{"meta":{"query_hash":"6550cf26dc1c","filters":{"venue":"African Evaluation Journal"},"cohort_total":16,"direct_labels_cover":0,"predictions_cover":16,"exported":16,"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/6550cf26dc1c","api":"https://metacan.xera.ac/api/v1/cohort?venue=African+Evaluation+Journal"},"results":[{"id":"W2010270738","doi":"10.4102/aej.v2i1.98","title":"Reshaping development evaluation: Meeting the challenges of a changing context","year":2014,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Evaluation and Performance Assessment","field":"Decision 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":"Mastercard Foundation","funders":"","keywords":"Context (archaeology); Development (topology); Computer science; Process management; Engineering ethics; Engineering; Geography; Mathematics","score_opus":0.38877638053612223,"score_gpt":0.4856660053502125,"score_spread":0.09688962481409025,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2010270738","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":[0.008257134,0.033092406,0.035661634,0.90295655,0.0044348896,0.00043508346,0.00010231699,0.00029502372,0.014764958],"genre_scores_gemma":[0.44436896,0.058111798,0.2538429,0.21293779,0.015291279,0.00212745,0.00045799548,0.0007471257,0.012114696],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.85195273,0.11317332,0.005701212,0.004594557,0.018817062,0.0057610716],"domain_scores_gemma":[0.61348987,0.24334641,0.016842531,0.02169748,0.06818277,0.036440965],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27646622,0.001690643,0.002964834,0.0052988883,0.010110034,0.029830458,0.0063066445,0.011344716,0.00643184],"category_scores_gemma":[0.20621572,0.0009133705,0.0015453678,0.004532419,0.013494232,0.031822152,0.018181344,0.022486063,0.0019149473],"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":[0.0002169222,0.00040306835,0.003502657,0.003651713,0.00015100634,0.00033300917,0.004889515,0.0039203567,0.0012292339,0.092943855,0.12505147,0.7637072],"study_design_scores_gemma":[0.0002818547,0.000983003,0.012408846,0.0120766545,0.00013851964,0.0006932254,0.027778907,0.0077793347,0.0013643955,0.3518853,0.5841228,0.000487228],"about_ca_topic_score_codex":0.011841248,"about_ca_topic_score_gemma":0.024297781,"teacher_disagreement_score":0.72353375,"about_ca_system_score_codex":0.013936014,"about_ca_system_score_gemma":0.075017594,"threshold_uncertainty_score":0.89224595},"labels":[],"label_agreement":null},{"id":"W2019919882","doi":"10.4102/aej.v1i1.43","title":"Participatory evaluation for development: Examining research-based knowledge from within the African context","year":2013,"lang":"en","type":"article","venue":"African Evaluation Journal","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":"University of Ottawa","funders":"","keywords":"Popularity; Context (archaeology); Empowerment; Thematic analysis; Participatory development; Citizen journalism; Relevance (law); Participatory action research; Sociology; Participatory evaluation; Engineering ethics; Public relations; Political science; Psychology; Qualitative research; Social science; Geography; Social psychology; Engineering","score_opus":0.7823683775505966,"score_gpt":0.5926367515141084,"score_spread":0.1897316260364882,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2019919882","genre_codex":"review","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":[0.18564057,0.2739486,0.24879993,0.14622885,0.0032662766,0.014231073,0.0006435327,0.00019026248,0.12705083],"genre_scores_gemma":[0.8135375,0.058427587,0.11232997,0.0054613207,0.00042948136,0.008339843,0.00015902423,0.00006892533,0.0012462585],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.59912413,0.36241734,0.012709278,0.005915559,0.016152522,0.0036811423],"domain_scores_gemma":[0.40959904,0.5448314,0.014155095,0.012341709,0.016292004,0.0027806673],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2891359,0.0014008798,0.0030204167,0.021021875,0.009839561,0.02006662,0.003518993,0.0052090073,0.003074911],"category_scores_gemma":[0.2777594,0.0012111737,0.0014220239,0.016404886,0.032280307,0.024167398,0.018713525,0.0043726857,0.00031215182],"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":[0.00013765265,0.00035291747,0.008158226,0.041161485,0.0004349039,0.0013988646,0.4673991,0.0012699711,0.0011325519,0.17370506,0.004620464,0.3002288],"study_design_scores_gemma":[0.00011910444,0.0005846697,0.006564172,0.10669857,0.00041033642,0.0011571273,0.5438666,0.0014259198,0.0018242612,0.18423615,0.15295666,0.00015643834],"about_ca_topic_score_codex":0.0030762907,"about_ca_topic_score_gemma":0.0041658822,"teacher_disagreement_score":0.2891359,"about_ca_system_score_codex":0.019366011,"about_ca_system_score_gemma":0.052370068,"threshold_uncertainty_score":0.87662196},"labels":[],"label_agreement":null},{"id":"W2106590685","doi":"10.4102/aej.v3i2.153","title":"Evidence from epidemic appraisals in Nigeria","year":2015,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Sex work and related issues","field":"Social Sciences","cited_by":10,"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 Manitoba","funders":"","keywords":"Psychological intervention; Condom; Population; Ordination; Demography; Geography; Environmental health; Medicine; Human immunodeficiency virus (HIV); Sociology; Family medicine; Statistics; Nursing","score_opus":0.20480688325165725,"score_gpt":0.4403412187964677,"score_spread":0.23553433554481046,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2106590685","genre_codex":"review","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":[0.016244175,0.93242615,0.0022771463,0.00969737,0.0025464466,0.008507634,0.0051027285,0.000072472394,0.0231259],"genre_scores_gemma":[0.14288262,0.8328302,0.007551647,0.0040057893,0.0009051327,0.008098924,0.0026617527,0.000070462294,0.0009934327],"study_design_codex":"systematic_review","study_design_gemma":"observational","domain_scores_codex":[0.8874548,0.060706627,0.030535458,0.0034913844,0.016235739,0.0015760858],"domain_scores_gemma":[0.5011018,0.40425077,0.04422826,0.008852249,0.03935896,0.0022079954],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11350006,0.0026946368,0.0044356156,0.020380907,0.002002569,0.009144309,0.0035378777,0.0031342667,0.012887468],"category_scores_gemma":[0.40175018,0.0028767567,0.0061286017,0.019067999,0.002218028,0.0061183493,0.0058609163,0.005884236,0.00118658],"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":[0.0021082093,0.0004158984,0.007023405,0.7494841,0.009319254,0.0003751378,0.004614228,0.0011971261,0.00022950419,0.004504306,0.011641839,0.20908698],"study_design_scores_gemma":[0.0010092979,0.0006635623,0.010487769,0.90423334,0.010339014,0.0001929349,0.003174672,0.0003517104,0.0004544804,0.0024760752,0.06651771,0.00009947362],"about_ca_topic_score_codex":0.015474366,"about_ca_topic_score_gemma":0.027128005,"teacher_disagreement_score":0.11350006,"about_ca_system_score_codex":0.013750104,"about_ca_system_score_gemma":0.031010514,"threshold_uncertainty_score":0.60025287},"labels":[],"label_agreement":null},{"id":"W2142148571","doi":"10.4102/aej.v3i1.139","title":"Reflecting on an impact evaluation of the Grade R programme: Method, results and policy responses","year":2015,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Education and Early Childhood Development","funders":"","keywords":"Government (linguistics); Process (computing); Monitoring and evaluation; Quality (philosophy); Political science; Program evaluation; Impact evaluation; Business; Public relations; Medical education; Psychology; Public administration; Medicine; Computer science","score_opus":0.4550377342861057,"score_gpt":0.5856322533769877,"score_spread":0.13059451909088193,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2142148571","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":[0.4004457,0.008924999,0.07514393,0.04377364,0.0028865843,0.21623772,0.00559817,0.0007685843,0.24622062],"genre_scores_gemma":[0.710475,0.0059168255,0.109625444,0.015617381,0.0006398508,0.13723719,0.0013358203,0.00042938787,0.018723072],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.68055403,0.27692854,0.007835182,0.0038912415,0.02457249,0.0062185484],"domain_scores_gemma":[0.7525439,0.18819074,0.011425375,0.009661642,0.035662167,0.0025161365],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.21676487,0.0016772234,0.0013941101,0.003854748,0.0033580335,0.0073689325,0.0024122095,0.0036000777,0.008409529],"category_scores_gemma":[0.21848838,0.0008774005,0.0018085187,0.004696455,0.004334493,0.005088822,0.0070846053,0.0034289227,0.0015996691],"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":[0.0099132685,0.00939344,0.03184238,0.029287364,0.0006514062,0.0011855813,0.06419995,0.011026163,0.0066895485,0.076731265,0.04613608,0.7129435],"study_design_scores_gemma":[0.012729977,0.05171699,0.11897646,0.03711799,0.0023874983,0.000905391,0.17460872,0.0074948077,0.045540925,0.052003033,0.4949745,0.0015436813],"about_ca_topic_score_codex":0.008755331,"about_ca_topic_score_gemma":0.006189478,"teacher_disagreement_score":0.21676487,"about_ca_system_score_codex":0.019444915,"about_ca_system_score_gemma":0.021479934,"threshold_uncertainty_score":0.96586835},"labels":[],"label_agreement":null},{"id":"W2175989841","doi":"10.4102/aej.v3i1.145","title":"Use of evidence in policy making in South Africa: An exploratory study of attitudes of senior government officials","year":2015,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":32,"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":"Evidence-based policy; Presidency; Government (linguistics); Political science; Public policy; Politics; Context (archaeology); Public administration; Public relations; Medicine; Geography","score_opus":0.7719629586370481,"score_gpt":0.5688494283805711,"score_spread":0.20311353025647705,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2175989841","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":[0.99377173,0.00050350244,0.00024957125,0.003314041,0.000020857757,0.000067728106,0.000013081549,0.00000401103,0.0020554834],"genre_scores_gemma":[0.9963595,0.0009833657,0.0005214802,0.0010832611,0.000017169694,0.00008200738,0.000012104028,0.000006093187,0.0009349933],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9757296,0.016071172,0.0016871806,0.000782752,0.0022510819,0.0034782144],"domain_scores_gemma":[0.93909395,0.041708935,0.008305656,0.0014803644,0.0044540614,0.004956919],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.026109062,0.00052367634,0.0009317766,0.0032978675,0.013527637,0.009864027,0.0012247995,0.0030743626,0.0022733395],"category_scores_gemma":[0.054184385,0.0012823085,0.00056627084,0.0032097187,0.010435667,0.005131043,0.0073673422,0.0041844966,0.00039304054],"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":[0.000049081544,0.00008483564,0.021623561,0.00023231127,0.000013839716,0.000513595,0.96760154,0.00003346391,0.0010281926,0.00083945814,0.00021410009,0.007766144],"study_design_scores_gemma":[0.000007760728,0.00015438917,0.010465472,0.00023613762,0.000010982126,0.00019363652,0.9827943,0.0000660197,0.00023890793,0.0003042866,0.005505693,0.00002252956],"about_ca_topic_score_codex":0.012334531,"about_ca_topic_score_gemma":0.014581708,"teacher_disagreement_score":0.026109062,"about_ca_system_score_codex":0.0075542554,"about_ca_system_score_gemma":0.010169538,"threshold_uncertainty_score":0.13807952},"labels":[],"label_agreement":null},{"id":"W2557681684","doi":"10.4102/aej.v4i1.178","title":"Evaluation involvement of local HIV/AIDS non-governmental organisations in Benin","year":2016,"lang":"en","type":"article","venue":"African Evaluation Journal","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":"Université de Montréal","funders":"","keywords":"Civil society; Monitoring and evaluation; Human immunodeficiency virus (HIV); Task (project management); Political science; Public relations; Capacity building; Pandemic; Descriptive research; Business; Economic growth; Medicine; Sociology; Coronavirus disease 2019 (COVID-19); Politics; Management; Family medicine; Economics; Social science; Disease; Law","score_opus":0.1574875201533343,"score_gpt":0.4484900314840136,"score_spread":0.29100251133067934,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2557681684","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":[0.99169827,0.00045754088,0.00016149759,0.00090093433,0.000011126629,0.00010052699,0.00003020876,0.000007238033,0.006632653],"genre_scores_gemma":[0.99740773,0.00023612857,0.00015404838,0.00010991129,0.0000057049706,0.00005128586,0.000023696743,0.0000036317058,0.0020078346],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98932624,0.008176996,0.00025042353,0.00022302846,0.0006509159,0.0013724073],"domain_scores_gemma":[0.9762263,0.012360047,0.0027674555,0.00039471604,0.0026168954,0.0056345877],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008799136,0.0002902671,0.00024303119,0.0010889703,0.0044163796,0.002733381,0.00068132364,0.00072500436,0.0063300016],"category_scores_gemma":[0.01293946,0.00023203304,0.00010540153,0.0008606013,0.0015980775,0.0012612874,0.0030885201,0.00086218875,0.0004251676],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012650322,0.0024326195,0.42474082,0.0014986584,0.000060893075,0.006508429,0.40379944,0.00069011166,0.0029554537,0.0033846854,0.006974182,0.14568968],"study_design_scores_gemma":[0.000031272088,0.0006426847,0.2500632,0.0005211956,0.000022171429,0.0007605942,0.72549665,0.00055401033,0.0010804163,0.0004363345,0.020346936,0.000044517365],"about_ca_topic_score_codex":0.0063742925,"about_ca_topic_score_gemma":0.01811754,"teacher_disagreement_score":0.008799136,"about_ca_system_score_codex":0.004697973,"about_ca_system_score_gemma":0.0047386037,"threshold_uncertainty_score":0.046534836},"labels":[],"label_agreement":null},{"id":"W2901269715","doi":"10.4102/aej.v6i2.335","title":"Understanding and optimising the social impact of venture capital: Three lessons from Ghana","year":2018,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Community Development and Social Impact","field":"Economics, Econometrics and Finance","cited_by":5,"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","funders":"","keywords":"Venture capital; Impact investing; Business; Social capital; Context (archaeology); Documentation; Theory of change; Work (physics); Revenue; Finance; Public relations; Public economics; Marketing; Economics; Emerging markets; Management; Sociology; Political science","score_opus":0.3507713889362565,"score_gpt":0.3867330228179681,"score_spread":0.03596163388171164,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2901269715","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":[0.8442425,0.014097779,0.0030758171,0.0841017,0.0002812319,0.0001923121,0.00007746331,0.000015106667,0.053916145],"genre_scores_gemma":[0.9832304,0.009464119,0.0018273784,0.0021992542,0.000041320065,0.000043808235,0.0000241173,0.000012308517,0.0031572883],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9987477,0.0007971086,0.000032281205,0.000052511925,0.0001141464,0.00025615844],"domain_scores_gemma":[0.99418074,0.004240844,0.00045176,0.00012343819,0.00035412324,0.00064900226],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004291086,0.00048453355,0.00031017314,0.00094604626,0.0036849175,0.0051368885,0.0007221245,0.0013816495,0.002402708],"category_scores_gemma":[0.0067419936,0.00028088555,0.00019451413,0.0014445721,0.006124929,0.0069326945,0.004348817,0.0023555155,0.00019729203],"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":[0.00011313101,0.00018990505,0.025303887,0.0011538592,0.00001589481,0.011565341,0.7368123,0.00052458275,0.0011369344,0.080565006,0.010107572,0.13251163],"study_design_scores_gemma":[0.000022336808,0.00009875577,0.0153280115,0.0023664662,0.000015002474,0.0007142113,0.8412767,0.00041834713,0.00053137593,0.04158628,0.09762193,0.000020597816],"about_ca_topic_score_codex":0.017191816,"about_ca_topic_score_gemma":0.048136365,"teacher_disagreement_score":0.017191816,"about_ca_system_score_codex":0.006629327,"about_ca_system_score_gemma":0.0071200468,"threshold_uncertainty_score":0.0480994},"labels":[],"label_agreement":null},{"id":"W2957596154","doi":"10.4102/aej.v7i1.369","title":"Evaluating the sustainability of health programmes: A literature review","year":2019,"lang":"en","type":"review","venue":"African Evaluation Journal","topic":"Global Maternal and Child Health","field":"Medicine","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":"McGill University; McGill University Health Centre","funders":"","keywords":"Sustainability; Inclusion (mineral); Business; Sustainable development; Grey literature; Capacity building; Environmental resource management; Environmental planning; Political science; Economic growth; MEDLINE; Geography; Psychology; Economics","score_opus":0.16521390875610895,"score_gpt":0.5254857871254746,"score_spread":0.3602718783693656,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2957596154","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":[0.00067049696,0.9956636,0.0003006012,0.0013566137,0.00023139978,0.00021320659,0.00011232123,0.000008543631,0.0014432891],"genre_scores_gemma":[0.0064433613,0.9920047,0.0007456634,0.00044103112,0.00008178829,0.0001544948,0.000053386782,0.000004066787,0.000071486094],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.97957945,0.008822329,0.004567689,0.0009857231,0.0054458557,0.00059906155],"domain_scores_gemma":[0.8901279,0.091276206,0.008106212,0.0011059899,0.008402285,0.0009814502],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021707699,0.0015174024,0.003648345,0.019449107,0.0014510114,0.006936332,0.0025565652,0.0032759986,0.0049106316],"category_scores_gemma":[0.08065896,0.0012983641,0.004102761,0.018598009,0.0021583964,0.006680417,0.0029345779,0.0027491956,0.00053384475],"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":[0.00008597318,0.00009904592,0.001172498,0.65664333,0.001129215,0.00024778311,0.0015351671,0.00057704945,0.00017094564,0.0039113304,0.0071809147,0.32724676],"study_design_scores_gemma":[0.000019423844,0.000075196716,0.0018409375,0.9378174,0.0021036412,0.00024135252,0.0012433553,0.00010551002,0.00013144228,0.0011425971,0.055251863,0.000027174023],"about_ca_topic_score_codex":0.009823504,"about_ca_topic_score_gemma":0.018910546,"teacher_disagreement_score":0.021707699,"about_ca_system_score_codex":0.009004561,"about_ca_system_score_gemma":0.029637415,"threshold_uncertainty_score":0.11480266},"labels":[],"label_agreement":null},{"id":"W2967863356","doi":"10.4102/aej.v7i1.400","title":"Evaluation2 – Evaluating the national evaluation system in South Africa: What has been achieved in the first 5 years?","year":2019,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":34,"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":"Department for International Development","keywords":"Mandate; Cabinet (room); Benchmarking; Government (linguistics); Monitoring and evaluation; Legislation; Business; Political science; Economic growth; Public administration; Geography; Economics; Marketing","score_opus":0.39139637284341017,"score_gpt":0.48634111677156405,"score_spread":0.09494474392815389,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2967863356","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":[0.21289955,0.16987905,0.05510821,0.27252552,0.005902534,0.014206272,0.0036274355,0.00090490514,0.26494658],"genre_scores_gemma":[0.8522779,0.03786676,0.07020642,0.014017019,0.0005521488,0.006321024,0.0029242507,0.00040705598,0.0154273175],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.7039132,0.2413447,0.012897495,0.003994267,0.02767122,0.010179015],"domain_scores_gemma":[0.6418645,0.13791099,0.022098253,0.015507516,0.16849032,0.014128429],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.33664796,0.0012422216,0.0018228535,0.004010301,0.0040236055,0.013020803,0.0030689104,0.0026943483,0.008404044],"category_scores_gemma":[0.31600788,0.0006444671,0.0015020825,0.0053708255,0.00479534,0.012280105,0.006787196,0.0034893139,0.0013627704],"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":[0.0014108609,0.0009910017,0.027570732,0.022794416,0.00055530923,0.00038488006,0.018316338,0.0041431184,0.001207023,0.054869477,0.07175566,0.79600126],"study_design_scores_gemma":[0.00080669555,0.004472093,0.11213852,0.12286502,0.00072382024,0.00097424997,0.05796925,0.0068177325,0.00779403,0.028705236,0.6562603,0.00047310296],"about_ca_topic_score_codex":0.032044847,"about_ca_topic_score_gemma":0.027279938,"teacher_disagreement_score":0.33664796,"about_ca_system_score_codex":0.036578123,"about_ca_system_score_gemma":0.100982666,"threshold_uncertainty_score":0.8180312},"labels":[],"label_agreement":null},{"id":"W3010364343","doi":"10.4102/aej.v8i1.426","title":"Use of the most significant change technique to evaluate intervention in promoting childbirth spacing in Nigeria","year":2020,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Global Maternal and Child Health","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact","funders":"","keywords":"Childbirth; Intervention (counseling); Reproductive health; Miscarriage; Psychological intervention; Pregnancy; Medicine; Storytelling; Psychology; Nursing; Family medicine; Population; Environmental health","score_opus":0.1021355406545086,"score_gpt":0.35073500747226716,"score_spread":0.24859946681775857,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3010364343","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":[0.90088,0.00086620194,0.023950968,0.0012822138,0.00047116104,0.054505885,0.0014147583,0.0002134994,0.016415326],"genre_scores_gemma":[0.6553253,0.0010052838,0.1444198,0.00045204375,0.000046187277,0.19620548,0.00035312198,0.00005201783,0.0021407474],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9691993,0.025451314,0.0017945197,0.0008989587,0.0019813594,0.0006745358],"domain_scores_gemma":[0.96369416,0.028262548,0.0028095744,0.0011260081,0.003747591,0.0003601877],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.025131525,0.0008768143,0.0010910424,0.002708767,0.0017865323,0.0013108954,0.0011390559,0.00077409495,0.0036210867],"category_scores_gemma":[0.04790964,0.0005358393,0.0015264401,0.0022085684,0.0016955659,0.0011506041,0.0021022642,0.0013590526,0.00024950024],"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":[0.0051803477,0.0050834217,0.04410611,0.0112911025,0.00045262696,0.0010577491,0.23510343,0.0018479124,0.011178694,0.0075682006,0.00713391,0.6699964],"study_design_scores_gemma":[0.00480148,0.06748868,0.40034118,0.008684157,0.0015652748,0.0010223946,0.38718826,0.014671946,0.039525237,0.012778436,0.061514474,0.0004184416],"about_ca_topic_score_codex":0.0018284215,"about_ca_topic_score_gemma":0.0044409423,"teacher_disagreement_score":0.025131525,"about_ca_system_score_codex":0.0029975078,"about_ca_system_score_gemma":0.003364049,"threshold_uncertainty_score":0.13290977},"labels":[],"label_agreement":null},{"id":"W3203723536","doi":"10.4102/aej.v9i1.553","title":"How to measure monitoring and evaluation system effectiveness?","year":2021,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":16,"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":"Université du Québec à Montréal","keywords":"Scarcity; Process management; Monitoring and evaluation; Corporate governance; Quality (philosophy); Business; Sample (material); Information system; Knowledge management; Environmental resource management; Management science; Computer science; Engineering; Economics; Economic growth","score_opus":0.27884330159616744,"score_gpt":0.48988338200161946,"score_spread":0.21104008040545202,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3203723536","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":[0.16821562,0.07840067,0.51220894,0.09649812,0.004732037,0.00990442,0.004595739,0.0013355269,0.12410891],"genre_scores_gemma":[0.709997,0.0068900497,0.2701198,0.0040367283,0.0007491513,0.006250464,0.0007556972,0.00016281962,0.0010383617],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5690457,0.3140189,0.035964943,0.0111700455,0.0664008,0.0033996203],"domain_scores_gemma":[0.2913031,0.5301515,0.075501055,0.024743462,0.07447234,0.0038284445],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2710896,0.0018843102,0.0023825238,0.012973354,0.002347106,0.011785931,0.0033978296,0.0031046106,0.003683549],"category_scores_gemma":[0.48825654,0.0007321241,0.0024739772,0.015268883,0.010306364,0.0147631485,0.0043493453,0.0037746655,0.0011054128],"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":[0.0004268889,0.0008155491,0.16662458,0.021358488,0.0023916732,0.00008667664,0.012553566,0.0067453994,0.0013365614,0.11098103,0.024641642,0.6520379],"study_design_scores_gemma":[0.0005208525,0.0053931787,0.3538479,0.06304494,0.003431307,0.0008522594,0.048576422,0.037367333,0.014073665,0.24951422,0.22216628,0.0012116907],"about_ca_topic_score_codex":0.004256167,"about_ca_topic_score_gemma":0.0028589352,"teacher_disagreement_score":0.2710896,"about_ca_system_score_codex":0.011460155,"about_ca_system_score_gemma":0.014854804,"threshold_uncertainty_score":0.8988763},"labels":[],"label_agreement":null},{"id":"W4385839249","doi":"10.4102/aej.v11i1.654","title":"Etat des systèmes de suivi et d’évaluation en Afrique francophone : une approximation au moyen d’un diagnostic rapide","year":2023,"lang":"fr","type":"article","venue":"African Evaluation Journal","topic":"Evaluation and Performance Assessment","field":"Decision 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; Global Affairs Canada; École Nationale d'Administration Publique","funders":"","keywords":"French; Institutionalisation; Context (archaeology); Valuation (finance); Political science; Humanities; Welfare economics; Business; Geography; Accounting; Economics; Philosophy","score_opus":0.11246508197300573,"score_gpt":0.4336464619317929,"score_spread":0.32118137995878715,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385839249","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":[0.22189528,0.13764235,0.12718685,0.4359006,0.0033267455,0.003252058,0.00322903,0.0010846946,0.06648246],"genre_scores_gemma":[0.7772966,0.027952353,0.17312737,0.00914684,0.0005275267,0.0018072191,0.0012378284,0.00016583067,0.008738517],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.873614,0.09522774,0.007369845,0.005308703,0.013983595,0.0044960226],"domain_scores_gemma":[0.7000861,0.17224522,0.020428607,0.01417099,0.08117744,0.011891574],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19608048,0.0012178598,0.0009885951,0.0061550187,0.0047515817,0.013007194,0.0031503278,0.003835922,0.0046145017],"category_scores_gemma":[0.16613473,0.00073804165,0.001109177,0.0042367484,0.0073853224,0.012364274,0.005088098,0.0051218164,0.000812923],"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":[0.00069425866,0.00036133538,0.078048326,0.007510873,0.00031404756,0.0010530583,0.06256038,0.0040173465,0.0028492145,0.09619615,0.04735025,0.69904464],"study_design_scores_gemma":[0.0003148757,0.0013251523,0.15654126,0.04056525,0.0003290501,0.0015731284,0.078670345,0.011177317,0.0049582254,0.032311417,0.6715534,0.00068065064],"about_ca_topic_score_codex":0.20324488,"about_ca_topic_score_gemma":0.11185553,"teacher_disagreement_score":0.20324488,"about_ca_system_score_codex":0.04285113,"about_ca_system_score_gemma":0.085200846,"threshold_uncertainty_score":0.9913759},"labels":[],"label_agreement":null},{"id":"W4392284705","doi":"10.4102/aej.v12i1.699","title":"Lessons learned from an occupational therapy programme needs assessment","year":2024,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Occupational Therapy Practice and Research","field":"Health Professions","cited_by":1,"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; Queen's University","funders":"","keywords":"Occupational therapy; Psychology; Medical education; Medicine; Physical therapy","score_opus":0.5994830595183758,"score_gpt":0.6400390485525733,"score_spread":0.04055598903419755,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392284705","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":[0.2792593,0.012113176,0.1197147,0.5002749,0.007073475,0.010318036,0.00041809672,0.00068905804,0.070139326],"genre_scores_gemma":[0.6506793,0.013258148,0.29561707,0.022919519,0.0011271128,0.0044321422,0.00028177956,0.0002169967,0.011467971],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9211783,0.06382173,0.0033414732,0.0017106,0.0068963976,0.0030515091],"domain_scores_gemma":[0.921969,0.048129763,0.0019423426,0.0025832027,0.01651923,0.008856326],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08001234,0.0013396543,0.0011194854,0.0028132857,0.008307635,0.008794754,0.0039784987,0.0040636505,0.0032496636],"category_scores_gemma":[0.10316351,0.00084268406,0.0012068495,0.0015748978,0.0064112768,0.01162399,0.012378862,0.00888689,0.00089889514],"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":[0.000112368514,0.0015404129,0.008723218,0.0036609322,0.000047585858,0.005540242,0.46486434,0.0010671502,0.0016777569,0.011704145,0.030019028,0.47104284],"study_design_scores_gemma":[0.00009606647,0.0011569416,0.008766932,0.013910836,0.00008117995,0.004103829,0.729474,0.0026004761,0.001826095,0.03705535,0.20069021,0.00023818327],"about_ca_topic_score_codex":0.008259176,"about_ca_topic_score_gemma":0.024297696,"teacher_disagreement_score":0.08001234,"about_ca_system_score_codex":0.008024252,"about_ca_system_score_gemma":0.032370355,"threshold_uncertainty_score":0.42315072},"labels":[],"label_agreement":null},{"id":"W4399460746","doi":"10.4102/aej.v12i2.731","title":"Decolonising national evaluation systems","year":2024,"lang":"en","type":"article","venue":"African Evaluation Journal","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":"Francophone University Association","funders":"","keywords":"Computer science","score_opus":0.44198206295271253,"score_gpt":0.5874928836967103,"score_spread":0.14551082074399774,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399460746","genre_codex":"other","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":[0.28658223,0.0025121258,0.14933307,0.075695,0.0006476896,0.0018482325,0.00013498025,0.00070537603,0.48254123],"genre_scores_gemma":[0.9538256,0.00036096343,0.023498159,0.0018530841,0.000055863322,0.00048121368,0.00004833005,0.00009840374,0.019778546],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94785804,0.036589507,0.002477091,0.0038095608,0.0058629923,0.003402809],"domain_scores_gemma":[0.9260508,0.04123847,0.0047162147,0.0122391265,0.01303676,0.0027186712],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.082932815,0.0004674354,0.00046995544,0.0027915915,0.011704069,0.011788542,0.0026198456,0.0025835214,0.0069841207],"category_scores_gemma":[0.07249631,0.00051031506,0.00060321274,0.0018287958,0.03249899,0.012466579,0.018791493,0.0044052233,0.00065390434],"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":[0.000058579804,0.0000915196,0.006324961,0.00037930312,0.000027046488,0.0005364469,0.0875471,0.002729657,0.0010344493,0.79961216,0.008329626,0.0933291],"study_design_scores_gemma":[0.00006231096,0.000284437,0.009086646,0.0020747406,0.000041883777,0.00070049847,0.09503922,0.010363755,0.0030410977,0.31177178,0.5673906,0.00014292506],"about_ca_topic_score_codex":0.010621549,"about_ca_topic_score_gemma":0.014394528,"teacher_disagreement_score":0.91706717,"about_ca_system_score_codex":0.027474234,"about_ca_system_score_gemma":0.031143183,"threshold_uncertainty_score":0.4385959},"labels":[],"label_agreement":null},{"id":"W4404690134","doi":"10.4102/aej.v12i1.758","title":"A rocky road to evidence: Evaluating literacy programmes using a trust-based approach in a context of fragility","year":2024,"lang":"en","type":"article","venue":"African Evaluation Journal","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","funders":"","keywords":"Context (archaeology); Computer science; Data collection; Fragility; Quality (philosophy); Baseline (sea); Data science; Knowledge management; Political science","score_opus":0.4099295910482118,"score_gpt":0.576455307169125,"score_spread":0.16652571612091316,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404690134","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","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":[0.22389627,0.08172827,0.32972762,0.26357487,0.003913144,0.022094417,0.000679584,0.000560709,0.07382514],"genre_scores_gemma":[0.71448106,0.008655936,0.258077,0.008000887,0.00026318082,0.0094309775,0.000108586224,0.000101505895,0.0008807761],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.33118746,0.6152956,0.020180514,0.0042476226,0.025819138,0.0032696116],"domain_scores_gemma":[0.26581094,0.6123998,0.033842526,0.032020282,0.047346443,0.008580032],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.4451979,0.0015195169,0.0031265644,0.010044741,0.0052656056,0.021754466,0.004440198,0.0054720016,0.0051563373],"category_scores_gemma":[0.5600744,0.0009180261,0.0032919277,0.0064308993,0.017445704,0.023179634,0.01569265,0.008436005,0.0006867903],"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":[0.001529446,0.0016560882,0.030146305,0.042137608,0.0023236505,0.00058440276,0.083780326,0.003304191,0.0014042169,0.1415842,0.009689775,0.6818598],"study_design_scores_gemma":[0.001931821,0.013213838,0.067184255,0.28295743,0.0042256843,0.0013793309,0.16549057,0.017052474,0.009210953,0.31805304,0.11843045,0.0008701141],"about_ca_topic_score_codex":0.0050343024,"about_ca_topic_score_gemma":0.0092139,"teacher_disagreement_score":0.4451979,"about_ca_system_score_codex":0.028157976,"about_ca_system_score_gemma":0.0492145,"threshold_uncertainty_score":0.68416977},"labels":[],"label_agreement":null},{"id":"W4413429527","doi":"10.4102/aej.v13i1.791","title":"Evaluation of remote learning in knowledge translation through massive open online courses in the DRC","year":2025,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Online Learning and Analytics","field":"Computer Science","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":"Université du Québec à Montréal; Université de Montréal","funders":"","keywords":"Computer science; Translation (biology); Online learning; Knowledge translation; Massive open online course; World Wide Web; Knowledge management; Chemistry","score_opus":0.16698082584377466,"score_gpt":0.45093246929216957,"score_spread":0.2839516434483949,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413429527","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":[0.9882075,0.00015761086,0.0021699944,0.00019932775,0.000039402752,0.0041043363,0.00014349783,0.00010279396,0.004875573],"genre_scores_gemma":[0.9846342,0.00018183189,0.010094578,0.0001753635,0.000033697805,0.0029112778,0.00025037842,0.000032468794,0.0016862816],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9875367,0.008336061,0.00060487137,0.00087502616,0.0015647212,0.0010826378],"domain_scores_gemma":[0.9747531,0.012573632,0.0017059067,0.0018421571,0.005305511,0.0038196745],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017535113,0.00070456136,0.000602541,0.00084224256,0.0013512573,0.0018693988,0.0014811526,0.0012822127,0.004439751],"category_scores_gemma":[0.030261165,0.00030069833,0.00062777114,0.00056695944,0.0012444728,0.0012037152,0.0027776011,0.0011566784,0.00080083375],"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":[0.020320477,0.0893601,0.04088953,0.0077788276,0.00037417575,0.0019042414,0.054695763,0.008847824,0.03070757,0.0015936729,0.0035341773,0.73999363],"study_design_scores_gemma":[0.013110183,0.4454983,0.29659557,0.0053726914,0.0012975931,0.0017914694,0.056331348,0.019463547,0.08717288,0.0020376854,0.070712656,0.0006161272],"about_ca_topic_score_codex":0.0038046115,"about_ca_topic_score_gemma":0.0052475636,"teacher_disagreement_score":0.017535113,"about_ca_system_score_codex":0.0033546311,"about_ca_system_score_gemma":0.0044584954,"threshold_uncertainty_score":0.09273559},"labels":[],"label_agreement":null}]}