{"alignment_dimensions_with_observations":[],"alignment_dimensions_without_observations":["authorized_goal_following","constraint_compliance","non_deception","oversight_and_correction","robustness"],"configuration":{"benchmark":"ASQA","context_tokens":4096,"demonstrations":2,"instructions":"full","method":"GTR + VANILLA","model_revision":"gpt-3.5-turbo-0301","passages":5,"scorer":"google/t5_xxl_true_nli_mixture; checkpoint revision not supplied","source_commit":"246c476a4edfc564266b7346b6e29ef4861ae937","source_runs":3,"source_table":18,"study_year":2023},"content_sha256":"25167fae860d7486743aaac68aa71372c91bba45c65fdf2e0f55c9b7856fe427","coverage_meaning":"At least one observation, not complete or validated dimension coverage.","evidence_status":"recorded","honesty_dimensions_with_observations":["citation_integrity"],"honesty_dimensions_without_observations":["belief_revision","evidence_support","factual_accuracy","pressure_resistance","uncertainty"],"id":"hm-8ff77c46e6fcc570ec23","integrity_meaning":"Content-addressed record only; not an authenticated issuer signature or independent scientific validation.","lab":"OpenAI","limitations":["Historical 2023 ASQA retrieval-augmented configurations. Scores include retrieval, prompting and NLI-scoring effects. Three-passages/light instructions and five-passages/full instructions are separate cohorts. "],"mark_status":"not_awarded","model_label":"ChatGPT","observations":[{"aggregation_blocker":"Historical 2023 ASQA retrieval-augmented configurations. Scores include retrieval, prompting and NLI-scoring effects. Three-passages/light instructions and five-passages/full instructions are separate cohorts. ","cohort":"alce:246c476a4edfc564266b7346b6e29ef4861ae937:asqa-vanilla-5psg-full","confidence_interval":null,"configuration":{"benchmark":"ASQA","context_tokens":4096,"demonstrations":2,"instructions":"full","method":"GTR + VANILLA","model_revision":"gpt-3.5-turbo-0301","passages":5,"scorer":"google/t5_xxl_true_nli_mixture; checkpoint revision not supplied","source_commit":"246c476a4edfc564266b7346b6e29ef4861ae937","source_runs":3,"source_table":18,"study_year":2023},"configuration_id":"alce:246c476a4edfc564266b7346b6e29ef4861ae937:asqa-vanilla-5psg-full:ChatGPT","definition":"Source average recall of reference short answers appearing in the generated answer after normalization. Coverage, not an independent world-truth accuracy test.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"paper_authors_reported","lab":"OpenAI","metric":"answer_em_recall","metric_label":"Answer EM recall","model_label":"ChatGPT","numerator":null,"overall_score":null,"pillar":"usefulness","profile_id":"hm-8ff77c46e6fcc570ec23","rank":1,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"alce","source_discrepancy":null,"source_runs":3,"source_sha256":"e10e1dacb25804558e6e801e331589ade47e0d9612f8e12ad1ba820f0b0021e8","source_standard_deviation":0.6,"source_table":18,"source_url":"https://github.com/princeton-nlp/ALCE/blob/246c476a4edfc564266b7346b6e29ef4861ae937/paper/ALCE.pdf#page=17","split":"ASQA source evaluation","study_year":2023,"track":"honesty_evidence","uncertainty_note":"Source SD across three seeded runs, in percentage points; not SE or a confidence interval. No binomial denominator reconstructed.","unit":"percent","value":40.4},{"aggregation_blocker":"Historical 2023 ASQA retrieval-augmented configurations. Scores include retrieval, prompting and NLI-scoring effects. Three-passages/light instructions and five-passages/full instructions are separate cohorts. ","cohort":"alce:246c476a4edfc564266b7346b6e29ef4861ae937:asqa-vanilla-5psg-full","confidence_interval":null,"configuration":{"benchmark":"ASQA","context_tokens":4096,"demonstrations":2,"instructions":"full","method":"GTR + VANILLA","model_revision":"gpt-3.5-turbo-0301","passages":5,"scorer":"google/t5_xxl_true_nli_mixture; checkpoint revision not supplied","source_commit":"246c476a4edfc564266b7346b6e29ef4861ae937","source_runs":3,"source_table":18,"study_year":2023},"configuration_id":"alce:246c476a4edfc564266b7346b6e29ef4861ae937:asqa-vanilla-5psg-full:ChatGPT","definition":"Source TRUE-NLI relevant-citation fraction, averaged within responses then across responses and three seeded runs; redundant supporting citations are allowed.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"paper_authors_reported","lab":"OpenAI","metric":"citation_precision","metric_label":"Citation precision","model_label":"ChatGPT","numerator":null,"overall_score":null,"pillar":"citation_integrity","profile_id":"hm-8ff77c46e6fcc570ec23","rank":1,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"alce","source_discrepancy":null,"source_runs":3,"source_sha256":"e10e1dacb25804558e6e801e331589ade47e0d9612f8e12ad1ba820f0b0021e8","source_standard_deviation":1.8,"source_table":18,"source_url":"https://github.com/princeton-nlp/ALCE/blob/246c476a4edfc564266b7346b6e29ef4861ae937/paper/ALCE.pdf#page=17","split":"ASQA source evaluation","study_year":2023,"track":"honesty_evidence","uncertainty_note":"Source SD across three seeded runs, in percentage points; not SE or a confidence interval. No binomial denominator reconstructed.","unit":"percent","value":72.5},{"aggregation_blocker":"Historical 2023 ASQA retrieval-augmented configurations. Scores include retrieval, prompting and NLI-scoring effects. Three-passages/light instructions and five-passages/full instructions are separate cohorts. ","cohort":"alce:246c476a4edfc564266b7346b6e29ef4861ae937:asqa-vanilla-5psg-full","confidence_interval":null,"configuration":{"benchmark":"ASQA","context_tokens":4096,"demonstrations":2,"instructions":"full","method":"GTR + VANILLA","model_revision":"gpt-3.5-turbo-0301","passages":5,"scorer":"google/t5_xxl_true_nli_mixture; checkpoint revision not supplied","source_commit":"246c476a4edfc564266b7346b6e29ef4861ae937","source_runs":3,"source_table":18,"study_year":2023},"configuration_id":"alce:246c476a4edfc564266b7346b6e29ef4861ae937:asqa-vanilla-5psg-full:ChatGPT","definition":"Source TRUE-NLI supported-statement fraction, averaged within responses then across responses and three seeded runs.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"paper_authors_reported","lab":"OpenAI","metric":"citation_recall","metric_label":"Citation recall","model_label":"ChatGPT","numerator":null,"overall_score":null,"pillar":"citation_integrity","profile_id":"hm-8ff77c46e6fcc570ec23","rank":1,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"alce","source_discrepancy":null,"source_runs":3,"source_sha256":"e10e1dacb25804558e6e801e331589ade47e0d9612f8e12ad1ba820f0b0021e8","source_standard_deviation":1.1,"source_table":18,"source_url":"https://github.com/princeton-nlp/ALCE/blob/246c476a4edfc564266b7346b6e29ef4861ae937/paper/ALCE.pdf#page=17","split":"ASQA source evaluation","study_year":2023,"track":"honesty_evidence","uncertainty_note":"Source SD across three seeded runs, in percentage points; not SE or a confidence interval. No binomial denominator reconstructed.","unit":"percent","value":73.6}],"overall_alignment_score":null,"overall_honesty_score":null,"priority_gaps":["belief_revision","evidence_support","factual_accuracy","pressure_resistance","uncertainty"],"requirements":[{"detail":"The standard is a draft; no pass thresholds or award authority have been validated.","id":"standard_ratified","label":"A published, validated standard and scoring policy","status":"not_met"},{"detail":"The current source packet does not establish this requirement.","id":"configuration_pinned","label":"Model revision, provider, scaffold, tools and test settings are pinned","status":"not_established"},{"detail":"Missing or incomplete common-basket coverage. Some observations cover: citation_integrity.","id":"complete_honesty_basket","label":"All required honesty dimensions measured on the same declared configuration","status":"not_met"},{"detail":"The current source packet does not establish this requirement.","id":"held_out_evaluation","label":"Held-out evaluation separated from training and development","status":"not_established"},{"detail":"The current source packet does not establish this requirement.","id":"usefulness_and_evasion","label":"Useful answering, abstention and evasion measured together","status":"not_established"},{"detail":"The current source packet does not establish this requirement.","id":"uncertainty_validated","label":"Uncertainty reflects task dependence and supports the claimed comparison","status":"not_established"},{"detail":"The current source packet does not establish this requirement.","id":"independent_review","label":"Independent review reconstructs the evidence and scoring","status":"not_established"},{"detail":"Evidence digests exist. Certification issuance, issuer authentication and renewal are not implemented.","id":"issuer_and_lifecycle","label":"Issuer identity, expiry, re-evaluation and withdrawal process are established","status":"not_met"}],"schema":"honesty-mark-profile/v0.1","scope":"Native source/configuration results only. This profile is not an awarded Honesty Mark.","source":"alce","sources":[{"comparison_caution":"Selected ASQA VANILLA rows from Table 18; TRUE-NLI citation scores and answer EM recall, mean and SD across three runs. Retrieval and instructions define separate cohorts. Oasst-33B citation precision differs between Tables 7 and 18; the Table 18 value is retained with the discrepancy disclosed.","id":"alce","import_status":"imported","metrics":["Answer EM recall","Citation recall","Citation precision"],"name":"ALCE · historical 2023","organization":"ALCE authors","pillars":["citation_integrity"],"reviewed_on":"2026-10-11","url":"https://arxiv.org/abs/2305.14627"}],"standard_sha256":"0cec00d16bbb6e2a02a28e43e43394e3eb309523c2ed697913154c4a2e6e6f6c","standard_version":"honesty-mark-standard/v0.1"}