{"meta":{"query_hash":"65d9b4f63912","filters":{"venue":"Assessment and Development Matters"},"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/65d9b4f63912","api":"https://metacan.xera.ac/api/v1/cohort?venue=Assessment+and+Development+Matters"},"results":[{"id":"W2626106683","doi":"10.53841/bpsadm.2011.3.2.28","title":"Re-validation of the Bem Sex Role Inventory","year":2011,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Gender Roles and Identity Studies","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Workplace Health, Safety and Compensation Commission","funders":"","keywords":"Bem Sex-Role Inventory; Computer science; Psychology; Social psychology","score_opus":0.056293444505796236,"score_gpt":0.3037491788556561,"score_spread":0.24745573434985987,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2626106683","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":[0.9554163,0.0006088696,0.015222037,0.0017780518,0.0012160703,0.0020731317,0.0017464929,0.00016282118,0.021776207],"genre_scores_gemma":[0.95026547,0.000417856,0.022525951,0.0012294045,0.00015843245,0.004570038,0.0031437748,0.00030979747,0.017379258],"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.98304915,0.00723326,0.0017762131,0.0014852043,0.005696778,0.0007593578],"domain_scores_gemma":[0.89918756,0.031780846,0.002649588,0.020509714,0.044309434,0.0015629268],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04229905,0.000712501,0.00089185766,0.0023184207,0.001906209,0.00208296,0.0020448498,0.0010480788,0.0038372872],"category_scores_gemma":[0.09553206,0.0005436302,0.0011489755,0.0010509709,0.0017007547,0.0019871746,0.0035194093,0.0021616204,0.0023686816],"study_design_candidate":"bench_or_experimental","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":[0.0014973562,0.0032046288,0.55391735,0.0005090144,0.00031473412,0.0004961285,0.043148883,0.00077126664,0.021706011,0.008614616,0.0154083455,0.35041168],"study_design_scores_gemma":[0.00027283956,0.0017541121,0.8685602,0.00050037407,0.00024897823,0.00083334185,0.01834893,0.003411622,0.014637101,0.00497806,0.08631097,0.00014348786],"about_ca_topic_score_codex":0.009044427,"about_ca_topic_score_gemma":0.013411999,"teacher_disagreement_score":0.04229905,"about_ca_system_score_codex":0.0017372502,"about_ca_system_score_gemma":0.0037916393,"threshold_uncertainty_score":0.22370142},"labels":[],"label_agreement":null},{"id":"W2891427919","doi":"10.53841/bpsadm.2015.7.2.21","title":"Assessing early development and learning across cultures: The East Asia Pacific - Early Child Development Scales","year":2015,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"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":"East Asia; Asia pacific; Geography; History; China; Ethnology; Archaeology","score_opus":0.03748743442441576,"score_gpt":0.34565329359519803,"score_spread":0.30816585917078226,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2891427919","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":[0.94634455,0.0016983672,0.002140194,0.00089483196,0.00020781897,0.0010596063,0.0028636926,0.000063736705,0.044727255],"genre_scores_gemma":[0.9634364,0.0031555155,0.017506726,0.00027653197,0.000045539768,0.0026496344,0.0027948706,0.000034279987,0.010100583],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9986058,0.0002617461,0.00030320746,0.00006749097,0.0005939611,0.00016787878],"domain_scores_gemma":[0.9959591,0.0008239111,0.00087717007,0.0001949686,0.0014796493,0.0006651422],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0042311014,0.00073549437,0.0005420607,0.001433769,0.001032231,0.001186232,0.00079734216,0.0003727164,0.0012733351],"category_scores_gemma":[0.008506258,0.00040697094,0.0010080059,0.0018325904,0.00053215853,0.0016500376,0.0033660799,0.0013710832,0.00027085052],"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":[0.0002416331,0.0004350707,0.837052,0.0003262064,0.00029457005,0.000274615,0.007004046,0.00083599525,0.00096420734,0.0013542586,0.0064933035,0.14472406],"study_design_scores_gemma":[0.000023259934,0.00015423934,0.9843737,0.00012629195,0.00007212787,0.00018148651,0.0035457916,0.00029750547,0.000675132,0.0006358797,0.009889098,0.000025586904],"about_ca_topic_score_codex":0.019601393,"about_ca_topic_score_gemma":0.05807589,"teacher_disagreement_score":0.019601393,"about_ca_system_score_codex":0.0012161841,"about_ca_system_score_gemma":0.0034454819,"threshold_uncertainty_score":0.038974643},"labels":[],"label_agreement":null},{"id":"W398879273","doi":"10.53841/bpsadm.2014.6.2.24","title":"Temporary but not trivial: Language and reading skills in children with ongoing glue ear","year":2014,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Hearing Impairment and Communication","field":"Psychology","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":"Impact","funders":"","keywords":"Reading (process); Psychology; Mathematics education; Audiology; Linguistics; Communication; Medicine; Philosophy","score_opus":0.012002093911324676,"score_gpt":0.30010926055894527,"score_spread":0.2881071666476206,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W398879273","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":[0.9967424,0.00018431488,0.000046272122,0.00013484739,0.000019413903,0.0000093067765,0.00005318808,0.000014194821,0.0027961314],"genre_scores_gemma":[0.99826485,0.0001544912,0.000118068245,0.00008447483,0.0000075749354,0.000023101047,0.00006795885,0.000010907448,0.0012685274],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99892443,0.0001027642,0.00012887391,0.00019997593,0.00032632958,0.00031763825],"domain_scores_gemma":[0.99749005,0.0006886008,0.0008420331,0.00013307687,0.0003043027,0.0005420096],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00094861904,0.0007731071,0.000633798,0.001735065,0.0013007406,0.0022617294,0.0006708631,0.0010032443,0.0047847177],"category_scores_gemma":[0.0075499886,0.00042169206,0.00054539647,0.0011507377,0.002583369,0.0023582848,0.0028664551,0.0014448176,0.0008347231],"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":[0.00043891644,0.0013170535,0.77521086,0.00032442424,0.00006843396,0.056221887,0.08829752,0.00028162688,0.00921151,0.0014498617,0.0033362485,0.06384173],"study_design_scores_gemma":[0.000039269813,0.00084121263,0.8706278,0.00023066354,0.00008085408,0.05389287,0.06886323,0.00022137683,0.0014467095,0.0013086887,0.0023622403,0.00008505204],"about_ca_topic_score_codex":0.017215746,"about_ca_topic_score_gemma":0.020298468,"teacher_disagreement_score":0.017215746,"about_ca_system_score_codex":0.000984754,"about_ca_system_score_gemma":0.0018826311,"threshold_uncertainty_score":0.034231126},"labels":[],"label_agreement":null},{"id":"W4315832394","doi":"10.53841/bpsadm.2012.4.3.17","title":"So what and then what? An integrated approach of vocational assessment and coaching","year":2012,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Educational and Psychological Assessments","field":"Psychology","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":"Canadian Armed Forces","funders":"","keywords":"Coaching; Vocational education; Process (computing); Work (physics); Process management; Psychology; Knowledge management; Applied psychology; Computer science; Engineering; Pedagogy; Mechanical engineering; Psychotherapist","score_opus":0.067947491083666,"score_gpt":0.3945954123103003,"score_spread":0.32664792122663433,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4315832394","genre_codex":"commentary","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":[0.024440592,0.021112673,0.2432812,0.49098298,0.006209104,0.0006429457,0.000061228755,0.0006274259,0.21264188],"genre_scores_gemma":[0.54218286,0.019540222,0.32284454,0.06732114,0.002934857,0.0011552294,0.00011476575,0.0002678449,0.04363858],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.98526996,0.010295588,0.00043125675,0.0008472648,0.0022622424,0.000893678],"domain_scores_gemma":[0.99171126,0.003988683,0.0004473397,0.00056635716,0.0015201175,0.0017662494],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013428855,0.00066668994,0.00105389,0.0030880517,0.0056181718,0.013101417,0.001762461,0.004917424,0.00225642],"category_scores_gemma":[0.015894052,0.0006827065,0.0006920116,0.0014936919,0.016175037,0.018445728,0.007296876,0.010631596,0.0012145205],"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":[0.00008786553,0.00060750725,0.007005752,0.0007098982,0.00010372469,0.0005243144,0.088209786,0.0007309687,0.0015449799,0.4408359,0.034322362,0.425317],"study_design_scores_gemma":[0.00004073892,0.00020031874,0.005832997,0.002405649,0.00006922005,0.0013286954,0.06483253,0.0019918098,0.0008352889,0.6503907,0.27193612,0.00013596033],"about_ca_topic_score_codex":0.006762458,"about_ca_topic_score_gemma":0.018758042,"teacher_disagreement_score":0.013428855,"about_ca_system_score_codex":0.0052437144,"about_ca_system_score_gemma":0.015786443,"threshold_uncertainty_score":0.07101941},"labels":[],"label_agreement":null},{"id":"W4315850908","doi":"10.53841/bpsadm.2016.8.4.14","title":"International Test Commission (ITC) Conference, Vancouver 2016 – Improving policy and practice: Opportunities and challenges in an international context","year":2016,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Higher Education Learning Practices","field":"Social 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":"Context (archaeology); Commission; Test (biology); Political science; Engineering ethics; Public administration; Engineering; Geography; Law","score_opus":0.15775339607159466,"score_gpt":0.42151775009226944,"score_spread":0.2637643540206748,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4315850908","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":[0.042516056,0.011488047,0.0034131007,0.7098354,0.010396056,0.00026928913,0.0012445066,0.00033085368,0.22050667],"genre_scores_gemma":[0.4403464,0.008082379,0.011739403,0.04783597,0.0014276998,0.00031174862,0.0028228008,0.0006664272,0.4867672],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98126465,0.0044202637,0.0008855087,0.00075241853,0.007672785,0.00500432],"domain_scores_gemma":[0.9399135,0.009192224,0.0020344798,0.0028077534,0.024867132,0.02118498],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03192276,0.0005818409,0.00065836956,0.0016442976,0.010256121,0.021844506,0.0026436034,0.011014765,0.020203372],"category_scores_gemma":[0.03627764,0.0005512021,0.00040402074,0.0043834415,0.0089674555,0.0045770654,0.007734697,0.010308685,0.002831292],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00038524918,0.00061852706,0.014166434,0.0005108999,0.000039607876,0.0005803917,0.011436555,0.001150857,0.0019609323,0.09874632,0.6974509,0.17295335],"study_design_scores_gemma":[0.0000402713,0.000116360956,0.030473277,0.0010626419,0.00001890061,0.00013152281,0.021600803,0.00037973892,0.0015176563,0.012767545,0.93182313,0.00006827525],"about_ca_topic_score_codex":0.5301893,"about_ca_topic_score_gemma":0.6460088,"teacher_disagreement_score":0.5301893,"about_ca_system_score_codex":0.029269766,"about_ca_system_score_gemma":0.16278504,"threshold_uncertainty_score":0.9451551},"labels":[],"label_agreement":null},{"id":"W4315850919","doi":"10.53841/bpsadm.2017.9.1.25","title":"Can combining assessments improve test fairness and enhance user experience?","year":2017,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Human-Automation Interaction and Safety","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Association of Physicists","funders":"","keywords":"Test (biology); Computer science; Reliability engineering; Psychology; Process management; Business; Engineering; Geology","score_opus":0.03281881503818991,"score_gpt":0.41408977587759743,"score_spread":0.38127096083940754,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4315850919","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":[0.68230987,0.0020964823,0.27878594,0.005051838,0.001015544,0.00075183494,0.00023312758,0.0026766614,0.027078727],"genre_scores_gemma":[0.9106774,0.00026899204,0.08544931,0.0008974366,0.00026600983,0.00026619053,0.0000997458,0.00019610464,0.0018788712],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.952604,0.029771568,0.003241767,0.0035536676,0.009462602,0.0013664138],"domain_scores_gemma":[0.72400457,0.18943489,0.01584145,0.0341816,0.03055488,0.0059827142],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.051425308,0.0011297147,0.0016937782,0.0021400251,0.0011815323,0.0056907856,0.0018629399,0.0021723164,0.005764836],"category_scores_gemma":[0.27092034,0.00089502655,0.0010852168,0.0012471997,0.001134209,0.008581338,0.0051665986,0.0017098597,0.0016795172],"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":[0.003929975,0.0026019532,0.17158668,0.0005742702,0.0005970793,0.00009092854,0.0036626973,0.0022001287,0.01724415,0.0037336505,0.0028260585,0.7909523],"study_design_scores_gemma":[0.001577746,0.015816234,0.7222164,0.0013693335,0.0023788065,0.0018713928,0.007907448,0.077391945,0.05689342,0.08747916,0.024090573,0.0010074999],"about_ca_topic_score_codex":0.0011047533,"about_ca_topic_score_gemma":0.0020302972,"teacher_disagreement_score":0.051425308,"about_ca_system_score_codex":0.0007963185,"about_ca_system_score_gemma":0.0021077844,"threshold_uncertainty_score":0.27196628},"labels":[],"label_agreement":null},{"id":"W4315851030","doi":"10.53841/bpsadm.2012.4.2.2","title":"It’s not what you know, it’s who you know: The importance of networks","year":2012,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Organizational Learning and Leadership","field":"Business, Management and Accounting","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":"Workplace Health, Safety and Compensation Commission","funders":"","keywords":"Friendship; Need to know; Variety (cybernetics); Public relations; Business; Social network (sociolinguistics); Internet privacy; Sociology; Psychology; Social psychology; Computer science; Political science; Social media; World Wide Web; Computer security","score_opus":0.026954990564052376,"score_gpt":0.25688986040515843,"score_spread":0.22993486984110606,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4315851030","genre_codex":"empirical","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5492273,0.0532151,0.020633928,0.15615141,0.0015239684,0.00010736767,0.0002652791,0.00010487779,0.21877073],"genre_scores_gemma":[0.97800726,0.012619297,0.0021406556,0.0020146414,0.00033748758,0.000037023943,0.000045493958,0.000026150707,0.004771936],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9943234,0.003773687,0.00012539873,0.00036987246,0.0009648599,0.00044282377],"domain_scores_gemma":[0.97540796,0.017661005,0.002520655,0.00054598943,0.0015100287,0.0023543718],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004908351,0.000227096,0.00030175576,0.0011660608,0.0041468167,0.0077132294,0.0006651496,0.0014618671,0.0031876825],"category_scores_gemma":[0.015402234,0.0002707653,0.00014633399,0.0015952004,0.007158974,0.010423492,0.0034134635,0.0020434298,0.00045893193],"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":[0.0002070593,0.00013790284,0.07693358,0.0014130733,0.0000955954,0.0019688678,0.25906804,0.000756025,0.0016721999,0.17979869,0.027205698,0.45074332],"study_design_scores_gemma":[0.000021495238,0.00021871386,0.0752958,0.0028604611,0.00011013115,0.004344771,0.32644224,0.001580524,0.0011774914,0.15532109,0.43250316,0.00012421657],"about_ca_topic_score_codex":0.0052249637,"about_ca_topic_score_gemma":0.006892174,"teacher_disagreement_score":0.0077132294,"about_ca_system_score_codex":0.002054546,"about_ca_system_score_gemma":0.0023760183,"threshold_uncertainty_score":0.02595818},"labels":[],"label_agreement":null},{"id":"W4316083273","doi":"10.53841/bpsadm.2010.2.1.31","title":"Book Review","year":2010,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Future Earth","funders":"","keywords":"Environmental science","score_opus":0.013567624517885436,"score_gpt":0.3073956609154393,"score_spread":0.29382803639755384,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4316083273","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":[0.0004039357,0.05561068,0.0013212555,0.025543276,0.04181561,0.00014449292,0.0006577626,0.00052022364,0.8739828],"genre_scores_gemma":[0.0006619107,0.009332516,0.0002458951,0.0027620823,0.0028463623,0.000025671185,0.00020441368,0.00007931283,0.9838418],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99939716,0.000054097804,0.0000146700295,0.000061149374,0.00042177847,0.000051183906],"domain_scores_gemma":[0.9982956,0.00029737214,0.00006455001,0.00013275331,0.00088206405,0.0003276645],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0006493052,0.00065582414,0.0007268591,0.0024342386,0.0013424552,0.0037149433,0.001081843,0.0018634896,0.37242532],"category_scores_gemma":[0.0032543577,0.00031748397,0.0004121409,0.002462877,0.0006332838,0.0021310241,0.0013300868,0.0022134841,0.3173707],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000037550803,0.000007587902,0.000020833853,0.000044925397,7.7618563e-7,0.000013471758,0.000007365017,0.000021917827,0.00003578733,0.0014500207,0.9502059,0.04818762],"study_design_scores_gemma":[0.0000015810633,0.0000031769393,0.00006443021,0.00005461451,9.650735e-7,0.000026070746,0.000012737835,0.000012784496,0.000017998416,0.0005216967,0.99928254,0.000001477228],"about_ca_topic_score_codex":0.004773575,"about_ca_topic_score_gemma":0.014629363,"teacher_disagreement_score":0.37242532,"about_ca_system_score_codex":0.0015632928,"about_ca_system_score_gemma":0.0028285363,"threshold_uncertainty_score":0.8951591},"labels":[],"label_agreement":null},{"id":"W4316083694","doi":"10.53841/bpsadm.2019.11.1.34","title":"Test user registration amongst educational practitioners","year":2019,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Educational and Psychological Assessments","field":"Psychology","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":"Nortel (Canada)","funders":"","keywords":"Test (biology); Register (sociolinguistics); Quality (philosophy); Quality assurance; Medical education; Psychology; Applied psychology; Medicine","score_opus":0.02988467187185596,"score_gpt":0.3707978010708839,"score_spread":0.34091312919902794,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4316083694","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":[0.93208456,0.00407328,0.0014215307,0.039717574,0.0005381334,0.0003898417,0.00089746446,0.00015571472,0.020721894],"genre_scores_gemma":[0.9661578,0.0022000659,0.0009386069,0.011354146,0.00042864666,0.00031478124,0.00034257607,0.000043002176,0.018220292],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9913833,0.0034781746,0.0008731051,0.0010517209,0.001984819,0.0012287324],"domain_scores_gemma":[0.9407453,0.025588011,0.011080337,0.0043476447,0.006572967,0.011665772],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010253566,0.0001167549,0.00031009625,0.0021511242,0.001747083,0.0016571813,0.00085891946,0.0024594234,0.025918217],"category_scores_gemma":[0.05087514,0.00038683484,0.00026106383,0.0018699575,0.0009009046,0.0021936442,0.0025052226,0.0017625407,0.004228691],"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":[0.00026226696,0.0004060712,0.76868325,0.00060847955,0.0000149666175,0.0013183656,0.043617144,0.000042473064,0.0015312543,0.0012685516,0.046261743,0.13598548],"study_design_scores_gemma":[0.00005761102,0.0010447366,0.8427275,0.0011583795,0.000021839252,0.0038037077,0.05759944,0.00038577663,0.00037604204,0.00062027894,0.092139386,0.00006525533],"about_ca_topic_score_codex":0.006525976,"about_ca_topic_score_gemma":0.007027923,"teacher_disagreement_score":0.025918217,"about_ca_system_score_codex":0.0014582659,"about_ca_system_score_gemma":0.003067853,"threshold_uncertainty_score":0.08670509},"labels":[],"label_agreement":null},{"id":"W4320302550","doi":"10.53841/bpsadm.2018.10.2.22","title":"Assessment in the digital age: Some challenges for Test Developers and Users","year":2018,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Psychometric Methodologies and Testing","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":"ASTER","funders":"","keywords":"Key (lock); Test (biology); Psychometrics; Computer science; Data science; Psychometric testing; Engineering ethics; Psychology; Applied psychology; Engineering; Clinical psychology; Computer security","score_opus":0.40412666237586337,"score_gpt":0.4711374105617989,"score_spread":0.06701074818593555,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4320302550","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004903088,0.009414121,0.022211853,0.9545563,0.0015243423,0.00005684807,0.000038964237,0.0002003607,0.0070941057],"genre_scores_gemma":[0.32055655,0.040532127,0.19929266,0.4122722,0.009824168,0.00092884724,0.0002461679,0.00059086975,0.015756391],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.925756,0.05003898,0.00504027,0.002328443,0.014681492,0.0021548078],"domain_scores_gemma":[0.6471632,0.2676624,0.007438462,0.014925391,0.038508043,0.024302507],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13082942,0.00061938365,0.0010733206,0.0019959062,0.005209227,0.014946193,0.0037056857,0.010829716,0.009276546],"category_scores_gemma":[0.2400442,0.0006669873,0.0007704265,0.0018299278,0.0144299995,0.024095003,0.014370228,0.015737554,0.005310631],"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":[0.00013365546,0.0003458302,0.011406644,0.0019386491,0.0000466942,0.0014132968,0.053722538,0.0007658447,0.0012856603,0.13279888,0.1754751,0.6206672],"study_design_scores_gemma":[0.00007214522,0.00027428544,0.0072036874,0.007354747,0.00004849027,0.005519757,0.123002805,0.0022735791,0.0009636533,0.35350204,0.49953866,0.000246079],"about_ca_topic_score_codex":0.0030881912,"about_ca_topic_score_gemma":0.005646418,"teacher_disagreement_score":0.86917055,"about_ca_system_score_codex":0.0030848049,"about_ca_system_score_gemma":0.012197465,"threshold_uncertainty_score":0.6919004},"labels":[],"label_agreement":null},{"id":"W4320399006","doi":"10.53841/bpsadm.2014.6.3.11","title":"Enabling individuals who have specific difficulties in the workplace: Part 1","year":2014,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Human Resource Development and Performance Evaluation","field":"Psychology","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":"Workplace Health, Safety and Compensation Commission","funders":"","keywords":"Strengths and weaknesses; Perception; Focus (optics); Psychology; Engineering ethics; Public relations; Political science; Sociology; Social psychology; Engineering","score_opus":0.046870376170526064,"score_gpt":0.3302876971337332,"score_spread":0.2834173209632071,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4320399006","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":[0.42455646,0.08923068,0.018598827,0.26551163,0.021656971,0.0033064706,0.00048038785,0.00022531842,0.17643328],"genre_scores_gemma":[0.6845315,0.14668901,0.022566367,0.066253446,0.0051537575,0.005713706,0.00057257584,0.00014842006,0.06837113],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9981957,0.0006413091,0.00017278797,0.000113573,0.00047137728,0.0004052767],"domain_scores_gemma":[0.9964204,0.0013966194,0.00060901325,0.00014364634,0.0004749839,0.0009553686],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0030328613,0.00053847476,0.000520489,0.0009821085,0.0029251853,0.003153119,0.00088155287,0.003446895,0.0057358784],"category_scores_gemma":[0.0065123173,0.00030523192,0.0004671739,0.00091762334,0.0020091552,0.0047453553,0.005337939,0.0035120407,0.00120767],"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":[0.0000788206,0.0022066387,0.041594375,0.0066908626,0.00004638766,0.0036261727,0.15769015,0.00032566983,0.0045094355,0.02692725,0.1844482,0.5718561],"study_design_scores_gemma":[0.000023315077,0.0007589551,0.080811895,0.0069514415,0.00007847496,0.0056491653,0.23251042,0.00015874131,0.0025708224,0.009287299,0.66109025,0.00010929073],"about_ca_topic_score_codex":0.0012837627,"about_ca_topic_score_gemma":0.005900982,"teacher_disagreement_score":0.0057358784,"about_ca_system_score_codex":0.0010387884,"about_ca_system_score_gemma":0.004998334,"threshold_uncertainty_score":0.019188404},"labels":[],"label_agreement":null},{"id":"W4320399249","doi":"10.53841/bpsadm.2014.6.3.28","title":"Inconsistency in dyslexia assessment","year":2014,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Reading and Literacy Development","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Dyslexia; Psychology; Natural language processing; Cognitive psychology; Computer science; Linguistics; Reading (process); Philosophy","score_opus":0.019039781001349662,"score_gpt":0.33675069962742893,"score_spread":0.31771091862607925,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4320399249","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":[0.89992356,0.013793556,0.04383676,0.011459157,0.00076114794,0.00034483857,0.00036334354,0.00031794992,0.029199615],"genre_scores_gemma":[0.9885853,0.00051981007,0.008821905,0.0010166763,0.00014839176,0.00008944436,0.000107953485,0.000090085894,0.00062047737],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.8734409,0.069104224,0.019268226,0.0093266,0.02737167,0.0014883311],"domain_scores_gemma":[0.46549803,0.43686944,0.029171152,0.014409663,0.05192115,0.002130512],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.10933344,0.0002977821,0.0012145408,0.006550064,0.0013588896,0.0054422757,0.0023603993,0.0015015805,0.0013557044],"category_scores_gemma":[0.43982685,0.00087501615,0.0006207961,0.00320852,0.00308327,0.004950738,0.0040020416,0.0022674503,0.00026876226],"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":[0.0007026389,0.00020812199,0.76207453,0.00064162177,0.00044850336,0.0005912898,0.008722385,0.0015288648,0.0008868205,0.013241952,0.0028199365,0.20813338],"study_design_scores_gemma":[0.00024412644,0.0011167488,0.8187607,0.003559477,0.0013193594,0.00862725,0.015294549,0.020398516,0.00890475,0.10019883,0.021262918,0.00031286813],"about_ca_topic_score_codex":0.004297437,"about_ca_topic_score_gemma":0.007512984,"teacher_disagreement_score":0.10933344,"about_ca_system_score_codex":0.0041587492,"about_ca_system_score_gemma":0.0044550574,"threshold_uncertainty_score":0.5782174},"labels":[],"label_agreement":null},{"id":"W4320400465","doi":"10.53841/bpsadm.2015.7.3.7","title":"Assessment of child development across cultures","year":2015,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"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":"Geography","score_opus":0.03323903704630559,"score_gpt":0.3882292987304678,"score_spread":0.3549902616841622,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4320400465","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":[0.9595594,0.0013449247,0.0017243329,0.00037138918,0.00005307858,0.00008723989,0.00036744814,0.000020992295,0.036471292],"genre_scores_gemma":[0.99602497,0.0006911508,0.0016466461,0.00006111148,0.000005534297,0.00008751178,0.00013309423,0.000010497292,0.0013394922],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.992044,0.003934294,0.00083279185,0.0006577586,0.0019001991,0.00063091615],"domain_scores_gemma":[0.988674,0.0031346828,0.0012180011,0.0014799512,0.004847839,0.0006456857],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006253832,0.00046755152,0.0005567301,0.0034735252,0.002006056,0.0035957051,0.0009163923,0.00035849167,0.0014297672],"category_scores_gemma":[0.026036303,0.00030894394,0.00044991056,0.0038549514,0.0019328626,0.0018896159,0.005583832,0.0007977734,0.00026985607],"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":[0.00020244076,0.00016438676,0.7469601,0.0002693112,0.00019240947,0.00060817425,0.093291596,0.00052948453,0.0033398774,0.005075742,0.0010697978,0.14829668],"study_design_scores_gemma":[0.0000076246197,0.00021295586,0.89281285,0.00040523685,0.00008141565,0.00056291185,0.093250714,0.00029916334,0.0037239697,0.0023087873,0.0062902467,0.000044200213],"about_ca_topic_score_codex":0.056372434,"about_ca_topic_score_gemma":0.06693182,"teacher_disagreement_score":0.056372434,"about_ca_system_score_codex":0.0029940568,"about_ca_system_score_gemma":0.0036400277,"threshold_uncertainty_score":0.11208862},"labels":[],"label_agreement":null},{"id":"W4320400475","doi":"10.53841/bpsadm.2015.7.4.13","title":"Testing dyslexic candidates using CAT administration: Taking time out of the test equation","year":2015,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Intelligent Tutoring Systems and Adaptive Learning","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":"Workplace Health, Safety and Compensation Commission","funders":"","keywords":"Test (biology); Computerized adaptive testing; Administration (probate law); Dyslexia; Division (mathematics); Psychology; Computer science; Audiology; Cognitive psychology; Developmental psychology; Reading (process); Medicine; Arithmetic; Psychometrics; Mathematics; Political science; Law","score_opus":0.11319900118690496,"score_gpt":0.318694181219585,"score_spread":0.20549518003268002,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4320400475","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":[0.99463356,0.00020942357,0.0015165675,0.0003960225,0.000045836176,0.00004816039,0.00006713448,0.000027682097,0.0030555632],"genre_scores_gemma":[0.9964347,0.00013233435,0.0018527901,0.00012463333,0.000015447118,0.000023992048,0.000084339576,0.0000096450485,0.0013221765],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.99776936,0.0008125816,0.00029064371,0.00024327912,0.0007061953,0.00017799562],"domain_scores_gemma":[0.98396456,0.0075958488,0.0045279865,0.0008707724,0.0018031762,0.001237485],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0028502187,0.00027083087,0.0004207597,0.0006543071,0.00058539794,0.0011972333,0.00058618095,0.00039016714,0.0030063232],"category_scores_gemma":[0.020677226,0.00016177549,0.00027097293,0.00047377034,0.00034879227,0.0009354455,0.0007208803,0.0007586335,0.0006520784],"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":[0.00082526397,0.0012156084,0.87270063,0.00013506906,0.000057943205,0.0004095177,0.0057036285,0.00017750228,0.0049253106,0.00055777,0.0014483377,0.11184328],"study_design_scores_gemma":[0.00004636262,0.0022937292,0.97800523,0.00011694487,0.0001153361,0.0011184347,0.005967665,0.001948245,0.005202224,0.00041009547,0.004740395,0.000035359768],"about_ca_topic_score_codex":0.00546282,"about_ca_topic_score_gemma":0.015207897,"teacher_disagreement_score":0.00546282,"about_ca_system_score_codex":0.000489223,"about_ca_system_score_gemma":0.0013001455,"threshold_uncertainty_score":0.015073597},"labels":[],"label_agreement":null},{"id":"W4386561694","doi":"10.53841/bpsadm.2023.15.3.4","title":"Investigating the presence of psychopathy bias in the general population and its impact on judgements of guilt","year":2023,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Personality Traits and Psychology","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Child, Adolescent and Family Mental Health","funders":"","keywords":"Psychopathy; Psychology; Population; Social psychology; Personality; Sociology; Demography","score_opus":0.12440099407185333,"score_gpt":0.43508871714243397,"score_spread":0.3106877230705806,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386561694","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":[0.99943334,0.000056367626,0.00004197196,0.000042638287,0.0000043987725,0.0000038918633,0.0000147703895,7.7036515e-7,0.00040182375],"genre_scores_gemma":[0.999775,0.000026642958,0.00003660892,0.000021858954,0.0000051550915,0.0000024557412,0.00001974118,0.0000010111161,0.00011150114],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9972006,0.0013559487,0.0001742507,0.00039388356,0.0006190955,0.000256204],"domain_scores_gemma":[0.9781431,0.011051314,0.0054019988,0.00189513,0.0022993276,0.0012091263],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0051399455,0.00029983238,0.0004382701,0.0011692947,0.0007667376,0.0012604551,0.00058481935,0.000791788,0.0028459677],"category_scores_gemma":[0.041251123,0.0002961863,0.00041584842,0.0007232495,0.0013551313,0.0010042335,0.0010009988,0.0010045384,0.0003262628],"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":[0.00015436117,0.00012833619,0.99413896,0.0000122788015,0.000049834165,0.00013900675,0.0009868164,0.000018371453,0.00021762206,0.000054320906,0.00008195972,0.004018096],"study_design_scores_gemma":[0.00000561772,0.00023308097,0.998178,0.0000054253296,0.000018602881,0.00043848384,0.0007631371,0.00007579418,0.00006849568,0.00014269604,0.00006683667,0.0000039107827],"about_ca_topic_score_codex":0.0037311797,"about_ca_topic_score_gemma":0.005264399,"teacher_disagreement_score":0.0051399455,"about_ca_system_score_codex":0.00041973474,"about_ca_system_score_gemma":0.00047408044,"threshold_uncertainty_score":0.027182937},"labels":[],"label_agreement":null},{"id":"W4391998433","doi":"10.53841/bpsadm.2024.16.1.48","title":"Large-scale testing in the face of AI","year":2024,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"First Nations University of Canada; Brock University","funders":"","keywords":"Expansive; Test (biology); Computer science; Construct (python library); Scale (ratio); Face validity; Set (abstract data type); Test design; Face (sociological concept); Construct validity; Data science; Psychology; Political science; Sociology; Psychometrics; Social science; Law","score_opus":0.1316125422972995,"score_gpt":0.44305571482263406,"score_spread":0.3114431725253346,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391998433","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":[0.23957267,0.0039148447,0.5133101,0.15041713,0.0025340386,0.0058329008,0.00044310113,0.0033928074,0.08058237],"genre_scores_gemma":[0.8051476,0.00057828764,0.17912978,0.007441495,0.0006659002,0.002395849,0.0002130089,0.0002778608,0.0041500935],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.8320197,0.13832001,0.003757276,0.0045777685,0.019638125,0.0016871004],"domain_scores_gemma":[0.34900773,0.5330047,0.013347119,0.0574923,0.03912006,0.008028042],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.20393395,0.00060643884,0.0007616223,0.0014992566,0.0022959865,0.0059806732,0.0046417555,0.0023493636,0.009974803],"category_scores_gemma":[0.3846985,0.00039849294,0.00054013694,0.0009653433,0.007835096,0.009601115,0.0071404898,0.004986406,0.00154151],"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":[0.0008369401,0.0018948184,0.037995078,0.001399291,0.00014850461,0.0012006004,0.01658512,0.008458165,0.0059872763,0.10474487,0.034685776,0.7860635],"study_design_scores_gemma":[0.0006234025,0.007439803,0.083181165,0.0050776964,0.0001906332,0.004353817,0.04319122,0.057460167,0.017313367,0.56564885,0.21513699,0.0003828592],"about_ca_topic_score_codex":0.0015958551,"about_ca_topic_score_gemma":0.0030116201,"teacher_disagreement_score":0.20393395,"about_ca_system_score_codex":0.0038863327,"about_ca_system_score_gemma":0.009955844,"threshold_uncertainty_score":0.9816912},"labels":[],"label_agreement":null}]}