{"id":"W3180000962","doi":"10.1108/jfp-02-2021-0004","title":"“That’s the way my Wednesdays always go”: reverse-order instructions insufficient to mitigate schema-consistent errors in alibi generation","year":2021,"lang":"en","type":"article","venue":"Journal of Forensic Practice","topic":"Deception detection and forensic psychology","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Bishop's University; Ontario Tech University","funders":"","keywords":"Alibi; Schema (genetic algorithms); Recall; Psychology; Mnemonic; Originality; Social psychology; Computer science; Cognitive psychology; Information retrieval","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001547141,0.0002287174,0.0003754757,0.0004088349,0.000221155,0.00009074537,0.0002387571,0.0002082151,0.001274571],"category_scores_gemma":[0.001759889,0.0001825155,0.0002326619,0.001165582,0.0002060328,0.000484968,0.0000701255,0.0009359239,0.0004242857],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002104261,"about_ca_system_score_gemma":0.0001812684,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006811255,"about_ca_topic_score_gemma":0.0005124134,"domain_scores_codex":[0.9969747,0.0007498917,0.000971298,0.0003909481,0.0005195464,0.0003935859],"domain_scores_gemma":[0.9970157,0.0003350203,0.0008821686,0.0006098805,0.0009333466,0.0002239257],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001321753,0.002040461,0.001001686,0.00002024573,0.0009750904,0.001735798,0.0197888,0.002335698,0.01576681,0.02425351,0.7999601,0.1308],"study_design_scores_gemma":[0.002741874,0.0004754931,0.0114295,0.00006213507,0.0002408656,0.01547865,0.02656911,0.0001916049,0.002473002,0.0004122665,0.9395957,0.0003298034],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8294793,0.0008058239,0.006652632,0.0746945,0.03382374,0.0004755023,0.00001680805,0.00005123917,0.05400043],"genre_scores_gemma":[0.9556996,0.0001986185,0.01394455,0.02034742,0.001168559,0.00003952649,0.00001212684,0.00005672532,0.008532931],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1396355,"threshold_uncertainty_score":0.9996384,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0572796402303015,"score_gpt":0.3396740422042591,"score_spread":0.2823944019739575,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}