{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":30,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":30,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"3be08ccc75e3","filters":{"venue":"Assessment in Education Principles Policy and Practice"}},"results":[{"id":"W2600733185","doi":"10.1080/0969594x.2017.1297010","title":"Developing assessment capable teachers in this age of accountability","year":2017,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":75,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Accountability; Psychology; Medical education; Political science; Medicine","authors":[{"name":"Christopher DeLuca","is_ca":true},{"name":"Sandra Johnson","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1225845974022681,"gpt":0.5148076355503635,"spread":0.3922230381480953,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005457025,0.0001511705,0.0002598889,0.0002510282,0.0006227701,0.0003924616,0.000512693,0.0001279713,0.0001255186],"category_scores_gemma":[0.003231292,0.0001624988,0.00003586689,0.0002870877,0.0003158816,0.002027341,0.000201889,0.0003605561,0.000003886596],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00124927,"about_ca_system_score_gemma":0.0057436,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.09320701,"about_ca_topic_score_gemma":0.01843984,"domain_scores_codex":[0.9975737,0.0007353768,0.0004992859,0.0003345431,0.0004962453,0.0003608429],"domain_scores_gemma":[0.9973277,0.001351779,0.000623671,0.0004520884,0.0001548377,0.00008987364],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001524136,0.0005197534,0.5562643,0.00003339005,0.0000186363,0.000001585196,0.02971618,0.000005175718,0.00009025372,0.4065339,0.00007300705,0.006728579],"study_design_scores_gemma":[0.0003475698,0.00001998301,0.7810145,0.00007257796,0.00001658217,0.000001020581,0.05467653,0.00001478507,0.00001131535,0.003633609,0.1600303,0.0001612413],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5109373,0.00006620627,0.00006757063,0.02198634,0.0003839033,0.000487625,0.000002447362,0.0000147035,0.4660539],"genre_scores_gemma":[0.9771204,0.00136459,0.01853381,0.0004878001,0.0002106914,0.0001035273,0.00001239934,0.0000102076,0.002156608],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.466183,"threshold_uncertainty_score":0.9998929,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1996435010","doi":"10.1080/09695940701272773","title":"Did we take the same test? Differing accounts of the Ontario Secondary School Literacy Test by first and second language test‐takers","year":2007,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":74,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University; Carleton University","funders":"","keywords":"Test (biology); Graduation (instrument); Psychology; Context (archaeology); Construct (python library); Literacy; Construct validity; Mathematics education; Fidelity; Test score; Standardized test; Pedagogy; Developmental psychology; Computer science; Psychometrics; Mathematics","authors":[{"name":"Janna Fox","is_ca":true},{"name":"Liying Cheng","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02100229501179931,"gpt":0.3845813286896781,"spread":0.3635790336778789,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002388958,0.0001786905,0.0001731892,0.0001097142,0.0006577378,0.0003841227,0.0003783061,0.00009920697,0.0003271223],"category_scores_gemma":[0.004567929,0.0001313146,0.00003937975,0.0004546433,0.0002520351,0.001184283,0.000179183,0.0005769877,0.000004157873],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005247691,"about_ca_system_score_gemma":0.0020187,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.04493859,"about_ca_topic_score_gemma":0.1139421,"domain_scores_codex":[0.9983174,0.0002050183,0.0004182057,0.0002808006,0.0004265653,0.0003519708],"domain_scores_gemma":[0.9892814,0.009708045,0.0004543165,0.0003119002,0.0001122827,0.0001321193],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001662022,0.0006631211,0.8745536,0.00004868594,0.00002720056,9.822992e-7,0.1114054,5.882509e-7,0.0006239841,0.003228653,0.001569173,0.007861994],"study_design_scores_gemma":[0.0002352006,0.00003407097,0.6397342,0.00005857932,0.00002670472,0.000005609869,0.04729789,0.000003764079,0.00004907619,0.0002276095,0.3122129,0.0001143526],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8785336,0.000717962,0.000005737945,0.03486021,0.0003170077,0.0005951427,0.00003316722,0.0000180155,0.08491918],"genre_scores_gemma":[0.9832307,0.0009199142,0.000621495,0.002271085,0.0003312866,0.00004199333,0.00001052814,0.00001383392,0.01255913],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3106437,"threshold_uncertainty_score":0.9614213,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2923740196","doi":"10.1080/0969594x.2019.1593105","title":"Conceptualising fairness in classroom assessment: exploring the value of organisational justice theory","year":2019,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":71,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Foundation (evidence); Economic Justice; Scholarship; Value (mathematics); Core (optical fiber); Sociology; Psychology; Social psychology; Epistemology; Political science; Computer science; Law","authors":[{"name":"Amirhossein Rasooli","is_ca":true},{"name":"Hamed Zandi","is_ca":false},{"name":"Christopher DeLuca","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07905727411628827,"gpt":0.4421515498552321,"spread":0.3630942757389438,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005509842,0.0001531339,0.0002152979,0.0002652954,0.000302508,0.0001768171,0.0003539386,0.00008157531,0.000181437],"category_scores_gemma":[0.001945079,0.0001376627,0.00003591063,0.0008488765,0.0002238656,0.001742413,0.0001319266,0.0004014618,0.00001018672],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005734127,"about_ca_system_score_gemma":0.003732229,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003472368,"about_ca_topic_score_gemma":0.0003561579,"domain_scores_codex":[0.996267,0.001942137,0.0005124431,0.0003023901,0.0006499005,0.000326108],"domain_scores_gemma":[0.991627,0.007442335,0.0004194252,0.0002589938,0.0001827617,0.0000694409],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00002865085,0.0003229735,0.1066119,0.0000401553,0.00002059672,4.183273e-7,0.03777248,0.0001840673,0.0001230797,0.8522458,0.00002172318,0.002628144],"study_design_scores_gemma":[0.0005097623,0.00003842234,0.563396,0.00009847654,0.00006077866,0.000002889843,0.3773915,0.0001447901,0.00001961376,0.01490691,0.04324486,0.0001859801],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7657065,0.0001308024,0.0003280732,0.01375263,0.001058244,0.0006923455,0.000003794297,0.00002087156,0.2183067],"genre_scores_gemma":[0.992701,0.0008593524,0.003746679,0.001014333,0.0003976421,0.0001081556,0.0000139136,0.00001486204,0.001144084],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8373389,"threshold_uncertainty_score":0.6620815,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2013464569","doi":"10.1080/0969594x.2013.776943","title":"Fair and equitable assessment practices for all students","year":2013,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":71,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Thompson Rivers University; University of Lethbridge; University of Alberta; University of Calgary","funders":"","keywords":"Intrusiveness; Equity (law); Psychology; Public relations; Best practice; Medical education; Applied psychology; Pedagogy; Social psychology; Political science; Medicine","authors":[{"name":"Shelleyann Scott","is_ca":true},{"name":"Charles F. Webber","is_ca":true},{"name":"Judy Lupart","is_ca":true},{"name":"Nola Aitken","is_ca":true},{"name":"Donald E. Scott","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1023168743994288,"gpt":0.5279200270248844,"spread":0.4256031526254556,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.004179651,0.0002085413,0.0002371054,0.0002521336,0.0006425898,0.001184108,0.0003592938,0.000132958,0.0001406804],"category_scores_gemma":[0.003173401,0.0002118671,0.0000362212,0.0003464833,0.0001532799,0.003488443,0.0002624991,0.0002812839,0.00001477248],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004633966,"about_ca_system_score_gemma":0.002181478,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0196238,"about_ca_topic_score_gemma":0.001214983,"domain_scores_codex":[0.997134,0.0006968656,0.0004453826,0.0004651367,0.000657997,0.0006006595],"domain_scores_gemma":[0.994364,0.004074993,0.0007698587,0.0002361321,0.0003138615,0.0002411873],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004274996,0.001842809,0.2229104,0.00009676772,0.0001433305,8.524243e-7,0.01804011,0.000005288103,0.000235441,0.7311828,0.003819997,0.0216795],"study_design_scores_gemma":[0.0006782885,0.0001218495,0.2919045,0.00003538582,0.00008289351,0.000004646491,0.08186039,0.00009393153,0.000003673005,0.007085557,0.6178655,0.0002633843],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4649363,0.0006306556,0.0005220761,0.1363678,0.001022394,0.004128053,0.00001065014,0.0001004109,0.3922817],"genre_scores_gemma":[0.9457006,0.004114734,0.03419917,0.003967003,0.0007121699,0.001296294,0.00003454959,0.00002473072,0.009950678],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7240973,"threshold_uncertainty_score":0.9998528,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2026041656","doi":"10.1080/09695940802164226","title":"Educational assessment in Canada","year":2008,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":66,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Brock University","funders":"Social Sciences and Humanities Research Council of Canada; Brock University","keywords":"Accountability; Independence (probability theory); Bureaucracy; Political science; Scale (ratio); Public administration; Multiculturalism; Educational assessment; Regional science; Economic growth; Geography; Sociology; Pedagogy; Economics; Law","authors":[{"name":"Louis Volante","is_ca":true},{"name":"Sonia Ben Jaafar","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1406251533691962,"gpt":0.4873947963953773,"spread":0.3467696430261811,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003451661,0.0002420646,0.0003079517,0.0007748446,0.0002619664,0.000187493,0.0005265267,0.00006428916,0.0005262162],"category_scores_gemma":[0.005543177,0.0002249535,0.00003959303,0.001473195,0.00008588162,0.00165235,0.0001753748,0.0005170541,0.00002473625],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00340097,"about_ca_system_score_gemma":0.07757267,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.701249,"about_ca_topic_score_gemma":0.5307347,"domain_scores_codex":[0.9952347,0.0008067368,0.001220214,0.0006697995,0.00163473,0.0004337633],"domain_scores_gemma":[0.9899126,0.00821395,0.0006158214,0.0005501459,0.0004694066,0.0002380547],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002175097,0.001274644,0.6957066,0.00001134497,0.00001565164,0.000004390798,0.001261436,0.0002466888,0.00007357115,0.2831981,0.006912924,0.01127289],"study_design_scores_gemma":[0.0003232677,0.00003405358,0.7957795,0.00002029313,0.000005895901,0.00006660655,0.01154044,0.0004376271,0.000009466598,0.01391034,0.1776639,0.0002086495],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7096179,0.0002043839,0.00009488407,0.16283,0.001205648,0.0005285803,0.00001498225,0.000008729876,0.125495],"genre_scores_gemma":[0.9741686,0.0005398653,0.01223469,0.006482939,0.0004153137,0.0003016885,0.00005448835,0.00001459151,0.005787834],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2692878,"threshold_uncertainty_score":0.9276567,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1978227392","doi":"10.1080/09695940903565362","title":"Teacher beliefs about the cognitive diagnostic information of classroom‐ versus large‐scale tests: implications for assessment literacy","year":2010,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":58,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Smiths Detection (Canada); University of Alberta","funders":"","keywords":"Test (biology); Mathematics education; Scale (ratio); Psychology; Literacy; Pedagogy","authors":[{"name":"Jacqueline P. Leighton","is_ca":true},{"name":"Rebecca Gokiert","is_ca":true},{"name":"M. Ken Cor","is_ca":false},{"name":"Colleen Heffernan","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05055241862711323,"gpt":0.48032179247732,"spread":0.4297693738502068,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002766606,0.0001561551,0.0001699181,0.0002073317,0.0006982924,0.0003364245,0.0003034465,0.0001319997,0.00006552644],"category_scores_gemma":[0.01135278,0.0001374429,0.00005792285,0.0005444637,0.0002056598,0.002331659,0.00009863017,0.0004452426,0.000006438774],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002192161,"about_ca_system_score_gemma":0.00280768,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001537609,"about_ca_topic_score_gemma":0.002459566,"domain_scores_codex":[0.9981667,0.000408867,0.0005237744,0.0002166098,0.0003367128,0.0003472974],"domain_scores_gemma":[0.9828265,0.01552949,0.0006018049,0.0002712926,0.0006609694,0.0001100038],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00008061616,0.001351589,0.1726081,0.00004318819,0.00006532221,5.765668e-8,0.05189255,0.000003925003,0.00007360176,0.7408808,0.0008353708,0.03216486],"study_design_scores_gemma":[0.0008289103,0.00007522754,0.6145263,0.00003505056,0.00008254273,0.000001276393,0.05829775,0.00004811789,0.00000421243,0.002400313,0.3235689,0.0001314548],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5479867,0.0002252819,0.00292921,0.1355857,0.002199988,0.004179873,0.0002059455,0.00008172852,0.3066056],"genre_scores_gemma":[0.9870275,0.0009119121,0.008294825,0.001265766,0.0007343136,0.001030403,0.0001699097,0.00001185452,0.0005534856],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7384805,"threshold_uncertainty_score":0.996975,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4322503168","doi":"10.1080/0969594x.2023.2182737","title":"Data literacy assessments: a systematic literature review","year":2023,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":57,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Literacy; Reliability (semiconductor); Computer science; Information literacy; Field (mathematics); Quality (philosophy); Systematic review; Psychology; Medical education; Data science; Mathematics education; Pedagogy; Political science; Medicine; MEDLINE","authors":[{"name":"Ying Cui","is_ca":true},{"name":"Fu Chen","is_ca":false},{"name":"Alina Lutsyk","is_ca":true},{"name":"Jacqueline P. Leighton","is_ca":true},{"name":"Maria Cutumisu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2563835509897077,"gpt":0.588095050840592,"spread":0.3317114998508843,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.01693645,0.0002879606,0.0005319071,0.001112119,0.0002346311,0.001416578,0.001494548,0.00009092638,0.0001802076],"category_scores_gemma":[0.02707397,0.0002244812,0.00006092557,0.00481967,0.00005403433,0.005758511,0.0008481569,0.0004486456,0.0002416726],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002923018,"about_ca_system_score_gemma":0.003166376,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008937122,"about_ca_topic_score_gemma":0.00002500518,"domain_scores_codex":[0.9932129,0.001977629,0.001650778,0.0009884859,0.001748299,0.0004219318],"domain_scores_gemma":[0.9839876,0.01162622,0.00110246,0.002156306,0.0009086838,0.0002187508],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00006922458,0.00368059,0.01459902,0.04968821,0.0002824512,0.00003850583,0.005548608,0.00003232995,0.0001581655,0.6062237,0.1742519,0.1454274],"study_design_scores_gemma":[0.0004375722,0.0001243095,0.0432422,0.02900849,0.0001667067,0.0001674075,0.0112015,0.004123088,0.000003251252,0.02032772,0.8906622,0.0005355408],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.03455831,0.1235802,0.00164963,0.6338046,0.007024258,0.01093865,0.0007978038,0.0005004221,0.1871461],"genre_scores_gemma":[0.2766623,0.4786966,0.06798269,0.09411259,0.002858414,0.003702938,0.005563969,0.000144727,0.07027572],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.7164104,"threshold_uncertainty_score":0.99962,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2124002552","doi":"10.1080/0969594x.2014.967168","title":"Instructional Rounds as a professional learning model for systemic implementation of Assessment for Learning","year":2014,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":52,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University","funders":"","keywords":"Professional learning community; Professional development; Psychology; Value (mathematics); Pedagogy; Mathematics education; Session (web analytics); Medical education; Medicine; Computer science","authors":[{"name":"Christopher DeLuca","is_ca":true},{"name":"Don A. Klinger","is_ca":true},{"name":"Jamie S. Pyper","is_ca":true},{"name":"Judy C. Woods","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05882398951401478,"gpt":0.5102705274787012,"spread":0.4514465379646864,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004186619,0.0001440104,0.0002139618,0.0002625659,0.0007346452,0.0001373319,0.0001540636,0.0001069032,0.00002831178],"category_scores_gemma":[0.001689007,0.0001549419,0.00006198201,0.0002695712,0.00008322777,0.0009538442,0.00005112768,0.0002593528,8.886276e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005231106,"about_ca_system_score_gemma":0.004707297,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001507303,"about_ca_topic_score_gemma":0.0002810442,"domain_scores_codex":[0.9976763,0.0007177723,0.0005208298,0.0003021267,0.0004740259,0.0003090028],"domain_scores_gemma":[0.9958019,0.002863585,0.000699274,0.00009800112,0.0004477827,0.00008950686],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00007823219,0.0002797426,0.03807133,0.0001374218,0.00004448113,4.582871e-8,0.01382124,0.002459109,0.0003312451,0.9211442,0.00009349727,0.02353945],"study_design_scores_gemma":[0.004096812,0.0007755842,0.09652027,0.0002966867,0.0002387414,0.0000309858,0.4364676,0.09097537,0.0000478515,0.03774159,0.3321038,0.0007047071],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8478959,0.00008822353,0.07919016,0.02741966,0.001110475,0.003647062,0.00001331097,0.00008674652,0.04054844],"genre_scores_gemma":[0.9406703,0.0002168565,0.05342697,0.0003297179,0.0005111466,0.0009101594,0.0001089054,0.00001760006,0.003808413],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8834026,"threshold_uncertainty_score":0.8350543,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3210068963","doi":"10.1080/0969594x.2021.1988510","title":"Formative assessment, growth mindset, and achievement: examining their relations in the East and the West","year":2021,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":50,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Mindset; Formative assessment; Psychology; Reading (process); Mainland China; Academic achievement; Pedagogy; Mathematics education; China; Political science; Computer science","authors":[{"name":"Zi Yan","is_ca":false},{"name":"Ronnel B. King","is_ca":false},{"name":"Joseph Y. Haw","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07796939436320567,"gpt":0.4219319294244714,"spread":0.3439625350612658,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004968341,0.0001533237,0.0001754987,0.0001525426,0.0008857095,0.0005765224,0.0002025273,0.00007532888,0.00003403229],"category_scores_gemma":[0.00169857,0.0001064081,0.00002166024,0.0007078957,0.0003685238,0.001411472,0.0001633556,0.0004507705,0.000001956606],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001779863,"about_ca_system_score_gemma":0.00148272,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002764485,"about_ca_topic_score_gemma":0.003752414,"domain_scores_codex":[0.9964328,0.002292171,0.0003665297,0.0002704898,0.0003588379,0.0002791052],"domain_scores_gemma":[0.9934272,0.005889819,0.000291428,0.0001922377,0.0001313969,0.00006790744],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.000018805,0.0002917472,0.1229567,0.00001306644,0.00002874021,0.000001213762,0.1644873,0.000001962929,0.000008965343,0.7088524,0.00009091223,0.003248253],"study_design_scores_gemma":[0.0006954217,0.00002293805,0.5520095,0.00003515187,0.00003501994,0.00001297605,0.4042856,0.00007671639,0.000001066389,0.006006738,0.0367156,0.0001032608],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5167763,0.0006542549,0.0001164499,0.1662666,0.000211887,0.0007681958,0.000008700095,0.00001501695,0.3151826],"genre_scores_gemma":[0.9904081,0.003560904,0.001925982,0.002895866,0.0002140522,0.0001988847,0.00002652764,0.000008007668,0.0007617036],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7028456,"threshold_uncertainty_score":0.6812251,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3081330604","doi":"10.1080/0969594x.2020.1801576","title":"Collaborating with teachers to design and implement assessments for self-regulated learning in the context of authentic classroom writing tasks","year":2020,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":41,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Context (archaeology); Task (project management); Mathematics education; Psychology; Self-regulated learning; Quality (philosophy); Pedagogy; Computer science; Engineering","authors":[{"name":"Nancy E. Perry","is_ca":true},{"name":"Simon Lisaingo","is_ca":true},{"name":"Nikki Yee","is_ca":true},{"name":"Natasha Parent","is_ca":true},{"name":"Xinke Wan","is_ca":true},{"name":"Krista R. Muis","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1044167634101325,"gpt":0.4946963088082171,"spread":0.3902795453980847,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00775729,0.0001549452,0.00022034,0.0001998185,0.0001896689,0.0001020855,0.0001273782,0.00006522617,0.00001008346],"category_scores_gemma":[0.002607319,0.0001284255,0.00001320019,0.0007722962,0.00003214009,0.0002180214,0.00004196748,0.0005504377,6.578499e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001374952,"about_ca_system_score_gemma":0.0007598583,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006374739,"about_ca_topic_score_gemma":0.00003739311,"domain_scores_codex":[0.994503,0.004238745,0.0004822578,0.0003308169,0.0001778276,0.0002673522],"domain_scores_gemma":[0.9937207,0.005434212,0.0004741483,0.0001425507,0.0001648065,0.00006357473],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0008541634,0.0009990353,0.1332768,0.0001846966,0.0002559451,0.00000361038,0.5294727,0.001196566,0.001726403,0.2017166,0.000331353,0.1299821],"study_design_scores_gemma":[0.002686199,0.002535509,0.2733774,0.0001999828,0.0001426199,0.00003712585,0.6137503,0.00406005,0.00009729156,0.0003075048,0.1024202,0.0003857483],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9182208,0.000079581,0.04620251,0.03052997,0.00007935795,0.001553156,0.000004358566,0.00003769673,0.003292631],"genre_scores_gemma":[0.8590552,0.00000734075,0.1378757,0.002583579,0.00007857729,0.0002832102,0.00001268557,0.00001808044,0.00008564818],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2014091,"threshold_uncertainty_score":0.5237039,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3004711011","doi":"10.1080/0969594x.2020.1719033","title":"Leveraging assessment to promote kindergarten learners’ independence and self-regulation within play-based classrooms","year":2020,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Queen's University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Curriculum; Leverage (statistics); Psychology; Pedagogy; Early childhood education; Independence (probability theory); Assessment for learning; Mathematics education; Formative assessment; Computer science","authors":[{"name":"Christopher DeLuca","is_ca":true},{"name":"Angela Pyle","is_ca":true},{"name":"Heather Braund","is_ca":true},{"name":"Laurie Faith","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04841793565652756,"gpt":0.3940303882998441,"spread":0.3456124526433165,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002244271,0.0001755362,0.0001685294,0.0002441187,0.0004982472,0.0004164467,0.0001703799,0.0001203035,0.00007749653],"category_scores_gemma":[0.001756252,0.0001929503,0.00001987975,0.0007084446,0.00006332792,0.0009758712,0.00008312597,0.0004103742,0.00001404414],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005630148,"about_ca_system_score_gemma":0.01007026,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002637736,"about_ca_topic_score_gemma":0.0003998358,"domain_scores_codex":[0.9974984,0.0006865133,0.0004097086,0.0004781988,0.000614142,0.0003130496],"domain_scores_gemma":[0.998294,0.0006177668,0.0003010856,0.0001644099,0.0001842298,0.0004385073],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00007988549,0.001188299,0.1119549,0.0001012479,0.00007543468,0.000003121623,0.5113803,0.002549357,0.0003049073,0.3407721,0.0005888059,0.03100166],"study_design_scores_gemma":[0.0005804726,0.0001619512,0.591415,0.00008844674,0.00003765075,0.000006814143,0.0856593,0.00180558,0.00003868042,0.001714298,0.3180762,0.0004156],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6063907,0.00006513009,0.002680054,0.3405067,0.0004499228,0.001235749,0.000002596262,0.0001407563,0.04852832],"genre_scores_gemma":[0.919952,0.0001115551,0.0657199,0.01339284,0.0003353278,0.00009201931,0.00001748237,0.00001543263,0.0003634629],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.47946,"threshold_uncertainty_score":0.9955417,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2994689616","doi":"10.1080/0969594x.2019.1703171","title":"A cross-cultural comparison of German and Canadian student teachers’ assessment competence","year":2019,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University","funders":"","keywords":"Competence (human resources); German; Multitude; Psychology; Pedagogy; Cross-cultural; Mathematics education; Sociology; Social psychology; Political science; Geography","authors":[{"name":"Christopher DeLuca","is_ca":true},{"name":"Christoph Schneider","is_ca":false},{"name":"Andrew Coombs","is_ca":true},{"name":"Marcela Pozas","is_ca":false},{"name":"Amirhossein Rasooli","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06255569384159228,"gpt":0.5279710097896466,"spread":0.4654153159480544,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001920984,0.0001635128,0.0002742514,0.0002653939,0.0003588272,0.0004221565,0.0002749393,0.0001094108,0.0001577346],"category_scores_gemma":[0.0004094581,0.0001647644,0.00003155009,0.0003885636,0.0002754328,0.001365613,0.0001114859,0.0003409048,0.00000925729],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007293606,"about_ca_system_score_gemma":0.00376512,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.2090689,"about_ca_topic_score_gemma":0.07102387,"domain_scores_codex":[0.9977324,0.0005685796,0.0004559019,0.0003293174,0.0005295817,0.0003842019],"domain_scores_gemma":[0.9979041,0.000999268,0.0003750068,0.0002144562,0.0002491403,0.000258076],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000006710908,0.0003214155,0.7012047,0.00001749208,0.00002681962,4.358511e-7,0.05310056,0.00001185364,0.00008801893,0.242436,0.00005887673,0.002727103],"study_design_scores_gemma":[0.0003228829,0.00006349017,0.7536595,0.0000357557,0.00002293777,0.000002467163,0.1006086,0.00005342795,0.00000431427,0.0002227041,0.1448531,0.0001508767],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7798889,0.0001802447,0.000004209724,0.007867324,0.0003349031,0.0005260509,0.000004842876,0.00001347776,0.21118],"genre_scores_gemma":[0.9930194,0.0006239272,0.003191533,0.0006330427,0.0001656385,0.00004987539,0.00002029541,0.000009715106,0.002286543],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2422133,"threshold_uncertainty_score":0.9459275,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2068327470","doi":"10.1080/09695940802164218","title":"Self‐assessment in a technology‐supported environment: the case of grade 9 geography","year":2008,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Gender and Technology in Education","field":"Social Sciences","cited_by":25,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto; Institute for Christian Studies","funders":"University of Toronto","keywords":"Self-efficacy; Psychology; Variance (accounting); Self-assessment; Perception; Outcome (game theory); Treatment and control groups; Applied psychology; Medical education; Mathematics education; Social psychology; Medicine; Statistics; Mathematics; Accounting","authors":[{"name":"John A. Ross","is_ca":true},{"name":"Michelle Starling","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04048734503115643,"gpt":0.4035206507773588,"spread":0.3630333057462024,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001690795,0.0001266053,0.0001533718,0.0007608384,0.0004579923,0.00002526072,0.0002442335,0.0001877648,0.00003292887],"category_scores_gemma":[0.0005278469,0.0001167478,0.00003069733,0.001190158,0.0005340779,0.0004294138,0.00007870288,0.0004983845,0.000002710742],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003637651,"about_ca_system_score_gemma":0.00276041,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008589511,"about_ca_topic_score_gemma":0.0009472634,"domain_scores_codex":[0.9982151,0.0005992239,0.0004091295,0.0002756413,0.0002021645,0.0002987329],"domain_scores_gemma":[0.998323,0.0008546485,0.0003380241,0.0003573967,0.00006637022,0.00006061131],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.000006900367,0.002195822,0.2833214,0.00002336054,0.0000395945,0.00003629063,0.02928803,0.00001812752,0.00004386529,0.6766735,0.00006825667,0.008284866],"study_design_scores_gemma":[0.0005917955,0.0001073308,0.4948461,0.00002797386,0.00007574538,0.001101893,0.2751256,0.00009959443,0.00008241062,0.04452461,0.18313,0.0002869923],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8857482,0.0003765052,0.00006154575,0.07181407,0.0001745559,0.0006709535,0.000003652693,0.00005388661,0.04109663],"genre_scores_gemma":[0.9832239,0.005353674,0.01055216,0.0003993922,0.00007317126,0.0002664578,0.000007982574,0.000008227792,0.0001150741],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6321489,"threshold_uncertainty_score":0.9980124,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3167522218","doi":"10.1080/0969594x.2021.1932736","title":"Conceptualising a Fairness Framework for Assessment Adjusted Practices for Students with Disability: An Empirical Study","year":2021,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Disability Education and Employment","field":"Social Sciences","cited_by":21,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Leverage (statistics); Psychology; Diversity (politics); Best practice; Pedagogy; Medical education; Mathematics education; Applied psychology; Sociology; Medicine; Computer science","authors":[{"name":"Amirhossein Rasooli","is_ca":true},{"name":"Maryam Razmjoee","is_ca":false},{"name":"Joy Cumming","is_ca":false},{"name":"Elizabeth Dickson","is_ca":false},{"name":"Amanda Webster","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2480842159998611,"gpt":0.6006909723976173,"spread":0.3526067563977562,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.005174986,0.0002110255,0.0003015543,0.00009106433,0.0007721467,0.0008668912,0.0003041171,0.0001300052,0.00006873978],"category_scores_gemma":[0.01542703,0.00020569,0.0000477437,0.0006500012,0.000293165,0.001841943,0.00008882834,0.0002930177,9.786309e-7],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001047752,"about_ca_system_score_gemma":0.008320802,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006081443,"about_ca_topic_score_gemma":0.01220345,"domain_scores_codex":[0.99525,0.002324725,0.000548038,0.0007218317,0.0007252398,0.0004301019],"domain_scores_gemma":[0.9879017,0.009868606,0.0006799976,0.0004999327,0.0007602822,0.000289454],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002167401,0.01878925,0.5453664,0.00009205576,0.00009763263,7.00254e-7,0.09996181,0.00003171208,0.000003816052,0.3237733,0.00009103559,0.01157555],"study_design_scores_gemma":[0.0007656207,0.0004151819,0.252482,0.0000497883,0.0001340595,0.00000376388,0.5556222,0.00004216296,0.000003239226,0.005748086,0.1845176,0.0002162856],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8838512,0.0000578593,0.004800655,0.09936462,0.0006794684,0.003719489,0.00002324439,0.00006105169,0.007442465],"genre_scores_gemma":[0.9204622,0.00008302528,0.07268649,0.002991743,0.0006537442,0.002212476,0.00009580753,0.00002454926,0.0007899427],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4556604,"threshold_uncertainty_score":0.9973011,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3185953117","doi":"10.1080/0969594x.2021.1951161","title":"Assessing young children’s self-regulation in school contexts","year":2021,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":15,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia; The King's University; Western University","funders":"","keywords":"Self-regulated learning; Psychology; Curriculum; Developmental psychology; Situated; Self-control; Pedagogy; Computer science","authors":[{"name":"Lynda R. Hutchinson","is_ca":true},{"name":"Nancy E. Perry","is_ca":true},{"name":"Jennifer D. Shapka","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05781325932187747,"gpt":0.4946531875119819,"spread":0.4368399281901044,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004570578,0.000158416,0.0001927399,0.0004160654,0.0001594266,0.0002548417,0.0000947048,0.0001375072,0.0003067468],"category_scores_gemma":[0.004860867,0.0001775623,0.0000242411,0.0008000803,0.00003419783,0.00102426,0.00005676802,0.000850583,0.00002413578],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004623732,"about_ca_system_score_gemma":0.001845835,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002396049,"about_ca_topic_score_gemma":0.0001142093,"domain_scores_codex":[0.994146,0.004500447,0.0004704322,0.0004190808,0.0001806432,0.0002833666],"domain_scores_gemma":[0.9972718,0.001808543,0.0003442819,0.0003149683,0.0001877265,0.00007269216],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003208567,0.001322343,0.5745565,0.0000216046,0.00006084693,0.000007504694,0.01077649,0.00006389436,0.0005126239,0.3619324,0.0001950608,0.05051862],"study_design_scores_gemma":[0.0005607689,0.00002407815,0.9618236,0.00006742032,0.00002147845,0.0002143136,0.008784736,0.00006716204,0.00003230603,0.001586666,0.02664114,0.0001763345],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7834067,0.0003884069,0.002110639,0.009064293,0.0007205586,0.0002391745,0.000001411501,0.00005872912,0.2040101],"genre_scores_gemma":[0.9518818,0.00006486766,0.04343246,0.001859197,0.0003928318,0.00007953762,0.00005783742,0.00002043027,0.002211034],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3872671,"threshold_uncertainty_score":0.7240782,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2317555153","doi":"10.1080/0969594x.2013.844094","title":"Students’ interpersonal trust and attitudes towards standardised tests: exploring affective variables related to student assessment","year":2013,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Communication in Education and Healthcare","field":"Psychology","cited_by":14,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Psychology; Interpersonal communication; Social psychology; Test (biology); Expectancy theory; Scale (ratio); Value (mathematics); Structural equation modeling; Interpersonal relationship; Computer science","authors":[{"name":"Man-Wai Chu","is_ca":true},{"name":"Qi Guo","is_ca":true},{"name":"Jacqueline P. Leighton","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09861461347341459,"gpt":0.526713327331136,"spread":0.4280987138577214,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001913421,0.0002579703,0.0002926248,0.0004973558,0.0003187491,0.0004184005,0.0003773436,0.0001135555,0.0009013367],"category_scores_gemma":[0.000871012,0.0002647522,0.00003409221,0.0005572312,0.00009000979,0.0009684197,0.0003668007,0.0005957325,0.00004620467],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001068321,"about_ca_system_score_gemma":0.00155269,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005553649,"about_ca_topic_score_gemma":0.0002055444,"domain_scores_codex":[0.9968011,0.001170024,0.0006501616,0.0005532033,0.0004529637,0.0003724926],"domain_scores_gemma":[0.9964465,0.001910085,0.0003220003,0.0006074894,0.0003802775,0.0003336244],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006022166,0.002885992,0.5794849,0.00007721096,0.0001976268,0.000001203325,0.062799,0.00001576382,0.00006883909,0.287419,0.0006434015,0.06634681],"study_design_scores_gemma":[0.0006109622,0.0001838622,0.8421533,0.0001167705,0.00003345983,0.00002933788,0.1053462,0.0000148162,0.000002065541,0.0007581683,0.05052903,0.0002220828],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8730057,0.000455146,0.0001341072,0.04579162,0.001026264,0.001601427,0.00001532757,0.00006759977,0.07790283],"genre_scores_gemma":[0.9792126,0.001019522,0.01120224,0.003246763,0.0001605314,0.003141379,0.00003553402,0.00003071424,0.001950664],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2866608,"threshold_uncertainty_score":0.9999804,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4386496830","doi":"10.1080/0969594x.2023.2255936","title":"Classroom assessment fairness inventory: a new instrument to support perceived fairness in classroom assessment","year":2023,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University; University of Saskatchewan; University of Alberta","funders":"","keywords":"Psychology; Grading (engineering); Equity (law); Perception; Legitimacy; Diversity (politics); Pedagogy; Medical education; Sociology; Political science","authors":[{"name":"Amirhossein Rasooli","is_ca":true},{"name":"Christopher DeLuca","is_ca":true},{"name":"Liying Cheng","is_ca":true},{"name":"Amin Mousavi","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08886918950009094,"gpt":0.4589116453380946,"spread":0.3700424558380037,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.006183858,0.0005577894,0.000667404,0.001718865,0.0006884076,0.000856847,0.0008797536,0.0003417651,0.0004127885],"category_scores_gemma":[0.00143961,0.0006220013,0.0001269998,0.003685141,0.0001939393,0.002342218,0.0005998237,0.0009779381,0.0001439553],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.004085341,"about_ca_system_score_gemma":0.0167424,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02242138,"about_ca_topic_score_gemma":0.01372358,"domain_scores_codex":[0.992456,0.001943049,0.001255944,0.001168533,0.00179198,0.001384467],"domain_scores_gemma":[0.9958108,0.001792061,0.0006003101,0.0006774919,0.0002954091,0.0008239021],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00009890275,0.001865883,0.4971001,0.00009196495,0.00009979896,0.00002897738,0.03683988,0.0002617112,0.0002231536,0.4235274,0.006652401,0.03320973],"study_design_scores_gemma":[0.001172407,0.0002104463,0.5966941,0.0001259892,0.00004972657,0.000005983081,0.09425754,0.0002687263,0.00000322489,0.004076625,0.3026088,0.000526513],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6615182,0.00004463403,0.0005469086,0.1122525,0.002284787,0.002894102,0.00002011361,0.0002846934,0.220154],"genre_scores_gemma":[0.9700794,0.001717387,0.01048918,0.003337954,0.001047132,0.000908075,0.0001872464,0.00006838489,0.01216519],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4194508,"threshold_uncertainty_score":0.9997378,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4205974171","doi":"10.1080/0969594x.2021.1999209","title":"Investigating the potential of NLP-driven linguistic and acoustic features for predicting human scores of children’s oral language proficiency","year":2021,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Reading and Literacy Development","field":"Psychology","cited_by":10,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Storytelling; Vocabulary; Grammar; Literacy; Psychology; Computer science; Linguistics; Natural language processing; Artificial intelligence; Narrative; Pedagogy","authors":[{"name":"Melissa Hunte","is_ca":true},{"name":"Samantha McCormick","is_ca":true},{"name":"Maitree Shah","is_ca":true},{"name":"Clarissa Lau","is_ca":true},{"name":"Eunice Eunhee Jang","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03072020722652336,"gpt":0.4230499314041326,"spread":0.3923297241776092,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008070465,0.00009726204,0.0001484744,0.0001125002,0.0001501834,0.00005010625,0.00009298134,0.00005176008,0.00001008149],"category_scores_gemma":[0.003451327,0.00008042021,0.00002165492,0.0001855934,0.0001021039,0.0001086908,0.0000652289,0.0001812937,1.598152e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003659694,"about_ca_system_score_gemma":0.0006446757,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001498436,"about_ca_topic_score_gemma":0.00007433668,"domain_scores_codex":[0.9988071,0.0003189886,0.0003692979,0.0002178448,0.000131146,0.0001556334],"domain_scores_gemma":[0.9980178,0.001194055,0.0004003185,0.000181843,0.000163752,0.00004221179],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000686478,0.001346671,0.5339083,0.000599134,0.0002595182,0.000004339291,0.1460677,0.0004407376,0.01841116,0.2774741,0.0002204195,0.02119933],"study_design_scores_gemma":[0.0004071127,0.00009694669,0.9744487,0.000178791,0.0001000248,0.0001134267,0.02246737,0.0002154485,0.0004799497,0.0009406016,0.0004464948,0.0001051192],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9926202,0.0007271887,0.0001722779,0.001572783,0.000241093,0.0003647325,0.00001821025,0.000009497217,0.004273991],"genre_scores_gemma":[0.9892986,0.0000453277,0.009508324,0.0002790754,0.0002013419,0.000083352,0.00005214135,0.00001002797,0.0005217761],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4405404,"threshold_uncertainty_score":0.413181,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4385416815","doi":"10.1080/0969594x.2023.2242004","title":"Educational assessment in Ghana: The influence of historical colonization and political accountability","year":2023,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Formative assessment; Summative assessment; Accountability; Context (archaeology); Politics; Political science; Educational assessment; Government (linguistics); Pedagogy; Sociology; Public administration; Geography; Law","authors":[{"name":"David Baidoo-Anu","is_ca":true},{"name":"Christopher DeLuca","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05672128565655436,"gpt":0.4671383429696339,"spread":0.4104170573130795,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003456059,0.0001105881,0.0001641855,0.000296137,0.0002838572,0.0001100015,0.0001986016,0.00009351248,0.00002947463],"category_scores_gemma":[0.003672015,0.0001011798,0.000020173,0.001237927,0.0002482886,0.0008246314,0.0001223152,0.000292027,0.000003204019],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001536083,"about_ca_system_score_gemma":0.004561805,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02435267,"about_ca_topic_score_gemma":0.003076569,"domain_scores_codex":[0.9976172,0.0008570211,0.0004404743,0.0002689261,0.0004929104,0.0003234837],"domain_scores_gemma":[0.9945495,0.00471593,0.0002319547,0.000180712,0.0002172959,0.0001045891],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00001125533,0.0003609131,0.393255,0.00002123173,0.000006795304,2.003859e-7,0.007785795,0.00003315591,0.00001752953,0.5977758,0.0001739815,0.000558323],"study_design_scores_gemma":[0.0001880977,0.00002821537,0.9192597,0.00002634244,0.00001633942,0.000001993891,0.04312395,0.000101032,0.000001117659,0.008023916,0.02913241,0.00009692396],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8525159,0.00006399259,0.000007599115,0.1138609,0.0001760295,0.0004538138,0.000002222905,0.00001592035,0.0329036],"genre_scores_gemma":[0.9964124,0.0005548155,0.001014288,0.0008304165,0.0002399588,0.0001574253,0.00001618586,0.000007011174,0.0007675307],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5897519,"threshold_uncertainty_score":0.9821442,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4286809449","doi":"10.1080/0969594x.2022.2103516","title":"The Education and Assessment System in Lithuania","year":2022,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Declaration; Accountability; Political science; Curriculum; Bologna declaration; Politics; Test (biology); Public administration; Independence (probability theory); Function (biology); European union; Educational assessment; Pedagogy; Public relations; Higher education; Sociology; Business; Law","authors":[{"name":"Irena Raudienė","is_ca":false},{"name":"Lina Kaminskienè","is_ca":false},{"name":"Liying Cheng","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04449651455830835,"gpt":0.4724211113539256,"spread":0.4279245967956172,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.007828544,0.0001210661,0.000124711,0.0002529895,0.001819772,0.0004965154,0.0002632953,0.00005274185,0.00005269183],"category_scores_gemma":[0.002814983,0.0001202163,0.00001603814,0.0008852546,0.0001420225,0.001161,0.000156563,0.0007141781,0.00000471344],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001876361,"about_ca_system_score_gemma":0.01389293,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02601653,"about_ca_topic_score_gemma":0.002290141,"domain_scores_codex":[0.9943877,0.004056946,0.0004103463,0.0003295867,0.0005113467,0.0003040039],"domain_scores_gemma":[0.9937803,0.005188851,0.0005004602,0.0002664685,0.0001377192,0.0001262745],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001871619,0.0006709625,0.05433899,0.00002758279,0.00001014815,6.039801e-7,0.03772618,0.00005467175,0.000006295922,0.8665367,0.0004384636,0.04017073],"study_design_scores_gemma":[0.00009058981,0.00002371889,0.1482698,0.00001692246,0.000010516,0.00001898452,0.2399789,0.00005153133,2.396134e-7,0.0006869495,0.6107686,0.00008314686],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.3871272,0.001789898,0.00001789206,0.2102024,0.002158863,0.0009949125,0.000003750087,0.00005961684,0.3976455],"genre_scores_gemma":[0.9851447,0.001710998,0.003973509,0.001446372,0.0003865814,0.0009178057,0.00001293639,0.00001402496,0.006393106],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8658497,"threshold_uncertainty_score":0.9994797,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2287186680","doi":"10.1080/0969594x.2015.1113929","title":"Brazilian national assessment data and educational policy: an empirical illustration","year":2015,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"School Choice and Performance","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Portuguese; Sample (material); Census; Latin Americans; Student achievement; Mathematics education; Geography; Developing country; Contrast (vision); Academic achievement; Political science; Psychology; Economic growth; Demography; Sociology; Population; Computer science; Economics","authors":[{"name":"Christine L. Paget","is_ca":false},{"name":"Lars‐Erik Malmberg","is_ca":false},{"name":"Dale R. Martelli","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2385888002248343,"gpt":0.5617100844317808,"spread":0.3231212842069465,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004557081,0.0001342293,0.0001223217,0.0003265335,0.000439066,0.0004193269,0.0003323815,0.0001194083,0.00005018128],"category_scores_gemma":[0.005857388,0.0001435657,0.00001068723,0.000597571,0.0001839121,0.004614574,0.0001367251,0.0003125532,0.00001017263],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007738037,"about_ca_system_score_gemma":0.0290053,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01092122,"about_ca_topic_score_gemma":0.006376652,"domain_scores_codex":[0.9973978,0.0008473707,0.0003372022,0.0004159443,0.0007312114,0.0002704479],"domain_scores_gemma":[0.9974141,0.001214767,0.0002306345,0.0003237921,0.0004307617,0.0003859303],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004109521,0.001023851,0.1949499,0.00002375055,0.00002140125,4.306387e-7,0.01619668,0.00007000154,0.00001386225,0.7627886,0.007822538,0.01704791],"study_design_scores_gemma":[0.0002768876,0.00005493312,0.3337388,0.00001621627,0.00001374189,0.00001926576,0.01798035,0.001039883,7.972032e-7,0.01740662,0.6292892,0.0001632861],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.250799,0.0002386713,0.0001659469,0.4483012,0.0006105926,0.0005058895,0.00005721628,0.0000442548,0.2992772],"genre_scores_gemma":[0.9663535,0.001042658,0.01780588,0.007729989,0.004515756,0.00008242649,0.0004806926,0.00001270016,0.001976416],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.745382,"threshold_uncertainty_score":0.9956651,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3007758995","doi":"10.1080/0969594x.2020.1728908","title":"Making feedback effective?","year":2020,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Kellogg's (Canada)","funders":"","keywords":"Computer science; Psychology","authors":[{"name":"Therese N. Hopfenbeck","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09632089257636077,"gpt":0.4810667836253344,"spread":0.3847458910489737,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001381374,0.0001292173,0.0001525728,0.0001251575,0.0003563722,0.0002650739,0.0002109142,0.00008271297,0.0001290017],"category_scores_gemma":[0.003328038,0.0001401274,0.00003207194,0.0006660628,0.0001122709,0.001060785,0.0001063519,0.0002838946,0.00004239959],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000298837,"about_ca_system_score_gemma":0.00158052,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001576815,"about_ca_topic_score_gemma":0.0002500586,"domain_scores_codex":[0.9979612,0.0008207741,0.0002518118,0.0003011059,0.0003666554,0.0002984733],"domain_scores_gemma":[0.9975784,0.001828932,0.0002057533,0.0001205467,0.0001052788,0.0001611057],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00007043136,0.0003948126,0.07271353,0.00004265333,0.00003571033,0.000002696451,0.05582862,0.00001748969,0.0001087276,0.8270962,0.001051365,0.0426378],"study_design_scores_gemma":[0.0002957752,0.00005367382,0.1924029,0.00003111214,0.00002582179,0.000002492834,0.05439495,0.00006102894,0.000005570266,0.001217276,0.7513362,0.0001731925],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.09940084,0.0002971933,0.0002383613,0.184577,0.0004564799,0.0008904234,0.000002436014,0.00008276053,0.7140545],"genre_scores_gemma":[0.9820929,0.0008390172,0.005939774,0.009044613,0.001135689,0.0001099461,0.000006019551,0.00001268344,0.0008193363],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8826921,"threshold_uncertainty_score":0.5714229,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2013643140","doi":"10.1080/0969594x.2014.953910","title":"The effects of key demographic variables on markers’ perceived ease of use and acceptance of onscreen marking","year":2014,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Hearing Impairment and Communication","field":"Psychology","cited_by":4,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"AGE-WELL","keywords":"Rasch model; Psychology; Usability; Developmental psychology; Computer science","authors":[{"name":"Zi Yan","is_ca":false},{"name":"David Coniam","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03750763857916641,"gpt":0.3903221768588904,"spread":0.352814538279724,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001729043,0.00009062927,0.0001491984,0.0001720576,0.00008981428,0.00003021338,0.0001596449,0.00006218397,0.000008683852],"category_scores_gemma":[0.003631893,0.00007674762,0.00002335372,0.0002373392,0.0001282819,0.0002629668,0.00006605095,0.0001674958,3.799369e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000280134,"about_ca_system_score_gemma":0.0001900192,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002203626,"about_ca_topic_score_gemma":0.00006980289,"domain_scores_codex":[0.9978947,0.001359972,0.0003395519,0.0001454111,0.0001305566,0.0001298205],"domain_scores_gemma":[0.9875383,0.01140206,0.0004554301,0.0004648199,0.00009532398,0.00004399859],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.000555491,0.001302699,0.3673622,0.0002754367,0.0001318737,1.929794e-7,0.00716,0.000008839377,0.002570871,0.5606055,0.0001174198,0.0599095],"study_design_scores_gemma":[0.0003935837,0.0002613241,0.9839348,0.0001803127,0.00004661522,0.000004074625,0.002304177,0.00009354104,0.00006363342,0.001053839,0.01159811,0.00006597199],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9784796,0.0002333103,0.00008732251,0.002356775,0.000118986,0.0003117999,0.000001885385,0.00000597974,0.0184044],"genre_scores_gemma":[0.9951404,0.001479817,0.002756436,0.0002784964,0.00003107584,0.00005235493,0.00000461242,0.000008210149,0.0002486095],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6165726,"threshold_uncertainty_score":0.4347978,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4416894890","doi":"10.1080/0969594x.2025.2591284","title":"An implementation of argument-based validation for assessing college major preferences with a hybrid of Likert-rating and forced-choice formats","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Data collection; Measure (data warehouse); Key (lock); Quality (philosophy)","authors":[{"name":"Sirui Wu","is_ca":true},{"name":"Yue Mao","is_ca":true},{"name":"Jake E. Stone","is_ca":false},{"name":"Amery D. Wu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04844669177838798,"gpt":0.4835110550125784,"spread":0.4350643632341904,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001777034,0.0001160883,0.0001861462,0.0003036003,0.0003287303,0.0001693488,0.0001227748,0.00004658314,0.00001108987],"category_scores_gemma":[0.0005783815,0.0001124965,0.00001960712,0.0004211101,0.0001028775,0.001961696,0.0000275883,0.00008201207,5.816528e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001543954,"about_ca_system_score_gemma":0.00296777,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00528487,"about_ca_topic_score_gemma":0.001728005,"domain_scores_codex":[0.9984451,0.0004021802,0.0004538211,0.0002187012,0.0002889973,0.0001912341],"domain_scores_gemma":[0.9965391,0.002300221,0.0006561158,0.0001279561,0.0003228883,0.00005369718],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002057682,0.0009727982,0.4774331,0.0006083126,0.0001095444,1.704726e-7,0.01404523,0.0001036723,0.002193719,0.4673372,0.00008502579,0.03690543],"study_design_scores_gemma":[0.003326574,0.000594447,0.568276,0.0005674294,0.0003521846,0.0000019917,0.3990984,0.001538567,0.006703575,0.006076785,0.01307826,0.0003858013],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9807323,0.00006987273,0.003294966,0.004250352,0.0001107337,0.001192372,0.00002554476,0.00001498977,0.01030884],"genre_scores_gemma":[0.9706818,0.0001255195,0.02850212,0.0002435917,0.00006391154,0.0002132193,0.00006602218,0.000006566947,0.00009725805],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4612604,"threshold_uncertainty_score":0.7989177,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4411558571","doi":"10.1080/0969594x.2025.2523162","title":"Placing students at the centre: new directions for student agency in assessment","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Agency (philosophy); Mathematics education; Psychology; Medical education; Pedagogy; Sociology; Medicine; Social science","authors":[{"name":"Christopher DeLuca","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05630634260107546,"gpt":0.535095557930777,"spread":0.4787892153297015,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004643497,0.0001446907,0.0001568545,0.000293601,0.000964939,0.0004588373,0.0003791182,0.00008866454,0.000129062],"category_scores_gemma":[0.005289496,0.0001364621,0.00003704906,0.001044219,0.00009465982,0.0009724473,0.0001642703,0.0003855267,0.000009993326],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002028685,"about_ca_system_score_gemma":0.007747408,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01897003,"about_ca_topic_score_gemma":0.01774202,"domain_scores_codex":[0.9966091,0.001717437,0.0004633525,0.0003757659,0.0004770172,0.0003573529],"domain_scores_gemma":[0.9913493,0.007694724,0.000388597,0.0002893841,0.0001597826,0.0001182519],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004252935,0.001708574,0.4653895,0.00003393335,0.00006794422,4.96656e-7,0.08850389,0.0002771229,0.00002504782,0.418513,0.004662332,0.0207756],"study_design_scores_gemma":[0.0002615833,0.00001147259,0.3819346,0.00004169804,0.00003352607,8.47319e-7,0.03942433,0.00001180541,0.00000244172,0.0009107744,0.5772885,0.00007840461],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.3738214,0.0006599255,0.0003062318,0.4041684,0.002270215,0.001993585,0.000004147945,0.00005379489,0.2167223],"genre_scores_gemma":[0.8521536,0.002973292,0.007535723,0.003955225,0.000566855,0.0005784818,0.00002234944,0.00001699057,0.1321975],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5726262,"threshold_uncertainty_score":0.9978777,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4417447051","doi":"10.1080/0969594x.2025.2602452","title":"The validity of the assessment for learning measurement instrument for Ethiopian middle school mathematics teachers","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Measure (data warehouse); Reliability (semiconductor); Test validity; Test (biology)","authors":[{"name":"Lake Yeworiew","is_ca":false},{"name":"Kim Koh","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.596064082416965,"gpt":0.5664535415641934,"spread":0.02961054085277159,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03872744,0.0001402789,0.0002441197,0.0002868176,0.0007114109,0.0003301754,0.0006645796,0.00008103788,0.000006017118],"category_scores_gemma":[0.3479875,0.00008251938,0.0001037405,0.001147127,0.0001068319,0.0003339389,0.0002158804,0.0003802642,4.486896e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004825642,"about_ca_system_score_gemma":0.003323106,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001932812,"about_ca_topic_score_gemma":0.00007542888,"domain_scores_codex":[0.9957099,0.001794594,0.0009756769,0.0003464481,0.0008991339,0.0002742465],"domain_scores_gemma":[0.892214,0.1045765,0.001400005,0.0007401768,0.0009897684,0.00007948784],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004936677,0.0007329295,0.09627533,0.0001606188,0.0001032397,3.339102e-8,0.002108771,0.0005693522,0.0003483622,0.665193,0.001789519,0.2326695],"study_design_scores_gemma":[0.0006510887,0.0001417461,0.2626672,0.0001786879,0.0000782771,0.000003444517,0.04758152,0.003221957,0.0002317051,0.1607387,0.5243675,0.0001382016],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.288953,0.000792162,0.3400139,0.2666278,0.004841681,0.006748653,0.00002496612,0.0000589729,0.09193895],"genre_scores_gemma":[0.6983185,0.0003085405,0.2967338,0.001180318,0.0001380889,0.0006762997,0.00000231742,0.00001061968,0.002631494],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.522578,"threshold_uncertainty_score":0.9898324,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4415029048","doi":"10.1080/0969594x.2025.2573160","title":"The importance of community in assessment","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Education Systems and Policy","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Set (abstract data type); Work (physics)","authors":[{"name":"Christopher DeLuca","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08192554165819252,"gpt":0.5204617613953847,"spread":0.4385362197371921,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007167247,0.00007803553,0.0001318999,0.0002284129,0.000575572,0.0001110837,0.0002682903,0.00006698917,0.00001304296],"category_scores_gemma":[0.003999252,0.00006868862,0.00002061048,0.000886704,0.000193476,0.0004205179,0.00007083978,0.0003984635,0.000001004399],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005095224,"about_ca_system_score_gemma":0.007792133,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.05381303,"about_ca_topic_score_gemma":0.02433403,"domain_scores_codex":[0.9963879,0.002620307,0.0004818859,0.0001015021,0.0002027489,0.0002056858],"domain_scores_gemma":[0.9931365,0.005973216,0.0003460287,0.0003268409,0.0001655653,0.00005180798],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000005715016,0.0003186913,0.1943385,0.00002357274,0.000007398851,4.951383e-8,0.02033752,0.000006299594,0.00000747542,0.7787505,0.000446403,0.005757898],"study_design_scores_gemma":[0.00009479978,0.000008501678,0.4013827,0.0000380631,0.000004597761,5.879573e-7,0.1602556,0.000005213053,0.000002547483,0.01016916,0.427997,0.00004125483],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4476966,0.0003143824,0.00002080142,0.1136601,0.000381352,0.0003481757,0.000002107402,0.00000817893,0.4375683],"genre_scores_gemma":[0.9914171,0.001817054,0.0008361955,0.002615766,0.0001272938,0.0001424384,0.000004493274,0.000003837904,0.003035799],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7685813,"threshold_uncertainty_score":0.9978328,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4413304054","doi":"10.1080/0969594x.2025.2549158","title":"Diversity: a necessary imperative for assessment research","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Diversity (politics); Data science; Engineering ethics; Management science; Computer science; Sociology; Engineering; Anthropology","authors":[{"name":"Christopher DeLuca","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4676913733950703,"gpt":0.6664621529759441,"spread":0.1987707795808739,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.02525773,0.0001768253,0.0002804151,0.001556178,0.001825767,0.0006116584,0.000613855,0.0001173123,0.0001691306],"category_scores_gemma":[0.01567403,0.0001535441,0.00006103333,0.002260363,0.0001579875,0.002082949,0.00112477,0.0005452217,0.00001669689],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008331924,"about_ca_system_score_gemma":0.01020887,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009658303,"about_ca_topic_score_gemma":0.0003781455,"domain_scores_codex":[0.9943515,0.002190822,0.0008040874,0.0006702925,0.001559229,0.0004240656],"domain_scores_gemma":[0.9763798,0.02039728,0.0003561782,0.0006250004,0.00210383,0.0001378779],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001079399,0.0007498134,0.05762774,0.00003452412,0.00004095611,6.210196e-7,0.004673721,0.0001712208,0.00007685571,0.8599273,0.006510029,0.0700793],"study_design_scores_gemma":[0.0008322081,0.0001620205,0.2861777,0.00005923019,0.0000280088,0.000005791066,0.04507035,0.005298485,0.00004607498,0.1125585,0.5495876,0.0001740563],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2229315,0.0003580497,0.01053343,0.2522392,0.001347215,0.002424146,0.00002960952,0.00004255144,0.5100943],"genre_scores_gemma":[0.9600778,0.0006724112,0.01874498,0.006185903,0.0002196507,0.0005732451,0.00002753024,0.000008870305,0.01348966],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7473688,"threshold_uncertainty_score":0.9994737,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W7118245466","doi":"10.1080/0969594x.2025.2607856","title":"Assessment quality","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Quality (philosophy); Quality assessment; Educational assessment; Standards-based assessment; Risk assessment","authors":[{"name":"Christopher DeLuca","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09899369132097026,"gpt":0.5616235891664735,"spread":0.4626298978455032,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004424936,0.00015398,0.0003421264,0.0002687155,0.0001252656,0.0001187739,0.00009035831,0.0001260699,0.00008103303],"category_scores_gemma":[0.01137784,0.0001452935,0.00005440595,0.000624461,0.0000696046,0.0006222947,0.00008937658,0.0006079813,0.00000856379],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004046254,"about_ca_system_score_gemma":0.006729583,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00147244,"about_ca_topic_score_gemma":0.00011755,"domain_scores_codex":[0.9972739,0.001076244,0.0007672189,0.0003773188,0.0002763967,0.0002289566],"domain_scores_gemma":[0.9915717,0.00704345,0.0004022627,0.0004968881,0.0003228628,0.0001628472],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0002148063,0.002043153,0.08875023,0.0001796482,0.00008801065,0.000002710492,0.0001837042,0.00000476088,0.0002924133,0.8654642,0.0003656472,0.04241073],"study_design_scores_gemma":[0.0008734473,0.00007078849,0.5200362,0.00007683155,0.00009641235,0.00001087166,0.001944038,0.0001169969,0.000008087694,0.002349391,0.4743272,0.00008971213],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1923656,0.0005430828,0.0008745148,0.309757,0.0005623841,0.0008463145,0.00001092087,0.00005609962,0.4949841],"genre_scores_gemma":[0.9325048,0.001131605,0.0204473,0.0404893,0.0002842409,0.0001326576,0.00003477881,0.00001024576,0.004965098],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8631148,"threshold_uncertainty_score":0.9989014,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2072764341","doi":"10.1080/09695940600563611","title":"Differential effects of global modifications to large‐scale high stakes examination programmes","year":2006,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta; Queen's University","funders":"","keywords":"Scholarship; Accountability; Scale (ratio); Session (web analytics); Test (biology); Empirical examination; Identification (biology); Psychology; Political science; Medical education; Medicine; Geography; Law; Actuarial science; Business; Advertising","authors":[{"name":"Don A. Klinger","is_ca":true},{"name":"W. Todd Rogers","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02697556704776101,"gpt":0.4168892970341898,"spread":0.3899137299864288,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006905198,0.000121077,0.000154245,0.0001473022,0.0002752735,0.0001648278,0.0001768419,0.00008220212,0.00002681152],"category_scores_gemma":[0.0008336612,0.0001286358,0.00002934244,0.0007467506,0.0000848611,0.0006062538,0.00006017355,0.0001003176,0.000004992301],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004017673,"about_ca_system_score_gemma":0.001068906,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01515504,"about_ca_topic_score_gemma":0.005486237,"domain_scores_codex":[0.9981055,0.0005899787,0.0003274526,0.0002528485,0.0004319766,0.0002922973],"domain_scores_gemma":[0.9983203,0.0009493798,0.0002561419,0.0001730527,0.0002073121,0.00009383414],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00001767984,0.002034014,0.09850462,0.00004563762,0.00001593483,3.029056e-7,0.008124891,0.00001722241,0.0002389501,0.88168,0.0002097574,0.009110951],"study_design_scores_gemma":[0.0003659113,0.00006621101,0.9349887,0.000034015,0.00004658782,6.742089e-7,0.02214925,0.00002527831,0.00009678698,0.003450616,0.03863537,0.000140544],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9377885,0.00007474767,0.001558819,0.009446842,0.0004132239,0.0009260873,0.00001408504,0.00003851156,0.04973916],"genre_scores_gemma":[0.9910107,0.0001607793,0.006846962,0.0001874566,0.0004142008,0.0002371277,0.00004560089,0.000007180022,0.001090026],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8782294,"threshold_uncertainty_score":0.9914031,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}