cohort_total":46,"direct_labels_cover":2,"predictions_cover":46,"exported":46,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/586ff0674e7a","api":"https://metacan.xera.ac/api/v1/cohort?venue=Evaluation"},"results":[{"id":"W1972177361","doi":"10.1177/1356389008095487","title":"A Transaction Cost-Based Approach to Partnership Performance Evaluation","year":2008,"lang":"en","type":"article","venue":"Evaluation","topic":"Banking stability, regulation, efficiency","field":"Economics, Econometrics and Finance","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Safety Canada","funders":"","keywords":"General partnership; Transaction cost; Corporate governance; Key (lock); Process management; Business; Database transaction; Limited partnership; Audit; Industrial organization; Risk analysis (engineering); Computer science; Accounting; Computer security; Finance","score_opus":0.12777864542751377,"score_gpt":0.29956379061594945,"score_spread":0.17178514518843568,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1972177361","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023190249,0.0022059474,0.8803383,0.0030482078,0.00027862596,0.002323039,0.00039745413,0.00013575726,0.08808245],"genre_scores_gemma":[0.6894829,0.0023737687,0.29580802,0.0003924295,0.00026670258,0.0027203637,0.00034198325,0.00010670245,0.008507021],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9275475,0.051355183,0.0026548563,0.0016889798,0.015351916,0.0014014719],"domain_scores_gemma":[0.9324392,0.04440291,0.005677561,0.0035047524,0.012924436,0.0010510718],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03633172,0.0018190009,0.0017853029,0.010439829,0.0014532738,0.0073467065,0.002601341,0.0022114664,0.0069340104],"category_scores_gemma":[0.083101995,0.00051799207,0.0012214438,0.010222125,0.0033574256,0.0072084977,0.0038040152,0.0029585457,0.0006545425],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016731744,0.00038800616,0.008230671,0.0007652243,0.0002950706,0.00019514399,0.00070016005,0.10166487,0.00064381526,0.6989733,0.004356648,0.18361978],"study_design_scores_gemma":[0.000091914204,0.0012023986,0.0080096405,0.0009387757,0.00021577162,0.00046869158,0.0033004135,0.40126112,0.0025442922,0.54790306,0.0338156,0.00024840634],"about_ca_topic_score_codex":0.003491951,"about_ca_topic_score_gemma":0.0036224166,"teacher_disagreement_score":0.03633172,"about_ca_system_score_codex":0.009311266,"about_ca_system_score_gemma":0.0051040202,"threshold_uncertainty_score":0.19214284},"labels":[],"label_agreement":null},{"id":"W2014366393","doi":"10.1177/1356389003094006","title":"Evaluability Assessment: A Tool for Incorporating Evaluation in Social Change Programmes","year":2003,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":61,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; University of Calgary","funders":"","keywords":"Behaviour change; Social change; Theory of change; Process management; Political science; Engineering ethics; Management science; Sociology; Public relations; Psychology; Engineering; Psychological intervention","score_opus":0.4539931329436821,"score_gpt":0.5881362452183663,"score_spread":0.13414311227468417,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2014366393","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0044954508,0.0010617063,0.9555759,0.0035221553,0.00036726243,0.0063881963,0.00054903305,0.002312003,0.025728332],"genre_scores_gemma":[0.060960572,0.00054980384,0.9281792,0.00034958695,0.00012824479,0.008108224,0.00028937365,0.00031920918,0.0011158743],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.57975996,0.34770653,0.028310927,0.0057741944,0.036660146,0.0017881923],"domain_scores_gemma":[0.33747485,0.5555155,0.022209913,0.035019938,0.04672412,0.003055727],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.37963408,0.0036938235,0.0037270857,0.027491584,0.004755317,0.013061464,0.0032286996,0.0035037997,0.010195148],"category_scores_gemma":[0.5303242,0.0014460792,0.0039296993,0.014002976,0.007301831,0.017505916,0.011603938,0.006069284,0.0014251556],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047461633,0.0006815336,0.00576243,0.0061724293,0.0007461093,0.00023649269,0.010514852,0.010772445,0.0012306385,0.19065608,0.016303523,0.75644875],"study_design_scores_gemma":[0.000716824,0.0021405742,0.01161693,0.01246665,0.0012255049,0.0006295101,0.0075276457,0.052888196,0.005996759,0.6773503,0.22658709,0.0008540774],"about_ca_topic_score_codex":0.0030443957,"about_ca_topic_score_gemma":0.0028333676,"teacher_disagreement_score":0.37963408,"about_ca_system_score_codex":0.008520317,"about_ca_system_score_gemma":0.016434336,"threshold_uncertainty_score":0},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"opus","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2026507234","doi":"10.1177/1356389009360471","title":"Peace and Conflict Impact Assessment (PCIA) in Community Development: A Case Study from Mozambique","year":2010,"lang":"en","type":"article","venue":"Evaluation","topic":"Environmental and Social Impact Assessments","field":"Environmental Science","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Situational ethics; Government (linguistics); Work (physics); Quality (philosophy); Public relations; Political science; Public administration; Sociology; Business; Engineering; Law","score_opus":0.04812356254503167,"score_gpt":0.4044966418320075,"score_spread":0.3563730792869758,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2026507234","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99162614,0.0002192662,0.0006381439,0.0007871125,0.0000071263053,0.00024576075,0.000048918464,0.0000073943866,0.006420199],"genre_scores_gemma":[0.99623954,0.00038350123,0.0013949177,0.00011300622,0.000004783198,0.00017436655,0.000033132706,0.0000046791383,0.0016520522],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.99745315,0.0015751339,0.000061112965,0.000096585405,0.00017114647,0.000642875],"domain_scores_gemma":[0.9970092,0.0018116868,0.00030235393,0.000115468814,0.00034828184,0.0004130277],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0033426986,0.00043907444,0.00033962593,0.0016047698,0.011462944,0.002162145,0.00083511975,0.0015906101,0.0019779848],"category_scores_gemma":[0.0056662858,0.00040976194,0.00031558488,0.0024746952,0.0038392362,0.0010149166,0.0027389615,0.0018906578,0.0001537552],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029903045,0.0014284644,0.098347336,0.0010818454,0.00007572837,0.06456869,0.71663195,0.0033267874,0.0052440558,0.009326487,0.0031072015,0.09656242],"study_design_scores_gemma":[0.0000657104,0.00079256017,0.13301355,0.00056329067,0.00004380643,0.005060173,0.8228008,0.002228982,0.001417728,0.0016484067,0.03228157,0.00008347303],"about_ca_topic_score_codex":0.17340434,"about_ca_topic_score_gemma":0.41980797,"teacher_disagreement_score":0.17340434,"about_ca_system_score_codex":0.009329124,"about_ca_system_score_gemma":0.006382164,"threshold_uncertainty_score":0.34479004},"labels":[],"label_agreement":null},{"id":"W2036342338","doi":"10.1177/1356389003009002006","title":"Reducing Anxiety and Resistance in Policy and Programme Evaluations","year":2003,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University; Simon Fraser University","funders":"","keywords":"Resistance (ecology); Anxiety; Process (computing); Psychology; Political science; Public relations; Social psychology; Computer science; Ecology; Biology","score_opus":0.22745225598283272,"score_gpt":0.5341188055677645,"score_spread":0.3066665495849318,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2036342338","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.948603,0.0008649025,0.012650855,0.010411099,0.00014568155,0.00045198243,0.000021360576,0.00018386549,0.026667317],"genre_scores_gemma":[0.99461824,0.0002148039,0.003357214,0.0006983005,0.0000591351,0.000199164,0.0000117521395,0.000023383825,0.00081798545],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.78100806,0.18757641,0.0047590504,0.0024813293,0.016573105,0.0076020546],"domain_scores_gemma":[0.73470676,0.2101716,0.024635408,0.010634522,0.0135920355,0.0062595815],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08652134,0.0008055718,0.00094758475,0.0026385002,0.0071459888,0.010823206,0.0018331183,0.0033268633,0.0022085863],"category_scores_gemma":[0.27980137,0.00068632053,0.0010693978,0.0015145758,0.006990858,0.005445364,0.009671984,0.005279315,0.00032127756],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011806397,0.0015837519,0.07174745,0.0012225688,0.00033333944,0.005810428,0.57220954,0.004079384,0.005961954,0.028573962,0.01358641,0.29371056],"study_design_scores_gemma":[0.00046550186,0.0056484044,0.12476489,0.0034806957,0.00056264095,0.0061595184,0.6435657,0.016570166,0.011878895,0.07161893,0.1144569,0.0008277791],"about_ca_topic_score_codex":0.0021523149,"about_ca_topic_score_gemma":0.0040367334,"teacher_disagreement_score":0.08652134,"about_ca_system_score_codex":0.0059464676,"about_ca_system_score_gemma":0.0054370677,"threshold_uncertainty_score":0.45757407},"labels":[],"label_agreement":null},{"id":"W2052280380","doi":"10.1177/1356389012461192","title":"The ethical sensitivity of evaluators: A qualitative study using a vignette design","year":2012,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université Laval","funders":"Government of Canada","keywords":"Vignette; Context (archaeology); Psychology; Qualitative research; Quality (philosophy); Social psychology; Engineering ethics; Sociology; Epistemology; Social science","score_opus":0.6117213930755443,"score_gpt":0.6475090385779075,"score_spread":0.03578764550236324,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2052280380","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9828075,0.00037212853,0.008296171,0.0028464766,0.00007119823,0.0010970206,0.00008698393,0.000025135438,0.004397353],"genre_scores_gemma":[0.9915211,0.00041524423,0.004702604,0.0007696755,0.000020763278,0.000950862,0.000040029754,0.000021967946,0.0015576968],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.894472,0.09567827,0.001544741,0.0019052462,0.0026225394,0.0037772206],"domain_scores_gemma":[0.86118495,0.11549512,0.005722998,0.0023124646,0.009820793,0.0054637208],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.056250177,0.000871408,0.000993387,0.0030954445,0.017222315,0.006149689,0.0026099514,0.0034982187,0.0021197996],"category_scores_gemma":[0.10507646,0.0010431146,0.00042865297,0.0030225124,0.014648759,0.0045227115,0.0058538276,0.0044818823,0.00031915915],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000062927145,0.000104327584,0.0019082787,0.000119074044,0.0000044654316,0.00079634925,0.9926843,0.000070336275,0.000581553,0.0013650743,0.00028231973,0.002021097],"study_design_scores_gemma":[0.000014162769,0.000121466524,0.0011168097,0.00015852363,0.0000048404295,0.0002853507,0.99090755,0.00024373192,0.00051875436,0.00046756846,0.006133094,0.000028148736],"about_ca_topic_score_codex":0.015170216,"about_ca_topic_score_gemma":0.021925915,"teacher_disagreement_score":0.94374985,"about_ca_system_score_codex":0.014677329,"about_ca_system_score_gemma":0.00860697,"threshold_uncertainty_score":0.2974829},"labels":[],"label_agreement":null},{"id":"W2081322617","doi":"10.1177/1356389005058475","title":"Polity, Politics and Policy Evaluation in Belgium","year":2005,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Polity; Parliament; Politics; Institutionalisation; Political science; Public administration; Corporate governance; Government (linguistics); Public policy; Phenomenon; Political economy; Comparative politics; Sociology; Economics; Law","score_opus":0.2900194241285808,"score_gpt":0.5822176039709627,"score_spread":0.2921981798423819,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2081322617","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5743333,0.057028968,0.014096531,0.038932,0.00028299988,0.00015148066,0.00017647099,0.00019183276,0.3148065],"genre_scores_gemma":[0.98903614,0.0032770385,0.001817755,0.0007396768,0.00005289467,0.00003127304,0.000032918804,0.000033199016,0.0049790223],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9696508,0.022322398,0.0008906882,0.0010374276,0.0033410017,0.002757761],"domain_scores_gemma":[0.97305185,0.018533735,0.0041285516,0.00054584397,0.0018015841,0.0019384319],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016366353,0.0003055496,0.00058819546,0.0043202187,0.0041769645,0.014665694,0.0005694314,0.0023873965,0.002987075],"category_scores_gemma":[0.0245152,0.00038543204,0.00035218333,0.006219079,0.011685513,0.00296895,0.0032183959,0.001900902,0.00026892836],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029508822,0.00029700354,0.03210975,0.0007955185,0.00015740245,0.00093656173,0.016443972,0.017737119,0.0014318207,0.69946915,0.006900477,0.22342621],"study_design_scores_gemma":[0.00018097542,0.0003433475,0.20979828,0.0021461756,0.00012817532,0.0008301794,0.03098326,0.009590842,0.003214349,0.2882807,0.4542285,0.00027526825],"about_ca_topic_score_codex":0.08274098,"about_ca_topic_score_gemma":0.077115394,"teacher_disagreement_score":0.08274098,"about_ca_system_score_codex":0.026289627,"about_ca_system_score_gemma":0.021293238,"threshold_uncertainty_score":0.19074547},"labels":[],"label_agreement":null},{"id":"W2093062308","doi":"10.1177/1356389012442445","title":"A socio-political framework for evaluability assessment of participatory evaluations of partnerships: Making sense of the power differentials in programs that involve the state and civil society","year":2012,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; University of Ottawa","funders":"","keywords":"General partnership; Citizen journalism; Participatory evaluation; Politics; Public relations; Civil society; Participatory action research; Power (physics); State (computer science); Political science; Public administration; Sociology; Computer science","score_opus":0.5679951806189625,"score_gpt":0.5888862518149395,"score_spread":0.020891071195976996,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2093062308","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029935243,0.0007761016,0.871553,0.011195491,0.00021951097,0.0069669955,0.00023952292,0.00011898012,0.07899517],"genre_scores_gemma":[0.54965854,0.00025971592,0.43607563,0.0007108762,0.000144375,0.012008344,0.000119557815,0.000053128497,0.0009698019],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.50948113,0.42882356,0.016832136,0.0060575465,0.034971997,0.0038336916],"domain_scores_gemma":[0.5638793,0.35609242,0.029009813,0.016387176,0.032031152,0.0026001723],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.36775017,0.0021336025,0.0020038881,0.016314343,0.008208381,0.012610477,0.0030825343,0.0034041135,0.004113114],"category_scores_gemma":[0.36053178,0.0010109291,0.0025961194,0.0063925805,0.03298356,0.010938004,0.009901284,0.004633101,0.00026983136],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000118305456,0.00021047791,0.004066558,0.0009191658,0.00021018244,0.00015304345,0.01210965,0.008263263,0.00052365847,0.9271727,0.001023208,0.045229703],"study_design_scores_gemma":[0.00017188776,0.0004704977,0.004934634,0.0013016319,0.00016747149,0.00012157394,0.010611308,0.022386475,0.0014809316,0.9394794,0.018749353,0.00012472311],"about_ca_topic_score_codex":0.004340884,"about_ca_topic_score_gemma":0.004862678,"teacher_disagreement_score":0.36775017,"about_ca_system_score_codex":0.018244892,"about_ca_system_score_gemma":0.023639826,"threshold_uncertainty_score":0.7796766},"labels":[],"label_agreement":null},{"id":"W2111087525","doi":"10.1177/1356389009360478","title":"Conceptual and Methodological Challenges in Producing Research Syntheses for Decision-and Policy-Making: An Illustrative Case in Primary Healthcare","year":2010,"lang":"en","type":"article","venue":"Evaluation","topic":"Primary Care and Health Outcomes","field":"Health Professions","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université de Montréal; Université de Sherbrooke","funders":"Canadian Institutes of Health Research; Canada Research Chairs; Ministère de la Santé et des Services sociaux; Canadian Health Services Research Foundation","keywords":"Rigour; Operationalization; Viewpoints; Management science; Health care; Engineering ethics; Foundation (evidence); Process (computing); Deliberation; Knowledge management; Computer science; Process management; Political science; Engineering; Epistemology; Politics","score_opus":0.7415406961974293,"score_gpt":0.6603945259424397,"score_spread":0.0811461702549896,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111087525","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.056985173,0.032771695,0.44672754,0.3933248,0.0027012443,0.008805201,0.0005751138,0.0003234389,0.05778585],"genre_scores_gemma":[0.38365406,0.010989236,0.5764247,0.014419668,0.0004933095,0.0095569715,0.00014217803,0.00013075962,0.0041891094],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.529311,0.43305558,0.012409027,0.0038194836,0.017569473,0.0038354385],"domain_scores_gemma":[0.43053913,0.53857255,0.007582135,0.008641593,0.012313501,0.002351123],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2819184,0.0019183471,0.002985262,0.00823264,0.020357084,0.022185717,0.0059832335,0.01823833,0.003415],"category_scores_gemma":[0.30951178,0.001822862,0.0027456929,0.013956006,0.028476452,0.016633376,0.013281797,0.0110187605,0.0007886242],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00031543776,0.00034623302,0.0018832362,0.010174438,0.00018352822,0.012134513,0.31368586,0.0078540845,0.0013838086,0.5816787,0.013444291,0.05691594],"study_design_scores_gemma":[0.0003693576,0.0003115917,0.0011717083,0.01619952,0.0002923646,0.003807174,0.33426172,0.010666898,0.0028974751,0.41677397,0.21293384,0.0003143166],"about_ca_topic_score_codex":0.022297857,"about_ca_topic_score_gemma":0.03539134,"teacher_disagreement_score":0.7180816,"about_ca_system_score_codex":0.029128289,"about_ca_system_score_gemma":0.0396597,"threshold_uncertainty_score":0.8855225},"labels":[],"label_agreement":null},{"id":"W2111340440","doi":"10.1177/1356389009105883","title":"How Legitimate and Justified are Judgments in Program Evaluation?","year":2009,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":43,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Montréal","funders":"","keywords":"Value (mathematics); Function (biology); Psychology; Order (exchange); Social psychology; Computer science; Economics","score_opus":0.3428563984356187,"score_gpt":0.5551089461910942,"score_spread":0.21225254775547547,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111340440","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.14069268,0.2171744,0.2139037,0.28454116,0.008625108,0.0017580839,0.0012290365,0.00056380586,0.13151208],"genre_scores_gemma":[0.9288931,0.013372257,0.040563304,0.0125349015,0.0022365672,0.0012660321,0.0002969689,0.00018817016,0.00064868113],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.21329382,0.62209463,0.051167805,0.013633848,0.09583285,0.0039769993],"domain_scores_gemma":[0.071455926,0.8244242,0.04321363,0.024262654,0.034096293,0.002547249],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.52377,0.00185709,0.0053121373,0.022469051,0.0056337076,0.025494363,0.0050307596,0.010169312,0.002633051],"category_scores_gemma":[0.8515333,0.0021686722,0.0037368683,0.014330476,0.046161134,0.036339577,0.009387719,0.010070787,0.0008635753],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021840748,0.0004163203,0.03317363,0.01643217,0.0063291183,0.0005519332,0.051958244,0.0034692585,0.00072628463,0.50113124,0.020996233,0.36263153],"study_design_scores_gemma":[0.0006303364,0.00044681778,0.012721209,0.018737458,0.0013638346,0.00044817402,0.014241929,0.0031861004,0.0016557575,0.91176003,0.03436797,0.00044039206],"about_ca_topic_score_codex":0.0046575675,"about_ca_topic_score_gemma":0.0051678387,"teacher_disagreement_score":0.47623003,"about_ca_system_score_codex":0.015915113,"about_ca_system_score_gemma":0.015994484,"threshold_uncertainty_score":0.58727646},"labels":[],"label_agreement":null},{"id":"W2116343005","doi":"10.1177/1356389012453289","title":"Towards an evidence base of theory-driven evaluations: Some questions for proponents of theory-driven evaluation","year":2012,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University; University of Toronto; St. Michael's Hospital","funders":"","keywords":"Management science; Causality (physics); Psychological intervention; Promotion (chess); Development theory; Theory of change; Computer science; Psychology; Engineering ethics; Sociology; Political science; Economics; Economic growth; Engineering","score_opus":0.37736500390818084,"score_gpt":0.5722308654359345,"score_spread":0.19486586152775365,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2116343005","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005944031,0.045866482,0.42661086,0.49001426,0.00454776,0.0030316361,0.00023837354,0.00040004795,0.023346554],"genre_scores_gemma":[0.22087756,0.019469554,0.688532,0.056789517,0.002238032,0.009718457,0.00028558576,0.00035721943,0.00173209],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.38596827,0.47650623,0.0368845,0.012077136,0.084698655,0.003865279],"domain_scores_gemma":[0.093531035,0.81195825,0.012566281,0.026263174,0.052769825,0.0029114229],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7093204,0.003338019,0.011049779,0.018128056,0.006242353,0.034778666,0.014798819,0.020857604,0.005633152],"category_scores_gemma":[0.755809,0.003597219,0.005564645,0.010140999,0.053368337,0.050387163,0.019303992,0.041862708,0.0016817125],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002916491,0.0003736715,0.0014694607,0.011839301,0.00080756156,0.00014119702,0.008957516,0.003707259,0.00021478682,0.82492435,0.0088014575,0.13847171],"study_design_scores_gemma":[0.00048433206,0.00030816876,0.00048708235,0.024020977,0.00029787613,0.000072758114,0.0034087112,0.008307198,0.00081186666,0.93023306,0.03142024,0.00014783336],"about_ca_topic_score_codex":0.005641839,"about_ca_topic_score_gemma":0.0038871567,"teacher_disagreement_score":0.29067957,"about_ca_system_score_codex":0.029953213,"about_ca_system_score_gemma":0.048209228,"threshold_uncertainty_score":0.3584597},"labels":[],"label_agreement":null},{"id":"W2128827052","doi":"10.1177/1356389009105884","title":"Critical Connections between Participatory Evaluation, Organizational Learning and Intentional Change in Pluralistic Organizations","year":2009,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":84,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Praxis; Citizen journalism; Sociology; Resistance (ecology); Process (computing); Organizational learning; Participatory action research; Organization development; Epistemology; Psychology; Knowledge management; Political science; Computer science","score_opus":0.34842097867393707,"score_gpt":0.5590078097974384,"score_spread":0.2105868311235013,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2128827052","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0983326,0.018882055,0.3412369,0.30042115,0.0016708452,0.0019339933,0.0000670235,0.00014167282,0.23731373],"genre_scores_gemma":[0.94488907,0.0025823081,0.044630416,0.0036754066,0.0002919383,0.00091636064,0.0000131855795,0.0000532627,0.0029481521],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.7553987,0.21400574,0.0047267308,0.0049787317,0.016574673,0.0043153805],"domain_scores_gemma":[0.6947962,0.27248624,0.009657795,0.009481447,0.010049526,0.0035288224],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17820369,0.0009752236,0.0013478224,0.0057972535,0.01540491,0.025658652,0.0031435844,0.007066301,0.0026253262],"category_scores_gemma":[0.15626729,0.0008135266,0.0008467783,0.0040158643,0.118422896,0.025773462,0.018267203,0.010014355,0.00015424789],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003520462,0.000048764472,0.0005022213,0.00034386857,0.00002265627,0.00020891122,0.061298415,0.0009140056,0.00011570704,0.91793716,0.0012238728,0.017349234],"study_design_scores_gemma":[0.00003349171,0.000054872184,0.00037793015,0.000853592,0.000013636602,0.0001401305,0.033532124,0.0010153848,0.0003858795,0.9343658,0.029190764,0.00003632091],"about_ca_topic_score_codex":0.0032006283,"about_ca_topic_score_gemma":0.0036153537,"teacher_disagreement_score":0.17820369,"about_ca_system_score_codex":0.017770458,"about_ca_system_score_gemma":0.021755563,"threshold_uncertainty_score":0.9424424},"labels":[],"label_agreement":null},{"id":"W2130276747","doi":"10.1177/1356389015593357","title":"Towards a comprehensive framework for the evaluation of small and medium enterprise policy","year":2015,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Additionality; Relevance (law); Public policy; Policy analysis; Process management; Business; Management science; Knowledge management; Computer science; Political science; Economics; Public economics; Public administration; Economic growth","score_opus":0.4988657924257035,"score_gpt":0.574468999590146,"score_spread":0.07560320716444247,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2130276747","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012468694,0.0069510406,0.86633265,0.039660245,0.00034110842,0.00464399,0.00035222768,0.00040323893,0.068846814],"genre_scores_gemma":[0.24532843,0.0022439333,0.7409336,0.0021372335,0.00013534365,0.007300219,0.00019340987,0.00010954882,0.0016182094],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5754796,0.37740567,0.01459867,0.00610842,0.021615827,0.0047917953],"domain_scores_gemma":[0.55871826,0.36790317,0.016020365,0.022944754,0.030291267,0.0041222745],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3932894,0.0033844572,0.0051880316,0.022925155,0.007913848,0.036278676,0.00807224,0.010393492,0.005701813],"category_scores_gemma":[0.27374813,0.0019447383,0.003503525,0.013841,0.038207684,0.02794675,0.015787859,0.010699354,0.0007298402],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004229904,0.0001586976,0.00084010296,0.0011022176,0.000090222806,0.000051009196,0.0038979691,0.0073241894,0.00017133106,0.95587194,0.0010423397,0.029407531],"study_design_scores_gemma":[0.000114507515,0.00017996188,0.00071483164,0.0039534434,0.00009260173,0.000040121893,0.0062822667,0.009573683,0.00066860125,0.959644,0.018665906,0.00007006644],"about_ca_topic_score_codex":0.0120386835,"about_ca_topic_score_gemma":0.009454791,"teacher_disagreement_score":0.3932894,"about_ca_system_score_codex":0.042247888,"about_ca_system_score_gemma":0.06585893,"threshold_uncertainty_score":0.7481822},"labels":[],"label_agreement":null},{"id":"W2133487638","doi":"10.1177/1356389012452052","title":"Applications of contribution analysis to outcome planning and impact evaluation","year":2012,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Accountability; General partnership; Participatory evaluation; Context (archaeology); Participatory planning; Citizen journalism; Theory of change; Process management; Process (computing); Value (mathematics); Management science; Appeal; Conceptual framework; Knowledge management; Public relations; Sociology; Computer science; Political science; Business; Public administration; Environmental planning; Engineering","score_opus":0.3089908516284891,"score_gpt":0.6239675731195479,"score_spread":0.31497672149105876,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133487638","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0071448027,0.0011473516,0.92384565,0.0041861725,0.0003296195,0.0036385586,0.0004352779,0.0008000658,0.058472533],"genre_scores_gemma":[0.2260898,0.0013535673,0.7579619,0.0007846472,0.00024685886,0.008554532,0.0005679262,0.00044933773,0.0039914325],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6667181,0.27741677,0.010400525,0.0074346466,0.03407228,0.0039576883],"domain_scores_gemma":[0.6047398,0.31994292,0.014417935,0.022383625,0.035514895,0.0030008792],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19090745,0.0036673518,0.0032145155,0.021391341,0.005064007,0.012978594,0.0049339524,0.003295998,0.011089493],"category_scores_gemma":[0.29873714,0.0015358435,0.004464129,0.017858671,0.013978583,0.013015957,0.015522866,0.0053552184,0.001847885],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003083694,0.0003523819,0.005052297,0.002855855,0.00044245485,0.00025332088,0.010625223,0.024713328,0.0003561834,0.63095874,0.006163663,0.31791815],"study_design_scores_gemma":[0.00019863395,0.0002985232,0.0019534687,0.0014916442,0.00017089954,0.00017527277,0.003610638,0.038749386,0.0013046905,0.9166563,0.035253223,0.00013728392],"about_ca_topic_score_codex":0.005633098,"about_ca_topic_score_gemma":0.0038716432,"teacher_disagreement_score":0.19090745,"about_ca_system_score_codex":0.013402027,"about_ca_system_score_gemma":0.022422764,"threshold_uncertainty_score":0.99775517},"labels":[],"label_agreement":null},{"id":"W2133783179","doi":"10.1177/1356389008101967","title":"Using Logic Analysis to Evaluate Knowledge Transfer Initiatives","year":2009,"lang":"en","type":"article","venue":"Evaluation","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"Canadian Institutes of Health Research; U.S. Public Health Service; Canadian Health Services Research Foundation","keywords":"Popularity; Knowledge transfer; Knowledge management; Process (computing); Politics; Order (exchange); Technology transfer; Logic model; Institutional logic; Management science; Computer science; Political science; Sociology; Engineering; Business; Social science; Public administration","score_opus":0.8591549218648961,"score_gpt":0.7677565598744323,"score_spread":0.0913983619904638,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133783179","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.78486586,0.0009326953,0.16712981,0.0015581901,0.00012229005,0.0041998876,0.0010882274,0.00036165852,0.03974142],"genre_scores_gemma":[0.85388297,0.00030898597,0.13934174,0.00019377087,0.000031505297,0.0046743494,0.0006037332,0.000050210154,0.000912535],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.8804599,0.08506446,0.0073819715,0.0026273255,0.022714732,0.0017516405],"domain_scores_gemma":[0.5857716,0.36820853,0.01658104,0.0066009024,0.021618292,0.001219607],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.10165584,0.00132294,0.0011981615,0.0124367215,0.0018309279,0.005547573,0.0017066151,0.0014743396,0.0047142194],"category_scores_gemma":[0.21989271,0.00035552893,0.0018662731,0.008913301,0.0040637488,0.0073475763,0.0037723838,0.0013660736,0.00037393215],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00942169,0.0036932535,0.12240378,0.007323072,0.0025150327,0.00054303074,0.020753056,0.070734516,0.010763038,0.17624483,0.0034928927,0.57211185],"study_design_scores_gemma":[0.0028917356,0.018204775,0.07977485,0.0024075103,0.0029400063,0.00049543293,0.03410289,0.4634137,0.059057914,0.31661078,0.019396026,0.00070435903],"about_ca_topic_score_codex":0.0022079707,"about_ca_topic_score_gemma":0.0018633336,"teacher_disagreement_score":0.10165584,"about_ca_system_score_codex":0.010218309,"about_ca_system_score_gemma":0.0054099755,"threshold_uncertainty_score":0.53761387},"labels":[],"label_agreement":null},{"id":"W2143790740","doi":"10.1177/1356389008090857","title":"Health Technology Assessment in the Canadian Health Policy Arena","year":2008,"lang":"en","type":"article","venue":"Evaluation","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université de Montréal","funders":"Canadian Institutes of Health Research; Canada Research Chairs","keywords":"Autonomy; Agency (philosophy); Accountability; Health technology; Public relations; Politics; Health care; Perspective (graphical); Public administration; Political science; Sociology; Social science","score_opus":0.45063542929334766,"score_gpt":0.5225140240842393,"score_spread":0.07187859479089165,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143790740","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13530888,0.022994293,0.005145581,0.27439454,0.00045029857,0.0004105945,0.0004998949,0.00006807432,0.56072783],"genre_scores_gemma":[0.9648448,0.008932916,0.003918286,0.0078101195,0.00012990164,0.00012952609,0.0001126403,0.000028671508,0.014093118],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.9557688,0.016414326,0.0010196071,0.0011129075,0.015868062,0.009816324],"domain_scores_gemma":[0.96860933,0.016871883,0.0014846158,0.000702684,0.01001036,0.0023212016],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027945999,0.0005773759,0.0006519141,0.00650171,0.034340035,0.019159866,0.0022644773,0.004891887,0.0049654697],"category_scores_gemma":[0.043636143,0.0005203317,0.00067182985,0.014413272,0.018695448,0.0043498185,0.006171641,0.0038936462,0.00021967171],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.00005471861,0.00008199749,0.015256463,0.0004675961,0.000047340516,0.0010943264,0.02749687,0.004569307,0.00037416964,0.86973214,0.025362547,0.05546251],"study_design_scores_gemma":[0.000051381117,0.00007501648,0.07679653,0.0012960864,0.000083627245,0.00028161312,0.124688685,0.004980687,0.00054113485,0.10980154,0.68118227,0.00022142952],"about_ca_topic_score_codex":0.9855832,"about_ca_topic_score_gemma":0.987186,"teacher_disagreement_score":0.673395,"about_ca_system_score_codex":0.32660502,"about_ca_system_score_gemma":0.37899223,"threshold_uncertainty_score":0.7810427},"labels":[],"label_agreement":null},{"id":"W2154807644","doi":"10.1177/1356389009341729","title":"Evaluating Service Organization Models","year":2009,"lang":"en","type":"article","venue":"Evaluation","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Agency of Canada; Université de Montréal; École Nationale d'Administration Publique","funders":"Canadian Institutes of Health Research; U.S. Public Health Service; Ministère de la Santé et des Services sociaux","keywords":"Computer science; Relevance (law); Set (abstract data type); Management science; Context (archaeology); Service (business); Contrast (vision); Data science; Artificial intelligence; Engineering","score_opus":0.8545316963181638,"score_gpt":0.762880564368063,"score_spread":0.09165113195010088,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2154807644","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.32065725,0.00239096,0.52290803,0.0065163174,0.0003226648,0.0019985489,0.0008882532,0.00047798952,0.14384],"genre_scores_gemma":[0.91293883,0.00048251176,0.084313735,0.00018250599,0.00004263704,0.0008531773,0.00020897613,0.000056555888,0.0009210405],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.8972409,0.085790224,0.0018558037,0.0022006982,0.011377172,0.0015352673],"domain_scores_gemma":[0.84103596,0.13242224,0.00572959,0.009249044,0.010457291,0.001105884],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.059133485,0.0014898026,0.0012297021,0.0044978927,0.0018109747,0.005884764,0.002127193,0.002162288,0.0065352954],"category_scores_gemma":[0.14689207,0.0005630734,0.0014690771,0.0050235926,0.005079167,0.0072616227,0.0042812726,0.001984963,0.00035801608],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000487331,0.00047460344,0.016972736,0.00091767637,0.0005593979,0.00015842218,0.0023624012,0.20465226,0.00076603977,0.6817762,0.0023283877,0.088544525],"study_design_scores_gemma":[0.000264462,0.0018050021,0.006501135,0.0009102803,0.00043387074,0.00010825024,0.0041333996,0.35614255,0.0028613985,0.61311084,0.013575458,0.00015340543],"about_ca_topic_score_codex":0.0053029587,"about_ca_topic_score_gemma":0.0049358467,"teacher_disagreement_score":0.059133485,"about_ca_system_score_codex":0.011285472,"about_ca_system_score_gemma":0.0061649326,"threshold_uncertainty_score":0.3127315},"labels":[],"label_agreement":null},{"id":"W2155601053","doi":"10.1177/1356389004046292","title":"What Counts is not Falling... but Landing","year":2004,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Douglas Mental Health University Institute; Douglas College","funders":"Canadian Institutes of Health Research; U.S. Public Health Service","keywords":"Strengths and weaknesses; Process management; Falling (accident); Process (computing); Management science; Computer science; Operations research; Political science; Business; Psychology; Engineering; Medicine; Environmental health","score_opus":0.36712162891403827,"score_gpt":0.545401205321608,"score_spread":0.1782795764075697,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2155601053","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027594019,0.026616888,0.048957177,0.46076447,0.020757444,0.00049637374,0.0016668249,0.00077996997,0.4123668],"genre_scores_gemma":[0.6693222,0.03929623,0.08001427,0.09431594,0.0054858034,0.0013212416,0.002035558,0.001281994,0.10692672],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.9749287,0.010318783,0.001563319,0.002707989,0.009180043,0.0013011938],"domain_scores_gemma":[0.9755397,0.008681223,0.0028886274,0.002381665,0.008224673,0.0022842707],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012828072,0.0012470774,0.0021178992,0.0033387996,0.006935574,0.014809238,0.0022377581,0.003294974,0.0139810145],"category_scores_gemma":[0.048383974,0.0004800576,0.0007347087,0.0055898535,0.023648513,0.023503395,0.0050527295,0.006341313,0.005440849],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012140962,0.00017336338,0.00976546,0.0026764073,0.00018595939,0.00019942949,0.014361924,0.0006834501,0.0013831268,0.40574318,0.17779209,0.38691416],"study_design_scores_gemma":[0.000022492548,0.00021378814,0.006975629,0.004411328,0.000107061634,0.00029376394,0.022392843,0.00051700603,0.0009399406,0.34211692,0.6218417,0.00016755812],"about_ca_topic_score_codex":0.014002357,"about_ca_topic_score_gemma":0.016054459,"teacher_disagreement_score":0.014809238,"about_ca_system_score_codex":0.006348735,"about_ca_system_score_gemma":0.014527685,"threshold_uncertainty_score":0.067842126},"labels":[],"label_agreement":null},{"id":"W2157111091","doi":"10.1177/1356389010380001","title":"To Be or Not to Be a Profession: Pros, Cons and Challenges for Evaluation","year":2010,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"École Nationale d'Administration Publique; Université Laval","funders":"","keywords":"Professionalization; Institutionalisation; Certification; Charter; Political science; Public relations; Accreditation; Valuation (finance); Quality (philosophy); Sociology; Engineering ethics; Public administration; Law; Accounting; Business; Engineering","score_opus":0.6053607172548194,"score_gpt":0.6135973810690398,"score_spread":0.00823666381422039,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2157111091","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017786942,0.08067156,0.02283814,0.8052026,0.0017645931,0.00018185466,0.000034860226,0.000077391705,0.07144204],"genre_scores_gemma":[0.86710733,0.037716094,0.029999427,0.055899955,0.0035371054,0.00064878794,0.000033688233,0.0001884666,0.0048691453],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.53826064,0.3771663,0.011099495,0.0063180155,0.06118645,0.005969012],"domain_scores_gemma":[0.3363493,0.597455,0.010676372,0.011089872,0.03664458,0.007784816],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.45002288,0.00069830584,0.002218258,0.007978109,0.01340846,0.04227686,0.003095166,0.011899442,0.0016933744],"category_scores_gemma":[0.3774643,0.00073239586,0.0010323584,0.007014369,0.10130494,0.030979222,0.016399825,0.013466807,0.0005342013],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001230203,0.00010767899,0.005493326,0.0016972116,0.0000507318,0.0001590652,0.039452225,0.0006025484,0.00016693577,0.61236626,0.017196318,0.32258472],"study_design_scores_gemma":[0.00007153099,0.00019575216,0.0076897894,0.011591237,0.000068804526,0.00060052506,0.1323897,0.002404023,0.0005465635,0.6639637,0.18024644,0.00023195476],"about_ca_topic_score_codex":0.012949975,"about_ca_topic_score_gemma":0.017188547,"teacher_disagreement_score":0.45002288,"about_ca_system_score_codex":0.022223061,"about_ca_system_score_gemma":0.03881035,"threshold_uncertainty_score":0.67821974},"labels":[],"label_agreement":null},{"id":"W2171507080","doi":"10.1177/1356389011430371","title":"Evaluation models and evaluation use","year":2012,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":59,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Sherbrooke; Université de Montréal","funders":"Canadian Institutes of Health Research; U.S. Public Health Service","keywords":"Context (archaeology); Management science; Psychological intervention; Process (computing); Perspective (graphical); Field (mathematics); Evaluation methods; Impact evaluation; Systematic review; Affect (linguistics); Program evaluation; Computer science; Knowledge management; Process management; Psychology; Political science; Engineering; Artificial intelligence; Medicine; MEDLINE","score_opus":0.6626932025379391,"score_gpt":0.5990196685991209,"score_spread":0.06367353393881814,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2171507080","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011933271,0.027830807,0.63591367,0.055200666,0.0014540763,0.00227558,0.0005699868,0.0006620222,0.26415995],"genre_scores_gemma":[0.5661375,0.014625835,0.3948479,0.0059176707,0.00137693,0.006570905,0.00047345695,0.000369529,0.009680359],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6575161,0.28604418,0.016530424,0.008882196,0.028218009,0.0028089702],"domain_scores_gemma":[0.5127546,0.42227402,0.015722305,0.020130116,0.02688184,0.0022371116],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.21811213,0.0021421656,0.0022565222,0.0120012425,0.0033490832,0.020914659,0.0039469586,0.0061313994,0.007887036],"category_scores_gemma":[0.31215215,0.0011601847,0.0019412054,0.008572827,0.023895914,0.022778196,0.0073315394,0.0048001898,0.0013175242],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000040660158,0.000044341392,0.0011191156,0.00082955376,0.000099794495,0.000055512384,0.0016259224,0.0025244432,0.00005498298,0.9558261,0.0029778902,0.034801777],"study_design_scores_gemma":[0.00007330798,0.00008249567,0.0006350555,0.002016814,0.00007427326,0.00012171947,0.0010605067,0.0054169493,0.00031108523,0.940789,0.049357314,0.000061510196],"about_ca_topic_score_codex":0.0034015342,"about_ca_topic_score_gemma":0.0019283318,"teacher_disagreement_score":0.21811213,"about_ca_system_score_codex":0.014985848,"about_ca_system_score_gemma":0.012509117,"threshold_uncertainty_score":0.96420693},"labels":[],"label_agreement":null},{"id":"W2321004893","doi":"10.1177/1356389014540705","title":"Evaluations, international agencies and censorship: A field doer’s viewpoint","year":2014,"lang":"en","type":"article","venue":"Evaluation","topic":"Qualitative Research Methods and Ethics","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Field (mathematics); Sociology; Action (physics); Perspective (graphical); Epistemology; Computer science; Mathematics","score_opus":0.27986036555218136,"score_gpt":0.6030785631938922,"score_spread":0.32321819764171084,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2321004893","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06125787,0.030389775,0.034152616,0.50367326,0.0018189457,0.00026039497,0.00007169317,0.000091304195,0.3682841],"genre_scores_gemma":[0.96486783,0.004663484,0.0040253154,0.017202768,0.00055035646,0.0001704933,0.000011961208,0.000053035008,0.008454713],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.923418,0.06761998,0.0009735871,0.0024558222,0.0038413014,0.001691269],"domain_scores_gemma":[0.9173207,0.07119737,0.0025721113,0.0038508598,0.0038988355,0.0011601007],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05928275,0.0006451962,0.00085381296,0.0035194347,0.010480256,0.017408006,0.002029596,0.006229768,0.0032196566],"category_scores_gemma":[0.05803699,0.00045254143,0.00037211887,0.0022281979,0.100017086,0.021837788,0.007474901,0.008751724,0.00024491348],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009551759,0.000025628962,0.00034135566,0.00011375739,0.0000042116117,0.00010642216,0.05677416,0.00011080273,0.00006624547,0.9293451,0.004669436,0.0084333345],"study_design_scores_gemma":[0.00003719447,0.00007976419,0.00063180533,0.0016633517,0.000016422595,0.00033063642,0.2172892,0.00061477587,0.0005390051,0.533607,0.24515143,0.000039409017],"about_ca_topic_score_codex":0.013076328,"about_ca_topic_score_gemma":0.009411722,"teacher_disagreement_score":0.05928275,"about_ca_system_score_codex":0.01497435,"about_ca_system_score_gemma":0.009754621,"threshold_uncertainty_score":0.3135209},"labels":[],"label_agreement":null},{"id":"W2324664192","doi":"10.1177/1356389014564248","title":"The institutionalization of evaluation matters: Updating the <i>International Atlas of Evaluation</i> 10 years later","year":2015,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":114,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Institutionalisation; Political science; Regional science; Public administration; Geography; Law","score_opus":0.2102689892953537,"score_gpt":0.4898436435508309,"score_spread":0.2795746542554772,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2324664192","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02103524,0.20753466,0.029539097,0.52184224,0.06931586,0.00018590988,0.0047516217,0.0008259081,0.1449695],"genre_scores_gemma":[0.35571602,0.33597368,0.09988916,0.099599764,0.043334063,0.0007230941,0.010465785,0.002110677,0.052187815],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9763464,0.009292659,0.0041863057,0.0013490246,0.0071641174,0.0016614617],"domain_scores_gemma":[0.8875396,0.040835634,0.00913961,0.01132145,0.04796488,0.0031988036],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03410174,0.0006289698,0.00075902324,0.013752834,0.0022705616,0.016767738,0.0016349349,0.002262572,0.0043923636],"category_scores_gemma":[0.07674022,0.00033845936,0.0011046792,0.025225153,0.0070216926,0.024264283,0.0050988677,0.0055394736,0.0017534897],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007330417,0.00006888345,0.008193092,0.0012287305,0.00003634624,0.00008828385,0.006286728,0.00070177944,0.00050478295,0.14528847,0.44589725,0.39163232],"study_design_scores_gemma":[0.0000032473324,0.000019458797,0.00859359,0.0016901075,0.000016718426,0.000119407676,0.005427285,0.00021382877,0.0003560998,0.010856579,0.9726571,0.000046644473],"about_ca_topic_score_codex":0.023527319,"about_ca_topic_score_gemma":0.029300652,"teacher_disagreement_score":0.9658983,"about_ca_system_score_codex":0.01288418,"about_ca_system_score_gemma":0.015002983,"threshold_uncertainty_score":0},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W2605706618","doi":"10.1177/1356389017697620","title":"Evaluability assessment of a small NGO in water-based development","year":2017,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Waterloo","funders":"Mitacs","keywords":"Accountability; Context (archaeology); Political science; Public relations; Water development; Face (sociological concept); Qualitative research; Baseline (sea); Psychology; Sociology; Water resources; Geography; Social science","score_opus":0.4462252381466189,"score_gpt":0.5827030978504076,"score_spread":0.13647785970378873,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2605706618","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.83382714,0.0010520912,0.06588591,0.011221514,0.00021001366,0.011864076,0.00023796032,0.00013604149,0.07556521],"genre_scores_gemma":[0.97036386,0.00023406302,0.022955209,0.0005180095,0.000029400988,0.0041176337,0.00009634901,0.000045125154,0.0016403873],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.69410044,0.25325117,0.012141096,0.0051458594,0.027792258,0.0075691296],"domain_scores_gemma":[0.5683558,0.29229105,0.026843525,0.018267216,0.08294067,0.011301702],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.29702115,0.0006463114,0.0006238222,0.004436094,0.008569509,0.008420805,0.0023052655,0.0015355685,0.0031225695],"category_scores_gemma":[0.34518138,0.0004760814,0.0006635579,0.0032708817,0.008914454,0.0070521794,0.010156864,0.0018279256,0.0003238453],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013810945,0.0023364888,0.097398974,0.004667419,0.00021730234,0.002097357,0.50763196,0.0044082017,0.0075299772,0.03184525,0.0053571686,0.33512887],"study_design_scores_gemma":[0.0005626586,0.004919843,0.11655398,0.005615021,0.00035021952,0.00082681794,0.6634664,0.010267944,0.019357128,0.04074822,0.13698013,0.0003515466],"about_ca_topic_score_codex":0.0112557635,"about_ca_topic_score_gemma":0.01663321,"teacher_disagreement_score":0.29702115,"about_ca_system_score_codex":0.02323461,"about_ca_system_score_gemma":0.03236695,"threshold_uncertainty_score":0.86689806},"labels":[],"label_agreement":null},{"id":"W2606109332","doi":"10.1177/1356389017697615","title":"Network-based approaches for evaluating ambient assisted living (AAL) technologies","year":2017,"lang":"en","type":"article","venue":"Evaluation","topic":"Technology Use by Older Adults","field":"Social Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University; University of Toronto","funders":"Economic and Social Research Council","keywords":"Assisted living; Perspective (graphical); Activities of daily living; Computer science; Cognition; Bridge (graph theory); Independent living; Function (biology); Living lab; Human–computer interaction; Knowledge management; Psychology; Gerontology; Medicine; Artificial intelligence","score_opus":0.19563511380443538,"score_gpt":0.399164352199533,"score_spread":0.20352923839509762,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2606109332","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4378564,0.004914498,0.38691625,0.00491838,0.0003879538,0.015203898,0.001065815,0.00015883752,0.14857791],"genre_scores_gemma":[0.8777741,0.00085284445,0.10535719,0.00022735959,0.000034399603,0.013892858,0.00014727112,0.000019413434,0.0016945027],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9061307,0.08210578,0.003520038,0.0016374694,0.0058208164,0.0007851881],"domain_scores_gemma":[0.9130085,0.06896514,0.0060602985,0.002180026,0.0086804135,0.0011055628],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.046907816,0.0012775175,0.0007200523,0.0061956877,0.002092725,0.004704691,0.0013977867,0.0013203418,0.004413085],"category_scores_gemma":[0.0871791,0.00037058207,0.0011598101,0.003621959,0.0034569667,0.0053378455,0.0034779017,0.0011772049,0.00029861482],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017093915,0.0022432678,0.09600324,0.008763309,0.001419193,0.00058842654,0.08170696,0.025094194,0.0046066283,0.3498002,0.0049706222,0.42309466],"study_design_scores_gemma":[0.0011360014,0.0075114747,0.12201836,0.008252588,0.0017467671,0.00095232454,0.21526007,0.15908429,0.012828045,0.3782121,0.09238026,0.000617787],"about_ca_topic_score_codex":0.003900943,"about_ca_topic_score_gemma":0.0060557937,"teacher_disagreement_score":0.046907816,"about_ca_system_score_codex":0.0074121365,"about_ca_system_score_gemma":0.0025242765,"threshold_uncertainty_score":0.24807519},"labels":[],"label_agreement":null},{"id":"W2742194300","doi":"10.1177/1356389017715719","title":"Using online mining techniques to inform formative evaluations: An analysis of YouTube video comments about chronic pain","year":2017,"lang":"en","type":"article","venue":"Evaluation","topic":"Focus Groups and Qualitative Methods","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Economic and Social Research Council; Centre National de la Recherche Scientifique; McGill University","keywords":"Judgement; Formative assessment; Social media; Psychology; Applied psychology; Computer science; Public relations; World Wide Web; Political science","score_opus":0.3269929441657317,"score_gpt":0.5957748917242113,"score_spread":0.26878194755847956,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2742194300","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97251093,0.00040130096,0.016989917,0.0010489393,0.000076150594,0.0014194428,0.0016576176,0.00008473221,0.0058109662],"genre_scores_gemma":[0.95734966,0.00046269127,0.03492519,0.00024027812,0.00009138456,0.0022457754,0.0017670366,0.00009768965,0.0028202084],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9816097,0.01277817,0.0012832519,0.0008004969,0.0028717066,0.0006566485],"domain_scores_gemma":[0.8739744,0.10025244,0.008609301,0.0025211056,0.014009177,0.0006336645],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018383125,0.0006639748,0.0005330215,0.0084111085,0.002040406,0.0025949257,0.00091681414,0.00080415374,0.0013876165],"category_scores_gemma":[0.08031431,0.00022627252,0.0003552875,0.0062126457,0.0016495513,0.0032108282,0.002036926,0.0006845522,0.0004433678],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00064601714,0.0005156174,0.1186904,0.0018215617,0.00011643227,0.0017353016,0.5573323,0.0006258583,0.010510095,0.0034956394,0.0051918984,0.29931885],"study_design_scores_gemma":[0.000049975384,0.00069439656,0.29061246,0.0019092273,0.000110706285,0.0010477839,0.6464072,0.011756958,0.010103253,0.006255869,0.030860296,0.00019199023],"about_ca_topic_score_codex":0.005431764,"about_ca_topic_score_gemma":0.0119616445,"teacher_disagreement_score":0.018383125,"about_ca_system_score_codex":0.00221206,"about_ca_system_score_gemma":0.0019352086,"threshold_uncertainty_score":0.09722042},"labels":[],"label_agreement":null},{"id":"W2800395496","doi":"10.1177/1356389018763242","title":"The evaluation of social innovation: A review and integration of the current empirical knowledge base","year":2018,"lang":"en","type":"review","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":75,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Empirical research; Knowledge management; Knowledge base; Social innovation; Scale (ratio); Social learning; Perspective (graphical); Public relations; Computer science; Political science","score_opus":0.735361020503894,"score_gpt":0.6846952553233857,"score_spread":0.05066576518050825,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2800395496","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00030617567,0.99808216,0.0003514532,0.00044320885,0.00008042469,0.000029375418,0.000027971926,0.0000045799334,0.0006745398],"genre_scores_gemma":[0.0044263056,0.9943032,0.00079477247,0.00021269103,0.00009499476,0.000057734942,0.000039358893,0.000005106753,0.00006590821],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9845803,0.006508444,0.0036805845,0.0011432739,0.0037441934,0.00034327715],"domain_scores_gemma":[0.857755,0.1262142,0.0053807213,0.0017152412,0.008211971,0.0007228453],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02333668,0.0013906752,0.0041464544,0.016396664,0.0009389213,0.0045441594,0.0021740184,0.002329042,0.0034860673],"category_scores_gemma":[0.061837863,0.00093128585,0.0021228206,0.018113242,0.0029253254,0.0046537756,0.0024009289,0.0019898063,0.0005159995],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000095595526,0.000086301945,0.0011251137,0.23998241,0.0005982105,0.00012456409,0.00087112345,0.00048424004,0.00029540056,0.007730629,0.0055162995,0.74309015],"study_design_scores_gemma":[0.00004616699,0.0002443557,0.006436808,0.55128974,0.0032488906,0.00084314856,0.0018599471,0.00050572807,0.0008477698,0.010302963,0.42425877,0.00011564042],"about_ca_topic_score_codex":0.00524851,"about_ca_topic_score_gemma":0.008880314,"teacher_disagreement_score":0.02333668,"about_ca_system_score_codex":0.004908866,"about_ca_system_score_gemma":0.01402998,"threshold_uncertainty_score":0.123417616},"labels":[],"label_agreement":null},{"id":"W2801790340","doi":"10.1177/1356389018765487","title":"Theory-based evaluations: Framing the existence of a new theory in evaluation and the rise of the 5th generation","year":2018,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":57,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre Intégré Universitaire de Santé et de Services Sociaux du Saguenay–Lac-Saint-Jean; Université de Sherbrooke; University of Victoria","funders":"","keywords":"Epistemology; Framing (construction); Complementarity (molecular biology); Realism; Sociology; Computer science; Philosophy","score_opus":0.28830792381595954,"score_gpt":0.5359311604701663,"score_spread":0.24762323665420677,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2801790340","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04346855,0.029804766,0.40552318,0.19181998,0.003457002,0.0004191272,0.000094826646,0.00023761608,0.3251749],"genre_scores_gemma":[0.9022131,0.0059331986,0.07659367,0.007487084,0.0010445318,0.00053527014,0.000053691714,0.00013703562,0.0060023703],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.961203,0.02908615,0.0011095693,0.002073773,0.005136043,0.001391524],"domain_scores_gemma":[0.94309974,0.04292735,0.0026576547,0.0038671414,0.005511049,0.0019371447],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.056127205,0.0011681736,0.001061493,0.009611467,0.0063239634,0.018360147,0.0027672458,0.007123732,0.0033946033],"category_scores_gemma":[0.041600417,0.0007108085,0.0009811892,0.0037675805,0.08653028,0.024104992,0.00905633,0.009564678,0.00043205716],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000065720265,0.00000676817,0.00013742756,0.00003789789,0.000002742671,0.000020424774,0.003975489,0.000102174636,0.000024172583,0.9906763,0.00040815733,0.0046019647],"study_design_scores_gemma":[0.000011515715,0.000026118656,0.00022562324,0.00036612386,0.0000076182932,0.00005938372,0.003169345,0.0011135844,0.00014811472,0.97166145,0.023194278,0.000016864951],"about_ca_topic_score_codex":0.004042775,"about_ca_topic_score_gemma":0.0027846803,"teacher_disagreement_score":0.9438728,"about_ca_system_score_codex":0.019386759,"about_ca_system_score_gemma":0.009844198,"threshold_uncertainty_score":0.29683256},"labels":[],"label_agreement":null},{"id":"W2913663231","doi":"10.1177/1356389019827035","title":"Combining internal and external evaluations within a multilevel evaluation framework: Computational text analysis of lessons from the Asian Development Bank","year":2019,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"Lee Kuan Yew School of Public Policy, National University of Singapore; Whitney and Betty MacMillan Center for International and Area Studies; Yale University","keywords":"Macro; Psychological intervention; Micro level; Scenario analysis; Political science; Computer science; Process management; Psychology; Business; Economics; Finance; Microeconomics","score_opus":0.23131795111349623,"score_gpt":0.5268569038509368,"score_spread":0.2955389527374405,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2913663231","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9185853,0.0008849551,0.05880079,0.0037989893,0.000048462985,0.00066200853,0.0018963317,0.00028862103,0.015034518],"genre_scores_gemma":[0.9768325,0.00012383141,0.021481011,0.000070900816,0.000023569355,0.00022728412,0.0007707809,0.00003127442,0.00043882604],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.97633535,0.019013228,0.0013237274,0.00076437736,0.002112226,0.0004511756],"domain_scores_gemma":[0.8229765,0.15126517,0.00908312,0.003994525,0.011811471,0.0008692218],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.022033785,0.00048286814,0.0005962981,0.0069777346,0.00082470605,0.004298218,0.0008499399,0.0006279459,0.0018165403],"category_scores_gemma":[0.106103174,0.00018307604,0.00065236806,0.0064876396,0.0013795188,0.00505839,0.0025839466,0.0014282893,0.00022411188],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006427816,0.00076875224,0.2679621,0.0016563461,0.0006687848,0.00046248137,0.031859234,0.03934302,0.0015462419,0.026015053,0.006473978,0.6226013],"study_design_scores_gemma":[0.00021442636,0.0009781319,0.27304044,0.0019504201,0.000672844,0.0001777228,0.07413198,0.49635765,0.009904895,0.12267816,0.019607672,0.0002856523],"about_ca_topic_score_codex":0.0069511747,"about_ca_topic_score_gemma":0.011938746,"teacher_disagreement_score":0.9779662,"about_ca_system_score_codex":0.0036609804,"about_ca_system_score_gemma":0.002150115,"threshold_uncertainty_score":0.11652714},"labels":[],"label_agreement":null},{"id":"W2973640600","doi":"10.1177/1356389019870213","title":"Rapid impact evaluation","year":2019,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Summative assessment; Impact evaluation; Formative assessment; Counterfactual thinking; Credibility; Stakeholder; Stakeholder engagement; Impact assessment; Ex-ante; Salience (neuroscience); Program evaluation; Theory of change; Process management; Legitimacy; Computer science; Management science; Knowledge management; Psychology; Business; Political science; Public relations; Economics; Social psychology; Artificial intelligence","score_opus":0.3381382875021545,"score_gpt":0.5894724107344181,"score_spread":0.2513341232322636,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2973640600","genre_codex":"other","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01301791,0.004758947,0.21186943,0.011908615,0.0069468906,0.07462343,0.047750358,0.011937294,0.6171871],"genre_scores_gemma":[0.16660388,0.010524195,0.47018757,0.006306546,0.0019692343,0.13051403,0.03862126,0.0062144827,0.1690588],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.8959323,0.054467294,0.009642656,0.0037670026,0.03342541,0.002765273],"domain_scores_gemma":[0.6747469,0.101797,0.011734463,0.02834832,0.1780872,0.0052861297],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.119456105,0.0026444283,0.0022175277,0.010986332,0.0028248604,0.008377989,0.004698825,0.0027120446,0.19045697],"category_scores_gemma":[0.25054088,0.001076949,0.0031853346,0.0073625175,0.0017676203,0.007551064,0.008252852,0.0034523685,0.030281542],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016050546,0.0004897519,0.0028195754,0.008061211,0.00027316427,0.00023141954,0.001492261,0.0032433092,0.0009008144,0.041433435,0.34979934,0.5896505],"study_design_scores_gemma":[0.0005530839,0.00093560293,0.003669679,0.0064430484,0.00032555393,0.00017352098,0.0013325611,0.003134979,0.002793478,0.025079252,0.9553778,0.00018151179],"about_ca_topic_score_codex":0.003510656,"about_ca_topic_score_gemma":0.0043269764,"teacher_disagreement_score":0.19045697,"about_ca_system_score_codex":0.005884175,"about_ca_system_score_gemma":0.026714193,"threshold_uncertainty_score":0.63714206},"labels":[],"label_agreement":null},{"id":"W3020197475","doi":"10.1177/1356389020911060","title":"Evaluators in the Anthropocene","year":2020,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria; Université de Sherbrooke","funders":"","keywords":"Anthropocene; Biosphere; Environmental ethics; Action (physics); State (computer science); Natural (archaeology); Call to action; Earth system science; Work (physics); Non-human; Environmental planning; Environmental resource management; Political science; Ecology; History; Geography; Archaeology; Environmental science; Business; Law; Computer science; Biology; Engineering","score_opus":0.4706248224375595,"score_gpt":0.5890246793657072,"score_spread":0.11839985692814775,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3020197475","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07755572,0.047485765,0.0777216,0.4887574,0.005185091,0.0010580011,0.0001467578,0.0003717591,0.30171785],"genre_scores_gemma":[0.87231296,0.019278584,0.043494534,0.0352536,0.0023405952,0.0014750866,0.00011297063,0.00031017442,0.02542163],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6882272,0.28010762,0.006240508,0.00458375,0.015938804,0.004902114],"domain_scores_gemma":[0.73449445,0.18353356,0.012466437,0.008329009,0.047171976,0.014004554],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.200548,0.0009497775,0.0011125152,0.004101818,0.007989064,0.018856145,0.0019030274,0.003782603,0.0080542015],"category_scores_gemma":[0.24691017,0.0004148629,0.0005060859,0.0037171922,0.02165112,0.014788825,0.012365876,0.006504741,0.0013871322],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00043677987,0.00041885927,0.010992543,0.0019337044,0.000112153495,0.00036980122,0.15646133,0.002135282,0.0009558141,0.43032607,0.08589153,0.30996606],"study_design_scores_gemma":[0.00010666898,0.00041074323,0.005026236,0.00579526,0.000065037944,0.00024832544,0.12511535,0.002349137,0.001352774,0.23581572,0.62355924,0.00015550335],"about_ca_topic_score_codex":0.0029398815,"about_ca_topic_score_gemma":0.0038559402,"teacher_disagreement_score":0.799452,"about_ca_system_score_codex":0.011543928,"about_ca_system_score_gemma":0.015140758,"threshold_uncertainty_score":0.98586667},"labels":[],"label_agreement":null},{"id":"W3095230013","doi":"10.1177/1356389020952462","title":"Evaluation for planetary health","year":2020,"lang":"en","type":"article","venue":"Evaluation","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Dialogic; Psychological intervention; Field (mathematics); Management science; Human health; Natural (archaeology); Work (physics); Engineering ethics; Sociology; Computer science; Knowledge management; Psychology; Engineering; Medicine; Geography","score_opus":0.6745489268580942,"score_gpt":0.510194529002474,"score_spread":0.16435439785562023,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3095230013","genre_codex":"other","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022316063,0.13386966,0.2945984,0.2097017,0.010143403,0.016871579,0.0059945513,0.0008683128,0.30563623],"genre_scores_gemma":[0.66743904,0.035953667,0.22697161,0.023225581,0.0030265786,0.03030719,0.0022574884,0.0003084004,0.010510538],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6903457,0.28189883,0.008861837,0.002868768,0.014023457,0.0020014714],"domain_scores_gemma":[0.64024234,0.30150977,0.016460141,0.014650337,0.023327753,0.003809622],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17930931,0.001213807,0.0022560712,0.0044545247,0.0021198895,0.007611234,0.0017252382,0.003601419,0.026213756],"category_scores_gemma":[0.3905882,0.0004358318,0.002149617,0.0037001777,0.0066503673,0.006615499,0.0065841777,0.0037291113,0.0013840247],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017758873,0.00040080852,0.0065158205,0.012299262,0.0015664821,0.000111451605,0.0013143279,0.005635554,0.00016344435,0.56162477,0.049613204,0.35897896],"study_design_scores_gemma":[0.0015364224,0.0020536873,0.01056554,0.02978447,0.0016459045,0.00033999493,0.0030308403,0.0122116245,0.00097543304,0.70187867,0.23581456,0.0001628889],"about_ca_topic_score_codex":0.0042920457,"about_ca_topic_score_gemma":0.0032539046,"teacher_disagreement_score":0.17930931,"about_ca_system_score_codex":0.012341022,"about_ca_system_score_gemma":0.02240022,"threshold_uncertainty_score":0.9482896},"labels":[],"label_agreement":null},{"id":"W3100567947","doi":"10.1177/1356389020969714","title":"Applying the eight principles of developmental evaluation: Complex Care Hub as a case example","year":2020,"lang":"en","type":"article","venue":"Evaluation","topic":"Geriatric Care and Nursing Homes","field":"Health Professions","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary; Alberta Health Services","funders":"Alberta Health Services","keywords":"Context (archaeology); Health care; Psychology; Nursing; Medical education; Medicine; Political science","score_opus":0.33880428037014854,"score_gpt":0.46695976281898927,"score_spread":0.12815548244884073,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3100567947","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.34120893,0.017794577,0.20757951,0.15225357,0.0017351804,0.005508426,0.0001940704,0.00037546008,0.2733503],"genre_scores_gemma":[0.80467814,0.009889373,0.15680312,0.0075443923,0.00023638511,0.0020196512,0.00005939806,0.00012572196,0.018643798],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.9695358,0.025601998,0.0008186268,0.00041205712,0.002042072,0.0015894364],"domain_scores_gemma":[0.9838976,0.011988949,0.00053605903,0.0005604483,0.001695408,0.0013215195],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.028597474,0.0007205176,0.00052491244,0.0018892544,0.0076433206,0.006014059,0.0015026955,0.0034469373,0.0022508833],"category_scores_gemma":[0.026409093,0.00044302587,0.0007911819,0.0017018068,0.008424123,0.005575273,0.008399372,0.0051987604,0.0003489138],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028818066,0.0021271708,0.027474813,0.0023817525,0.00004806576,0.033909164,0.24120112,0.004033178,0.001563449,0.3283758,0.050361842,0.30823547],"study_design_scores_gemma":[0.0002378538,0.0015288083,0.013961205,0.008020423,0.000120476965,0.05152293,0.3235812,0.006567235,0.0063187126,0.09540018,0.49243566,0.00030531397],"about_ca_topic_score_codex":0.0061443844,"about_ca_topic_score_gemma":0.02368204,"teacher_disagreement_score":0.028597474,"about_ca_system_score_codex":0.009183618,"about_ca_system_score_gemma":0.011013394,"threshold_uncertainty_score":0.1512397},"labels":[],"label_agreement":null},{"id":"W3109011900","doi":"10.1177/1356389020969721","title":"How to normalize reflexive evaluation? Navigating between legitimacy and integrity","year":2020,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Athena Sustainable Materials Institute","funders":"","keywords":"Reflexivity; Normalization (sociology); Legitimacy; Epistemology; Process management; Political science; Sociology; Computer science; Knowledge management; Business; Social science; Politics","score_opus":0.4709169939777773,"score_gpt":0.5835940233142233,"score_spread":0.11267702933644597,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3109011900","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05000588,0.002781208,0.63760465,0.09111087,0.00082570134,0.0007708202,0.00006464371,0.0007264045,0.21610983],"genre_scores_gemma":[0.86407894,0.00076411816,0.12344603,0.0035648588,0.00021969905,0.00085690484,0.000039727514,0.00039274397,0.006636995],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.79009694,0.14481604,0.010117671,0.011620995,0.037885986,0.0054623596],"domain_scores_gemma":[0.7119609,0.17753115,0.021221288,0.046216074,0.038780358,0.004290233],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.18215804,0.000946833,0.0017924451,0.0050682705,0.007684204,0.029247569,0.0035524992,0.0044267094,0.0042018467],"category_scores_gemma":[0.30011588,0.0009048432,0.0011382992,0.004463451,0.07757322,0.037815787,0.0152475685,0.0100909015,0.00093312387],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000028581548,0.00004458386,0.0013062787,0.00016021139,0.000029194922,0.000044349854,0.013851622,0.0010037106,0.00036016948,0.93021303,0.0012514404,0.051706802],"study_design_scores_gemma":[0.00002763244,0.00003774617,0.00056416733,0.00051460735,0.000018549457,0.000053416246,0.005714875,0.0028410903,0.0016801709,0.96389526,0.02460365,0.00004880587],"about_ca_topic_score_codex":0.0042004865,"about_ca_topic_score_gemma":0.0031312034,"teacher_disagreement_score":0.81784195,"about_ca_system_score_codex":0.017055882,"about_ca_system_score_gemma":0.030438129,"threshold_uncertainty_score":0.9633553},"labels":[],"label_agreement":null},{"id":"W3139167773","doi":"10.1177/1356389020978501","title":"Developing an ethical rationale for collaborative approaches to evaluation","year":2021,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; University of Victoria","funders":"","keywords":"Reflexivity; Dialogic; Reciprocity (cultural anthropology); Inclusion (mineral); Sociology; Engineering ethics; Context (archaeology); Politics; Set (abstract data type); Epistemology; Political science; Computer science; Pedagogy; Social science; Engineering; Law","score_opus":0.8319796326770158,"score_gpt":0.5935660075730588,"score_spread":0.23841362510395703,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3139167773","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008616904,0.001531438,0.7389443,0.144749,0.0014012717,0.0018379099,0.00007750651,0.00015501797,0.10268663],"genre_scores_gemma":[0.488133,0.0009993988,0.47343135,0.019923368,0.00094489154,0.006530928,0.00009199064,0.00017787567,0.009767199],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5856849,0.326994,0.01795623,0.014201597,0.04954304,0.0056201587],"domain_scores_gemma":[0.6512765,0.25159207,0.013915262,0.030450428,0.04628352,0.0064821597],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2898375,0.001199052,0.001680309,0.00435949,0.012830922,0.02147766,0.0061656395,0.021583207,0.003134648],"category_scores_gemma":[0.25429597,0.0011992148,0.0023449145,0.002277789,0.08857335,0.021369347,0.018899372,0.02256438,0.001446932],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000004683843,0.00001170287,0.0000691656,0.000033042394,0.0000037458776,0.000029108463,0.0028391408,0.00019365866,0.0000607289,0.9940614,0.0007908812,0.001902594],"study_design_scores_gemma":[0.000036108017,0.00003237795,0.000103305945,0.0004594774,0.00001074824,0.00011249572,0.0023740593,0.0019754672,0.00047083228,0.9398475,0.054533355,0.000044200406],"about_ca_topic_score_codex":0.0036993814,"about_ca_topic_score_gemma":0.0028903247,"teacher_disagreement_score":0.7101625,"about_ca_system_score_codex":0.012777279,"about_ca_system_score_gemma":0.029269978,"threshold_uncertainty_score":0.8757568},"labels":[],"label_agreement":null},{"id":"W4220896026","doi":"10.1177/13563890221086306","title":"Contribution of practical wisdom to resolving ethical issues","year":2022,"lang":"en","type":"article","venue":"Evaluation","topic":"Ethics in medical practice","field":"Health Professions","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Montréal","funders":"","keywords":"Engineering ethics; Action (physics); Face (sociological concept); Ethical issues; Epistemology; Management science; Sociology; Psychology; Social science; Economics; Philosophy; Engineering","score_opus":0.22857419154794673,"score_gpt":0.6319014456775595,"score_spread":0.40332725412961273,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4220896026","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02738413,0.016489316,0.24368404,0.432162,0.0035922797,0.00056025066,0.000043494038,0.00014513257,0.27593935],"genre_scores_gemma":[0.81026477,0.00668244,0.14536743,0.027783249,0.0022661206,0.00068859814,0.00004338399,0.00012762177,0.0067763506],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8127723,0.14787167,0.0063177356,0.005240327,0.02445699,0.0033410604],"domain_scores_gemma":[0.6535211,0.28386015,0.01180536,0.021502266,0.023438003,0.0058730994],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.14523904,0.0013594719,0.0020190168,0.00808908,0.009989209,0.021330062,0.00372243,0.013330415,0.0032662584],"category_scores_gemma":[0.17677546,0.0007924727,0.0014789194,0.003409848,0.09083148,0.022679143,0.020942563,0.014573633,0.00063086185],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000020011874,0.0000551469,0.0007266469,0.0005078552,0.000046515644,0.000274709,0.018690998,0.00075381366,0.00014463523,0.93690753,0.004818005,0.037054073],"study_design_scores_gemma":[0.000024752917,0.000023832383,0.00020308844,0.0006641547,0.000015452295,0.00020546318,0.00611612,0.0008848414,0.000117011565,0.9599169,0.031792715,0.000035625406],"about_ca_topic_score_codex":0.0018830479,"about_ca_topic_score_gemma":0.0029222644,"teacher_disagreement_score":0.14523904,"about_ca_system_score_codex":0.012158963,"about_ca_system_score_gemma":0.03112759,"threshold_uncertainty_score":0.76810664},"labels":[],"label_agreement":null},{"id":"W4232707640","doi":"10.1177/1356389015580672","title":"Canadian Evaluation Society 36th Conference 2015: <i>Evaluation for the world we want</i>","year":2015,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Political science; Engineering ethics; Environmental ethics; Engineering; Philosophy","score_opus":0.5302600051866425,"score_gpt":0.5674058189989668,"score_spread":0.03714581381232429,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4232707640","genre_codex":"commentary","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0023060679,0.06168627,0.011952974,0.6586597,0.042820096,0.0014058219,0.0039196345,0.00074794336,0.21650149],"genre_scores_gemma":[0.075125046,0.07726471,0.049825072,0.14907362,0.015737973,0.0016711762,0.009641533,0.0020667447,0.6195941],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.904609,0.017633073,0.0065642893,0.0025472168,0.06000881,0.008637574],"domain_scores_gemma":[0.83367586,0.018532906,0.004577049,0.005965551,0.1145053,0.022743378],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.12508534,0.002200286,0.002950222,0.009027495,0.011851999,0.029311532,0.0067828577,0.019766515,0.026683474],"category_scores_gemma":[0.115095824,0.001803854,0.002806869,0.009114055,0.01272187,0.00687449,0.007977991,0.014371895,0.009610791],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000032711483,0.00004401924,0.0004173278,0.0003006222,0.000024141289,0.000030430032,0.00029955216,0.00028686968,0.00018063953,0.020603584,0.95313275,0.02464744],"study_design_scores_gemma":[0.0000161723,0.000014140049,0.001878767,0.0013578492,0.000025494213,0.000023704099,0.0005737602,0.00030073704,0.00030199823,0.008771777,0.98663735,0.00009824561],"about_ca_topic_score_codex":0.81071144,"about_ca_topic_score_gemma":0.8918662,"teacher_disagreement_score":0.9013859,"about_ca_system_score_codex":0.098614104,"about_ca_system_score_gemma":0.42592585,"threshold_uncertainty_score":0.71549875},"labels":[],"label_agreement":null},{"id":"W4236952314","doi":"10.1177/1356389014543594","title":"The 35 <sup>th</sup> Canadian Evaluation Society Annual Conference: ‘Celebrating Contributions to Canadian Evaluation’","year":2014,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Political science; Library science; Computer science","score_opus":0.11994209496311921,"score_gpt":0.48091336084166886,"score_spread":0.36097126587854966,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4236952314","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0006576053,0.023981016,0.0017453878,0.83514464,0.11251157,0.00014394074,0.0007011086,0.00015465917,0.02496014],"genre_scores_gemma":[0.068594694,0.04009298,0.01723581,0.34519377,0.06714757,0.00075634837,0.0026974361,0.0017408351,0.45654058],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.8900195,0.01659459,0.004296491,0.002769207,0.07441147,0.011908786],"domain_scores_gemma":[0.7863357,0.028426303,0.0048251287,0.0056554107,0.14141461,0.03334289],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.087604,0.0020365694,0.0022364599,0.005884655,0.014373721,0.022091243,0.005809788,0.019061765,0.025255809],"category_scores_gemma":[0.11918898,0.0015433076,0.0027426095,0.0069550727,0.018587481,0.0070577613,0.008861302,0.023111755,0.007989321],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000023284816,0.000012513747,0.00017717169,0.00013855293,0.000011688927,0.000034056917,0.00023590933,0.00006269112,0.000091390495,0.006090797,0.98190504,0.01121691],"study_design_scores_gemma":[0.000014177829,0.000009424478,0.0021708473,0.0003065621,0.000018160787,0.000029064811,0.0009867613,0.00009237404,0.00019958349,0.0027849595,0.99332446,0.00006358839],"about_ca_topic_score_codex":0.851636,"about_ca_topic_score_gemma":0.9492937,"teacher_disagreement_score":0.9065655,"about_ca_system_score_codex":0.09343453,"about_ca_system_score_gemma":0.27545196,"threshold_uncertainty_score":0.6779181},"labels":[],"label_agreement":null},{"id":"W4236954017","doi":"10.1177/135638900300900406","title":"Evaluability Assessment: A Tool for Incorporating Evaluation in Social Change Programmes","year":2003,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; University of Calgary","funders":"","keywords":"Participatory evaluation; Accountability; Process management; Promotion (chess); Process (computing); Social change; Monitoring and evaluation; Theory of change; Citizen journalism; Plan (archaeology); Political science; Theme (computing); Politics; Public relations; Management science; Knowledge management; Sociology; Business; Engineering; Computer science; Public administration","score_opus":0.4539931329436821,"score_gpt":0.5881362452183663,"score_spread":0.13414311227468417,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4236954017","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0044954508,0.0010617063,0.9555759,0.0035221553,0.00036726243,0.0063881963,0.00054903305,0.002312003,0.025728332],"genre_scores_gemma":[0.060960572,0.00054980384,0.9281792,0.00034958695,0.00012824479,0.008108224,0.00028937365,0.00031920918,0.0011158743],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.57975996,0.34770653,0.028310927,0.0057741944,0.036660146,0.0017881923],"domain_scores_gemma":[0.33747485,0.5555155,0.022209913,0.035019938,0.04672412,0.003055727],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.37963408,0.0036938235,0.0037270857,0.027491584,0.004755317,0.013061464,0.0032286996,0.0035037997,0.010195148],"category_scores_gemma":[0.5303242,0.0014460792,0.0039296993,0.014002976,0.007301831,0.017505916,0.011603938,0.006069284,0.0014251556],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047461633,0.0006815336,0.00576243,0.0061724293,0.0007461093,0.00023649269,0.010514852,0.010772445,0.0012306385,0.19065608,0.016303523,0.75644875],"study_design_scores_gemma":[0.000716824,0.0021405742,0.01161693,0.01246665,0.0012255049,0.0006295101,0.0075276457,0.052888196,0.005996759,0.6773503,0.22658709,0.0008540774],"about_ca_topic_score_codex":0.0030443957,"about_ca_topic_score_gemma":0.0028333676,"teacher_disagreement_score":0.37963408,"about_ca_system_score_codex":0.008520317,"about_ca_system_score_gemma":0.016434336,"threshold_uncertainty_score":0.7650216},"labels":[],"label_agreement":null},{"id":"W4285595442","doi":"10.1177/13563890221107044","title":"A theory-based approach to designing interventions for Planetary Health","year":2022,"lang":"en","type":"article","venue":"Evaluation","topic":"Climate Change and Health Impacts","field":"Environmental Science","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Psychological intervention; Action (physics); Management science; Theory of change; Engineering ethics; Field (mathematics); Human health; Public relations; Psychology; Computer science; Sociology; Political science; Medicine; Engineering","score_opus":0.285646917889852,"score_gpt":0.4283617954686286,"score_spread":0.1427148775787766,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285595442","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008295408,0.002644937,0.8945024,0.016854873,0.0007569492,0.015754452,0.0004382497,0.0003049752,0.06044771],"genre_scores_gemma":[0.09796052,0.0017436834,0.86723125,0.0025462552,0.000065485634,0.028141057,0.00020215173,0.00006355469,0.0020461609],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9262024,0.064830065,0.001963273,0.002599886,0.0035658272,0.000838575],"domain_scores_gemma":[0.94013107,0.05082987,0.0018389209,0.0030240023,0.0031804068,0.0009958093],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06520697,0.002900462,0.0023583644,0.0063084434,0.004291245,0.007956686,0.005179924,0.004257046,0.012790085],"category_scores_gemma":[0.05787364,0.0013777517,0.0027920748,0.003507831,0.013699803,0.0055806767,0.00521553,0.0064457688,0.0014229781],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00024914864,0.0013078867,0.0020956534,0.008150783,0.00049557775,0.00037772156,0.016935784,0.027994018,0.0014885329,0.79707015,0.00723927,0.13659546],"study_design_scores_gemma":[0.00071603263,0.0011466627,0.0013025207,0.006117912,0.00047292438,0.00019508321,0.011717251,0.02473524,0.0018755306,0.86854446,0.083041005,0.00013545301],"about_ca_topic_score_codex":0.005666189,"about_ca_topic_score_gemma":0.009298395,"teacher_disagreement_score":0.06520697,"about_ca_system_score_codex":0.015853252,"about_ca_system_score_gemma":0.02589076,"threshold_uncertainty_score":0.34485155},"labels":[],"label_agreement":null},{"id":"W4319439088","doi":"10.1177/13563890221149452","title":"The use of mediation analysis in evaluation of complex health interventions","year":2023,"lang":"en","type":"article","venue":"Evaluation","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Economic and Social Research Council; Bill and Melinda Gates Foundation","keywords":"Mediation; Causal inference; Intervention (counseling); Context (archaeology); Psychological intervention; Randomized controlled trial; Quality (philosophy); Psychology; Inference; Applied psychology; Medicine; Nursing; Computer science; Sociology; Social science","score_opus":0.9517514639979373,"score_gpt":0.7979085226517668,"score_spread":0.1538429413461705,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4319439088","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015055907,0.0073079793,0.94444984,0.007886418,0.0013757796,0.010031079,0.0006221766,0.0005274066,0.012743453],"genre_scores_gemma":[0.27753642,0.002637211,0.67388076,0.0028745057,0.00050009426,0.0412147,0.0002224367,0.0002583565,0.0008755008],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.1476134,0.8202204,0.009543229,0.006831057,0.014552805,0.0012391803],"domain_scores_gemma":[0.098476544,0.8588934,0.014346047,0.020247886,0.0074892105,0.00054697844],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5939302,0.0046972483,0.006574899,0.008295763,0.004225482,0.0084487945,0.0063716015,0.0048761996,0.011519307],"category_scores_gemma":[0.7509877,0.0023166288,0.009409927,0.008885584,0.010716407,0.014501058,0.012468956,0.010530751,0.0007439605],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0055214544,0.0012638976,0.019610105,0.027902003,0.030796554,0.0007178426,0.015586609,0.0283378,0.0011278063,0.4993029,0.0060984045,0.36373466],"study_design_scores_gemma":[0.0046921913,0.0069119027,0.008304295,0.010486897,0.010746864,0.00046012274,0.003913067,0.107159115,0.0049762973,0.8155981,0.026196264,0.00055507006],"about_ca_topic_score_codex":0.0053299377,"about_ca_topic_score_gemma":0.0040083737,"teacher_disagreement_score":0.40606982,"about_ca_system_score_codex":0.00818393,"about_ca_system_score_gemma":0.018539838,"threshold_uncertainty_score":0.5007564},"labels":[],"label_agreement":null},{"id":"W4322744890","doi":"10.1177/13563890231156954","title":"How can climate change and its interaction with other compounding risks be considered in evaluation? Experiences from Vietnam","year":2023,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; University of Alberta; University of Guelph","funders":"Australian Centre for International Agricultural Research; Canadian Institutes of Health Research; Consortium of International Agricultural Research Centers","keywords":"Climate change; Context (archaeology); Environmental resource management; Environmental planning; Public relations; Political science; Business; Psychology; Geography; Environmental science; Ecology","score_opus":0.6069973684791174,"score_gpt":0.5550850423223465,"score_spread":0.051912326156770994,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4322744890","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92416316,0.0025621313,0.0038811231,0.032298736,0.00018905532,0.00031136352,0.000046031884,0.000019006144,0.036529314],"genre_scores_gemma":[0.9961571,0.00072784506,0.001139933,0.0010100653,0.00001925573,0.000085821695,0.000008752284,0.000012235931,0.00083901465],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.95986426,0.034915518,0.00074662274,0.0004884483,0.0015049248,0.0024802648],"domain_scores_gemma":[0.9752891,0.016066154,0.0015763635,0.0005920887,0.003129593,0.0033465826],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.035040468,0.00031871797,0.0005115306,0.0006750119,0.006131114,0.0071241907,0.0010263474,0.0014282512,0.0021296772],"category_scores_gemma":[0.03249695,0.0003150387,0.0003965088,0.0009931391,0.0069712317,0.004882221,0.004570507,0.0024661887,0.0001418688],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019902275,0.00057655637,0.026659558,0.00088576745,0.000053116706,0.0034172395,0.9001906,0.000859536,0.0012989825,0.012690223,0.0034954506,0.049673975],"study_design_scores_gemma":[0.000032928492,0.00039362485,0.008940425,0.0011478424,0.00003162272,0.0005353032,0.9383713,0.00058698107,0.001095991,0.0049102446,0.043898303,0.00005532711],"about_ca_topic_score_codex":0.018253077,"about_ca_topic_score_gemma":0.027869068,"teacher_disagreement_score":0.035040468,"about_ca_system_score_codex":0.010900559,"about_ca_system_score_gemma":0.010444766,"threshold_uncertainty_score":0.18531394},"labels":[],"label_agreement":null},{"id":"W4382811084","doi":"10.1177/13563890231185167","title":"How to develop a realist programme theory using Margaret Archer’s structure–agency–culture framework: The case of adolescent accountability for sexual and reproductive health in urban resource-constrained settings","year":2023,"lang":"en","type":"article","venue":"Evaluation","topic":"Social Policy and Reform Studies","field":"Social Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"Fonds Wetenschappelijk Onderzoek","keywords":"Structure and agency; Agency (philosophy); Sociology; Critical realism (philosophy of perception); Reproductive health; Accountability; Context (archaeology); Resource (disambiguation); Epistemology; Social science; Political science; Realism; Population","score_opus":0.12633726763033404,"score_gpt":0.4286265921107111,"score_spread":0.30228932448037704,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4382811084","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028005246,0.005359626,0.37180713,0.35503796,0.001178069,0.003775844,0.00018118312,0.0002448377,0.2344102],"genre_scores_gemma":[0.7512515,0.0028443104,0.2139505,0.014078057,0.00022743753,0.0053248797,0.00008184035,0.00019312224,0.012048319],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9379204,0.05245798,0.0010848826,0.0022399053,0.0035515984,0.0027452188],"domain_scores_gemma":[0.9568037,0.032974727,0.0017063203,0.0031638239,0.0034443086,0.0019072082],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06394962,0.0013846825,0.0014808162,0.005382073,0.007991521,0.015807178,0.0051165586,0.009176149,0.008878578],"category_scores_gemma":[0.05135025,0.00094152585,0.0025031345,0.002303438,0.06199287,0.02936273,0.008902068,0.010575122,0.0008179312],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015153421,0.00009303822,0.00053023413,0.0002768484,0.00002120919,0.00016041257,0.012131071,0.0012718193,0.00006869203,0.9736265,0.002675119,0.009129907],"study_design_scores_gemma":[0.000080664184,0.000111131696,0.00054258376,0.0008101354,0.000029339968,0.00016010275,0.014458025,0.0037284356,0.00036679773,0.9367944,0.042877413,0.000041025545],"about_ca_topic_score_codex":0.011171556,"about_ca_topic_score_gemma":0.009907278,"teacher_disagreement_score":0.06394962,"about_ca_system_score_codex":0.021755153,"about_ca_system_score_gemma":0.02001015,"threshold_uncertainty_score":0.338202},"labels":[],"label_agreement":null},{"id":"W4389600041","doi":"10.1177/13563890231207115","title":"Interorganizational evaluation capacity building in the public, health and community sectors","year":2023,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Capacity building; Work (physics); Business; Public sector; Public relations; Joint (building); Knowledge management; Political science; Economic growth; Economics; Computer science; Engineering","score_opus":0.6482669341035596,"score_gpt":0.5566919701817297,"score_spread":0.09157496392182995,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389600041","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.71953106,0.0019806284,0.055642657,0.02062061,0.00021033341,0.002961866,0.00014518909,0.0003143984,0.19859338],"genre_scores_gemma":[0.99029046,0.00017619907,0.0059556523,0.00029427026,0.000017253229,0.00027548388,0.000041011117,0.000022373059,0.0029273499],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.90010726,0.070376344,0.0029354386,0.0031593086,0.008797896,0.014623731],"domain_scores_gemma":[0.84194916,0.086477645,0.012875489,0.013840771,0.024915027,0.01994194],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.086027876,0.000521513,0.0005251041,0.005542811,0.01179366,0.012944412,0.0029139232,0.0016949187,0.0045572924],"category_scores_gemma":[0.06629769,0.0006211604,0.0007667285,0.004582454,0.017344087,0.0063165906,0.02274171,0.0031855877,0.00037694606],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00035685286,0.0018043553,0.15304795,0.0012458983,0.00039432323,0.002480512,0.12895456,0.027119998,0.002675376,0.3190979,0.014735778,0.3480865],"study_design_scores_gemma":[0.0001814087,0.00075104454,0.17840785,0.002724314,0.00013988304,0.0010113303,0.42908213,0.020083921,0.003892472,0.121252656,0.24203539,0.000437595],"about_ca_topic_score_codex":0.036230728,"about_ca_topic_score_gemma":0.059224103,"teacher_disagreement_score":0.91397214,"about_ca_system_score_codex":0.033459127,"about_ca_system_score_gemma":0.076312676,"threshold_uncertainty_score":0.45496434},"labels":[],"label_agreement":null},{"id":"W4391389032","doi":"10.1177/13563890241227433","title":"Planetary health: Creating rapid impact assessment tools","year":2024,"lang":"en","type":"article","venue":"Evaluation","topic":"Climate Change and Health Impacts","field":"Environmental Science","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Future Earth; University of Victoria","funders":"","keywords":"Environmental planning; Environmental science","score_opus":0.17725812000104163,"score_gpt":0.4650921412998606,"score_spread":0.28783402129881897,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391389032","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.019892272,0.0031276178,0.845611,0.029856334,0.0012362733,0.005975021,0.0013701534,0.008504589,0.08442672],"genre_scores_gemma":[0.07134136,0.0016212102,0.91568226,0.0016266711,0.00027491048,0.0035511472,0.00081576465,0.00094549207,0.0041412106],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8619362,0.10682809,0.0072654733,0.0028918446,0.018619683,0.0024586786],"domain_scores_gemma":[0.8022492,0.13379794,0.008638935,0.020856326,0.030069008,0.004388663],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1758843,0.0016081203,0.0009464735,0.007926452,0.0026581467,0.016628213,0.0042735776,0.0033978748,0.009397503],"category_scores_gemma":[0.20405473,0.0011191687,0.0014409889,0.003426017,0.006667349,0.02128199,0.014245529,0.0037479105,0.0028520052],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029296646,0.0007856869,0.00574595,0.0033306205,0.00016597535,0.0006234779,0.017897267,0.012098839,0.0027716446,0.25713307,0.052541304,0.6466131],"study_design_scores_gemma":[0.0002964689,0.0008274862,0.0035725797,0.0071287635,0.0001475544,0.00061072985,0.014249295,0.014235021,0.009613496,0.3983864,0.5504733,0.00045887075],"about_ca_topic_score_codex":0.0019575423,"about_ca_topic_score_gemma":0.002875526,"teacher_disagreement_score":0.1758843,"about_ca_system_score_codex":0.0053671235,"about_ca_system_score_gemma":0.016493758,"threshold_uncertainty_score":0.93017626},"labels":[],"label_agreement":null},{"id":"W4401276261","doi":"10.1177/13563890241265859","title":"Post-truth and pathways for evaluators","year":2024,"lang":"en","type":"article","venue":"Evaluation","topic":"Viral Infections and Outbreaks Research","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Disinformation; Transformative learning; Skepticism; Post truth; Politics; Phenomenon; Narrative; Fake news; Sociology; Premise; Political science; Public relations; Epistemology; Social media; Law; Media studies","score_opus":0.0724581220496297,"score_gpt":0.4204130039843808,"score_spread":0.3479548819347511,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401276261","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18505138,0.01053208,0.097268045,0.35827017,0.0035174522,0.0016045093,0.00017998097,0.0005507244,0.34302568],"genre_scores_gemma":[0.9458624,0.0016727208,0.013807859,0.014972716,0.00047873595,0.00067650183,0.000057093486,0.00022667626,0.022245232],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8480664,0.12683825,0.0031403902,0.005454591,0.008800624,0.007699737],"domain_scores_gemma":[0.78154016,0.1482687,0.014493563,0.011665381,0.029549519,0.014482593],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.122275956,0.0007196036,0.00067705323,0.003082067,0.013015056,0.020061934,0.002166497,0.0058801346,0.01252516],"category_scores_gemma":[0.2086393,0.00075627817,0.00064910745,0.0017374031,0.028897548,0.018644568,0.01959853,0.008545632,0.0016427992],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003219645,0.00017938882,0.012681979,0.00079264503,0.00006396903,0.0012675718,0.34139115,0.00052819244,0.0013747708,0.46080977,0.022145744,0.15844283],"study_design_scores_gemma":[0.0001214814,0.0003950566,0.0062627327,0.0029233715,0.0000651556,0.00071782,0.24778438,0.0011881535,0.0020947438,0.33016813,0.40810683,0.00017216103],"about_ca_topic_score_codex":0.0033433135,"about_ca_topic_score_gemma":0.004060132,"teacher_disagreement_score":0.87772405,"about_ca_system_score_codex":0.010213594,"about_ca_system_score_gemma":0.020035122,"threshold_uncertainty_score":0.6466648},"labels":[],"label_agreement":null},{"id":"W4404873315","doi":"10.1177/13563890241289937","title":"Artificial intelligence and big data-driven evaluation research and practices: A systematic literature review","year":2024,"lang":"en","type":"article","venue":"Evaluation","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"École Nationale d'Administration Publique","funders":"","keywords":"Big data; Systematic review; Data science; Management science; Psychology; Political science; Engineering ethics; Computer science; MEDLINE; Engineering; Data mining","score_opus":0.7863038003097949,"score_gpt":0.6347017197247322,"score_spread":0.15160208058506264,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404873315","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020476396,0.99077284,0.0016277883,0.0023079792,0.00018344248,0.0014314097,0.0003377927,0.000021751688,0.0012694673],"genre_scores_gemma":[0.045415074,0.93957275,0.008158366,0.0017726473,0.00011940092,0.0043446207,0.0004156144,0.000029772757,0.00017179242],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.8903001,0.06455066,0.026852215,0.0025743397,0.01457648,0.0011461057],"domain_scores_gemma":[0.660991,0.27931032,0.027483514,0.004999712,0.02527837,0.0019370076],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10399675,0.0012875212,0.005738563,0.021460116,0.0018786856,0.006941498,0.0027016734,0.0024139476,0.003125309],"category_scores_gemma":[0.28013992,0.0013543274,0.004945094,0.022665096,0.0034420437,0.006510608,0.0045352466,0.0022833713,0.00037685805],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021058669,0.000109549524,0.0021609375,0.75170547,0.0024609512,0.00015672996,0.0033008042,0.00048373538,0.0001502077,0.0045104427,0.004660712,0.23008986],"study_design_scores_gemma":[0.00006700449,0.00009287777,0.0017158639,0.96735805,0.0038587651,0.00017852434,0.0018149615,0.00016363888,0.00013263676,0.0018368023,0.022738142,0.000042658612],"about_ca_topic_score_codex":0.009866173,"about_ca_topic_score_gemma":0.026789198,"teacher_disagreement_score":0.89600325,"about_ca_system_score_codex":0.013672749,"about_ca_system_score_gemma":0.058235902,"threshold_uncertainty_score":0.549994},"labels":[],"label_agreement":null},{"id":"W4417434618","doi":"10.1177/13563890251395056","title":"Unraveling the complexities of learning in community development evaluation","year":2025,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; University of Victoria","funders":"","keywords":"Neoliberalism (international relations); Community development; Focus (optics); Learning community; Key (lock)","score_opus":0.42157615803294757,"score_gpt":0.5579690605544919,"score_spread":0.13639290252154435,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417434618","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04969215,0.03920178,0.48286128,0.2706071,0.0021432387,0.0009412765,0.00007585939,0.00033481172,0.15414254],"genre_scores_gemma":[0.8469325,0.006884668,0.13222355,0.0062241224,0.0011899944,0.0011343852,0.000034691104,0.00020145635,0.0051744934],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.68761146,0.27039167,0.007576373,0.006416239,0.023755984,0.004248297],"domain_scores_gemma":[0.4651393,0.48827457,0.008076977,0.016486904,0.018475529,0.003546755],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.26339328,0.0012817913,0.0026998252,0.0069809663,0.0130240815,0.04050844,0.0054114894,0.010978236,0.0029413018],"category_scores_gemma":[0.28967163,0.0013641092,0.0011316254,0.0049009966,0.105614625,0.054486815,0.024180992,0.015196319,0.0004612012],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000034750163,0.000059657796,0.0012241699,0.0007365881,0.000035452013,0.00027695645,0.038719967,0.0017013242,0.00017807879,0.90094954,0.0019600273,0.05412351],"study_design_scores_gemma":[0.000035412737,0.00007754543,0.0005925562,0.002240956,0.000028007475,0.00028994816,0.023931904,0.0046139723,0.0005933383,0.912279,0.055239268,0.00007812099],"about_ca_topic_score_codex":0.005979077,"about_ca_topic_score_gemma":0.006340201,"teacher_disagreement_score":0.26339328,"about_ca_system_score_codex":0.017451476,"about_ca_system_score_gemma":0.021317255,"threshold_uncertainty_score":0.9083672},"labels":[],"label_agreement":null}]}