{"alignment_dimensions_with_observations":["constraint_compliance","robustness"],"alignment_dimensions_without_observations":["authorized_goal_following","non_deception","oversight_and_correction"],"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"content_sha256":"108501306eaf80ff5d9dd6661a05af122ee6f474e46324880958640f04304873","coverage_meaning":"At least one observation, not complete or validated dimension coverage.","evidence_status":"recorded","honesty_dimensions_with_observations":["uncertainty"],"honesty_dimensions_without_observations":["belief_revision","citation_integrity","evidence_support","factual_accuracy","pressure_resistance"],"id":"hm-22de079b73829026ac9e","integrity_meaning":"Content-addressed record only; not an authenticated issuer signature or independent scientific validation.","lab":"Anthropic","limitations":["Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent."],"mark_status":"not_awarded","model_label":"Claude Sonnet 4","observations":[{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Stability of confidence across runs, not correctness of confidence.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"C_conf","metric_label":"Confidence consistency","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"reliability","profile_id":"hm-22de079b73829026ac9e","rank":2,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0118762269209962,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.905945980649422},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Repeat agreement on success/failure; always failing can score 1.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"C_out","metric_label":"Outcome consistency","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"reliability","profile_id":"hm-22de079b73829026ac9e","rank":7,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0306945049056481,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.762962962962963},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Stability of resource consumption across repeated runs.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"C_res","metric_label":"Resource consistency","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"reliability","profile_id":"hm-22de079b73829026ac9e","rank":3,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0097495755631655,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.7592466085847418},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Similarity of action distributions across successful runs.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"C_traj_d","metric_label":"Action-distribution consistency","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"reliability","profile_id":"hm-22de079b73829026ac9e","rank":4,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0117015865537766,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.8717661760468421},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Similarity of action order across successful runs.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"C_traj_s","metric_label":"Action-sequence consistency","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"reliability","profile_id":"hm-22de079b73829026ac9e","rank":5,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.017351625493947,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.7485625460151788},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"AUROC of confidence distinguishing success from failure.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"P_auroc","metric_label":"Confidence discrimination","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"calibration","profile_id":"hm-22de079b73829026ac9e","rank":12,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0214159233483997,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.7122183052187665},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Inverted binary Brier score; higher is better. Not the raw Brier loss.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"P_brier","metric_label":"Brier quality: 1 − Brier","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"calibration","profile_id":"hm-22de079b73829026ac9e","rank":6,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.014068254870905,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.7337070707070708},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"One minus expected calibration error using source bins.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"P_cal","metric_label":"Calibration: 1 − ECE","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"calibration","profile_id":"hm-22de079b73829026ac9e","rank":6,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0210007656756455,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.774949494949495},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Source normalized excess area under the risk–coverage curve.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"P_rc","metric_label":"Risk–coverage quality","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"calibration","profile_id":"hm-22de079b73829026ac9e","rank":9,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":null,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.4196078855467356},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Perturbed success / baseline success, capped at 1. Read beside task success.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"R_fault","metric_label":"Tool-failure robustness","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"robustness","profile_id":"hm-22de079b73829026ac9e","rank":1,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0297510599037571,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":1.0},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Rephrased-prompt success / baseline success, capped at 1.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"R_prompt","metric_label":"Prompt robustness","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"robustness","profile_id":"hm-22de079b73829026ac9e","rank":7,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0483161754087452,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.8302583025830258},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Structural-perturbation success / baseline success, capped at 1.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"R_struct","metric_label":"Format robustness","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"robustness","profile_id":"hm-22de079b73829026ac9e","rank":1,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":0.0459717647973824,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":1.0},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Fraction of source tasks with zero detected constraint violations.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"S_comp","metric_label":"Constraint compliance","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"constraint_compliance","profile_id":"hm-22de079b73829026ac9e","rank":1,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":null,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":1.0},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"One minus severity conditional on a violation; equals 1 if no violations. Not the probability of avoiding harm.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"S_harm","metric_label":"Conditional severity: inverted","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"constraint_compliance","profile_id":"hm-22de079b73829026ac9e","rank":1,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":null,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":1.0},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Successful tasks under the source agent setup.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"accuracy","metric_label":"Task success","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"usefulness","profile_id":"hm-22de079b73829026ac9e","rank":6,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":2.23717962935121,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"percent","value":54.74747474747475},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Source category-weighted consistency aggregate. Consistent failure can score well.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"dim_Consistency","metric_label":"Consistency","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"reliability","profile_id":"hm-22de079b73829026ac9e","rank":4,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":null,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.7774579775262384},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"HAL uses the inverted Brier score as this aggregate.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"dim_Predictability","metric_label":"Predictability","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"calibration","profile_id":"hm-22de079b73829026ac9e","rank":6,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":null,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.7337070707070708},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Source mean of fault, format and prompt robustness ratios.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"dim_Robustness","metric_label":"Robustness","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"robustness","profile_id":"hm-22de079b73829026ac9e","rank":6,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":null,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.9434194341943419},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"One minus source violation frequency times conditional severity. Not universal safety.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"dim_Safety","metric_label":"Constraint-risk score","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"constraint_compliance","profile_id":"hm-22de079b73829026ac9e","rank":1,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":null,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":1.0},{"aggregation_blocker":"Agent + scaffold result on the named benchmark. Not a bare-model alignment score. Repeats and overlapping aggregate metrics are dependent.","cohort":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia","confidence_interval":null,"configuration":{"agent_id":"gaia_generalist_claude_sonnet_4","benchmark":"gaia","model_revision":null,"provider_name":"anthropic","reasoning_effort":null,"scaffold":"HAL benchmark-specific agent; source identity retained, not equated with a bare model.","source_agent_slug":"claude-sonnet-4"},"configuration_id":"hal:a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20:gaia:gaia_generalist_claude_sonnet_4","definition":"Source aggregate of consistency, predictability and robustness. Excludes safety.","denominator":null,"direction":"higher","evaluation_date":null,"evidence_class":"independent_benchmark_reported","lab":"Anthropic","metric":"overall_reliability","metric_label":"HAL reliability","model_label":"Claude Sonnet 4","model_release_date":"2025-05-22","numerator":null,"overall_score":null,"pillar":"reliability","profile_id":"hm-22de079b73829026ac9e","rank":6,"rank_meaning":"Descriptive native-metric order; not a significance or overall honesty rank.","source":"hal","source_runs":3,"source_sha256":"a0da5842bdd0ca4563dd0d9a67efd62e47e24f9ee7fd42090f196e8bd5bc3a20","source_standard_error":null,"source_task_count":165,"source_updated":null,"source_url":"https://hal.cs.princeton.edu/reliability/benchmark/gaia/","split":"source_benchmark_tasks","track":"agent_alignment_evidence","uncertainty_note":"Source standard error retained where supplied; not converted to a confidence interval or a significance rank.","unit":"fraction","value":0.8181948274758836}],"overall_alignment_score":null,"overall_honesty_score":null,"priority_gaps":["belief_revision","citation_integrity","evidence_support","factual_accuracy","pressure_resistance"],"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: uncertainty.","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 source reports standard errors for some metrics. No cross-benchmark or award uncertainty is established.","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":"hal","sources":[{"comparison_caution":"Agent/scaffold results on GAIA and clean airline τ-bench. HAL reliability excludes safety; metric aggregates overlap their components. Source standard errors are not confidence intervals. Model release dates are not evaluation dates.","id":"hal","import_status":"imported","metrics":["Task success","HAL reliability","Consistency","Outcome consistency","Action-distribution consistency","Action-sequence consistency","Resource consistency","Confidence consistency","Predictability","Calibration: 1 − ECE","Confidence discrimination","Brier quality: 1 − Brier","Risk–coverage quality","Robustness","Tool-failure robustness","Format robustness","Prompt robustness","Constraint-risk score","Constraint compliance","Conditional severity: inverted"],"name":"HAL Agent Reliability","organization":"SAgE / Princeton University","pillars":["calibration","constraint_compliance","robustness","reliability"],"reviewed_on":"2026-10-11","url":"https://hal.cs.princeton.edu/reliability/"}],"standard_sha256":"0cec00d16bbb6e2a02a28e43e43394e3eb309523c2ed697913154c4a2e6e6f6c","standard_version":"honesty-mark-standard/v0.1"}