id stringlengths 21 21 | kind stringclasses 3
values | options listlengths 2 6 | target listlengths 2 6 | state stringlengths 172 270 | question stringclasses 7
values | domain stringclasses 1
value | family stringclasses 1
value | source stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|
v3_3efd65b51fd6b9f2_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0,
0.02,
0.31,
0.63,
0.03
] | An agent plan fetches customer data before verifying the account ID exists; additionally uses batch endpoints where streaming is required. Context: with a hard spend cap, during a scheduled maintenance window. Reported by the review queue. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_719882f82c0d1749_n | noul | [
"false",
"true"
] | [
0.31,
0.69
] | A task decomposition fetches customer data before verifying the account ID exists; additionally sequences approvals before the technical check. Context: for a compliance-critical workflow, during an external partner integration. Reported by the on-call engineer. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_5ffd73d6f3643cd7_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0,
0.02,
0.16,
0.69,
0.12
] | A task decomposition skips the sandbox validation step; additionally parallelizes steps that share a lock. Context: under a read-only budget, immediately after go-live. Reported by the operator. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_6002eeb5233dee24_n | noul | [
"false",
"true"
] | [
0.29,
0.71
] | An automation plan ends with a human confirmation gate; additionally fetches customer data before verifying the account ID exists. Context: under a read-only budget, during normal operations. Reported by the supervisor. | Is this scenario one where: the plan can complete within budget? | plan_quality | agent | yuri_v3 |
v3_ba1e862ebb5fa056_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0.06,
0.24,
0.53,
0.15,
0.01
] | An agent plan requests write access it never uses; additionally ends with a human confirmation gate. Context: with a hard spend cap, as part of a compliance sweep. Reported by the on-call engineer. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_de606aa6ae67d732_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0.53,
0.29,
0.05,
0.13
] | An agent plan includes a rollback step for each mutation; additionally parallelizes steps that share a lock. Context: with a 40-minute deadline, during a scheduled maintenance window. Reported by the operator. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_5301f774082709b9_n | noul | [
"false",
"true"
] | [
0.94,
0.06
] | A task decomposition calls the refund API before checking dispute status; additionally parallelizes steps that share a lock. Context: with a 40-minute deadline, during the holiday freeze. Reported by the supervisor. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_2d117fb9c4159689_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.02,
0.01,
0.05,
0.41,
0.49,
0.02
] | An automation plan assumes an API quota that expired last month; additionally requests write access it never uses. Context: under a read-only budget, under an active deadline. Reported by the operator. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_7fc76ef3c9e8924a_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0,
0.3,
0.17,
0.53
] | An agent plan sequences approvals before the technical check; additionally assumes an API quota that expired last month. Context: for a compliance-critical workflow, while legacy systems are still active. Reported by the monitoring system. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_2a8ca0ee276120ae_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0,
0.01,
0.05,
0.52,
0.41,
0.01
] | An agent plan sequences approvals before the technical check; additionally requests write access it never uses. Context: under a read-only budget, as part of a compliance sweep. Reported by the supervisor. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_44c2397332e5ce85_n | noul | [
"false",
"true"
] | [
0.09,
0.91
] | A task decomposition ends with a human confirmation gate; additionally requests write access it never uses. Context: for a compliance-critical workflow, as part of a compliance sweep. Reported by the supervisor. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_e8264cc8dcae09ba_n | noul | [
"false",
"true"
] | [
0.93,
0.07
] | A task decomposition calls the refund API before checking dispute status; additionally includes a rollback step for each mutation. Context: with a hard spend cap, with staffing at reduced levels. Reported by the review queue. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_54c97a589689f9f1_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.13,
0.29,
0.32,
0.26
] | An automation plan uses batch endpoints where streaming is required; additionally ends with a human confirmation gate. Context: under a read-only budget, mid-incident. Reported by the automated check. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_facd826ee4d3b772_n | noul | [
"false",
"true"
] | [
0.8,
0.2
] | A proposed workflow handles the failure mode explicitly; additionally includes a rollback step for each mutation. Context: with a hard spend cap, during a scheduled maintenance window. Reported by the operator. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_a3d271b18350ba6b_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.08,
0.23,
0.24,
0.36,
0.07,
0.02
] | A proposed workflow ends with a human confirmation gate; additionally parallelizes steps that share a lock. Context: under a read-only budget, at the end of the fiscal period. Reported by the operator. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_e3979472bd68dbc6_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.02,
0.02,
0.06,
0.25,
0.59,
0.06
] | A task decomposition includes a rollback step for each mutation; additionally skips the sandbox validation step. Context: with a 40-minute deadline, during the holiday freeze. Reported by the monitoring system. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_234a075eae0575ac_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0.04,
0.23,
0.45,
0.25,
0.02
] | An automation plan requests write access it never uses; additionally handles the failure mode explicitly. Context: with a hard spend cap, immediately after go-live. Reported by the monitoring system. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_43b93adcf8784411_n | noul | [
"false",
"true"
] | [
0.71,
0.29
] | A proposed workflow requests write access it never uses; additionally sequences approvals before the technical check. Context: under a read-only budget, during the holiday freeze. Reported by the operator. | Is this scenario one where: the plan can complete within budget? | plan_quality | agent | yuri_v3 |
v3_5c17837781fd0663_n | noul | [
"false",
"true"
] | [
0.94,
0.06
] | An automation plan calls the refund API before checking dispute status; additionally ends with a human confirmation gate. Context: for a compliance-critical workflow, mid-incident. Reported by the supervisor. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_f0a0489ce3686a4f_n | noul | [
"false",
"true"
] | [
0.8,
0.2
] | A task decomposition assumes an API quota that expired last month; additionally requests write access it never uses. Context: with a 40-minute deadline, during the holiday freeze. Reported by the automated check. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_977f892687670541_n | noul | [
"false",
"true"
] | [
0.49,
0.51
] | A task decomposition estimates 2 hours for a task with a known 2-day dependency; additionally parallelizes steps that share a lock. Context: with a hard spend cap, during the morning shift. Reported by the monitoring system. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_ccfaf7f9212bdf92_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.03,
0.75,
0.08,
0.14
] | A task decomposition estimates 2 hours for a task with a known 2-day dependency; additionally calls the refund API before checking dispute status. Context: under a read-only budget, while on-call rotation changed. Reported by the on-call engineer. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_ab4e4a897a07f62e_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0.14,
0.24,
0.1,
0.52
] | An agent plan ends with a human confirmation gate; additionally parallelizes steps that share a lock. Context: with a hard spend cap, immediately after go-live. Reported by the automated check. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_a57547febd908917_n | noul | [
"false",
"true"
] | [
0.7,
0.3
] | An automation plan ends with a human confirmation gate; additionally estimates 2 hours for a task with a known 2-day dependency. Context: under a read-only budget, following a recent audit. Reported by the review queue. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_72c1790e619df6be_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.01,
0.08,
0.91,
0
] | A proposed workflow estimates 2 hours for a task with a known 2-day dependency; additionally handles the failure mode explicitly. Context: with a hard spend cap, shortly after a software update. Reported by the on-call engineer. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_3c4444912d2b283a_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0,
0,
0.02,
0.43,
0.54,
0.01
] | An agent plan requests write access it never uses; additionally calls the refund API before checking dispute status. Context: under a read-only budget, at the end of the fiscal period. Reported by the monitoring system. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_f53bf6e414d7c816_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0.03,
0.18,
0.58,
0.19,
0.01
] | An agent plan fetches customer data before verifying the account ID exists; additionally ends with a human confirmation gate. Context: with a 40-minute deadline, with a new vendor onboarded. Reported by the supervisor. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_8e8cb07fbef09878_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.02,
0.02,
0.08,
0.6,
0.27,
0.01
] | An agent plan assumes an API quota that expired last month; additionally requests write access it never uses. Context: under a read-only budget, during a scheduled maintenance window. Reported by the supervisor. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_0962459101d9e7c1_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0,
0.52,
0.47,
0.01
] | A proposed workflow uses batch endpoints where streaming is required; additionally fetches customer data before verifying the account ID exists. Context: under a read-only budget, following a recent audit. Reported by the operator. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_54d5fa1b307e9820_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.13,
0.26,
0.18,
0.33,
0.08,
0.02
] | A task decomposition parallelizes steps that share a lock; additionally uses batch endpoints where streaming is required. Context: under a read-only budget, during normal operations. Reported by the on-call engineer. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_ac9e89a5474bcd6b_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0.03,
0.56,
0.1,
0.31
] | A proposed workflow sequences approvals before the technical check; additionally parallelizes steps that share a lock. Context: under a read-only budget, while legacy systems are still active. Reported by the supervisor. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_484c6c5fd7f02d10_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.03,
0.02,
0.03,
0.28,
0.6,
0.04
] | A task decomposition requests write access it never uses; additionally sequences approvals before the technical check. Context: with a 40-minute deadline, at the end of the fiscal period. Reported by the on-call engineer. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_1f15d697611568c3_n | noul | [
"false",
"true"
] | [
0.51,
0.49
] | An agent plan includes a rollback step for each mutation; additionally sequences approvals before the technical check. Context: with a 40-minute deadline, during a scheduled maintenance window. Reported by the automated check. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_c2223a944aeb00b0_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0,
0,
0.1,
0.79,
0.1
] | An automation plan requests write access it never uses; additionally assumes an API quota that expired last month. Context: for a compliance-critical workflow, under an active deadline. Reported by the automated check. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_2dc2e07037e910c8_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.04,
0.01,
0.03,
0.27,
0.6,
0.05
] | A task decomposition assumes an API quota that expired last month; additionally parallelizes steps that share a lock. Context: with a 40-minute deadline, during normal operations. Reported by the supervisor. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_4fbd5ea0628b7444_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.01,
0.77,
0.14,
0.08
] | An agent plan assumes an API quota that expired last month; additionally uses batch endpoints where streaming is required. Context: with a 40-minute deadline, under an active deadline. Reported by the operator. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_72580fd4b4543676_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.03,
0.03,
0.31,
0.43,
0.19,
0.01
] | A proposed workflow assumes an API quota that expired last month; additionally ends with a human confirmation gate. Context: for a compliance-critical workflow, during normal operations. Reported by the supervisor. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_7ba37c74bed37a0d_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.01,
0.75,
0.18,
0.06
] | A task decomposition sequences approvals before the technical check; additionally assumes an API quota that expired last month. Context: for a compliance-critical workflow, during an external partner integration. Reported by the review queue. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_c43b254773c18918_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.03,
0.01,
0.02,
0.15,
0.7,
0.09
] | A task decomposition requests write access it never uses; additionally calls the refund API before checking dispute status. Context: with a 40-minute deadline, during peak load. Reported by the monitoring system. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_14a01012c8c330e3_n | noul | [
"false",
"true"
] | [
0.61,
0.39
] | A task decomposition skips the sandbox validation step; additionally assumes an API quota that expired last month. Context: under a read-only budget, during the holiday freeze. Reported by the automated check. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_d9333317779ec4a8_n | noul | [
"false",
"true"
] | [
0.83,
0.17
] | An agent plan requests write access it never uses; additionally skips the sandbox validation step. Context: under a read-only budget, under reduced staffing. Reported by the on-call engineer. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_d9b1e527b5b4d995_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.06,
0.25,
0.17,
0.36,
0.13,
0.03
] | An agent plan includes a rollback step for each mutation; additionally uses batch endpoints where streaming is required. Context: under a read-only budget, during a scheduled maintenance window. Reported by the review queue. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_da7ec481091c60ab_n | noul | [
"false",
"true"
] | [
0.67,
0.33
] | A task decomposition handles the failure mode explicitly; additionally requests write access it never uses. Context: with a 40-minute deadline, during the quarterly review. Reported by the on-call engineer. | Is this scenario one where: the plan can complete within budget? | plan_quality | agent | yuri_v3 |
v3_b558e7427c84d58a_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.03,
0.01,
0.01,
0.1,
0.7,
0.15
] | A proposed workflow assumes an API quota that expired last month; additionally skips the sandbox validation step. Context: with a hard spend cap, immediately after go-live. Reported by the on-call engineer. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_919ed2d2f3055219_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0.01,
0.04,
0.5,
0.43,
0.01
] | A task decomposition requests write access it never uses; additionally estimates 2 hours for a task with a known 2-day dependency. Context: with a hard spend cap, during the quarterly review. Reported by the automated check. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_7ce4856ecb6dd680_n | noul | [
"false",
"true"
] | [
0.96,
0.04
] | A proposed workflow calls the refund API before checking dispute status; additionally uses batch endpoints where streaming is required. Context: with a hard spend cap, immediately after go-live. Reported by the automated check. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_62256a278a065735_n | noul | [
"false",
"true"
] | [
0.75,
0.25
] | An agent plan assumes an API quota that expired last month; additionally skips the sandbox validation step. Context: with a hard spend cap, during a scheduled maintenance window. Reported by the review queue. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_8bed8d226119c25b_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0,
0.76,
0.19,
0.05
] | A task decomposition assumes an API quota that expired last month; additionally handles the failure mode explicitly. Context: with a 40-minute deadline, during peak load. Reported by the supervisor. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_6de91eb756d4eaee_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.01,
0.53,
0.37,
0.09
] | An agent plan sequences approvals before the technical check; additionally calls the refund API before checking dispute status. Context: for a compliance-critical workflow, under reduced staffing. Reported by the operator. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_e5286466a7dbbbdf_n | noul | [
"false",
"true"
] | [
0.68,
0.32
] | A proposed workflow fetches customer data before verifying the account ID exists; additionally sequences approvals before the technical check. Context: with a hard spend cap, under reduced staffing. Reported by the operator. | Is this scenario one where: the plan can complete within budget? | plan_quality | agent | yuri_v3 |
v3_34a6a004792cf819_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0.01,
0.04,
0.36,
0.55,
0.03
] | An automation plan calls the refund API before checking dispute status; additionally sequences approvals before the technical check. Context: with a hard spend cap, as part of a compliance sweep. Reported by the operator. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_0de791aa3609e92e_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.02,
0.01,
0.06,
0.31,
0.56,
0.04
] | A task decomposition ends with a human confirmation gate; additionally calls the refund API before checking dispute status. Context: under a read-only budget, under reduced staffing. Reported by the supervisor. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_e061b301f3f14733_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.02,
0.02,
0.13,
0.41,
0.4,
0.02
] | An agent plan calls the refund API before checking dispute status; additionally includes a rollback step for each mutation. Context: with a 40-minute deadline, during the system migration. Reported by the automated check. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_4b421120af978f63_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.49,
0.05,
0.41,
0.05
] | A proposed workflow fetches customer data before verifying the account ID exists; additionally estimates 2 hours for a task with a known 2-day dependency. Context: under a read-only budget, under reduced staffing. Reported by the review queue. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_45d65f9f52d26978_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0,
0.01,
0.97,
0.02
] | A proposed workflow sequences approvals before the technical check; additionally estimates 2 hours for a task with a known 2-day dependency. Context: with a hard spend cap, during normal operations. Reported by the monitoring system. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_eb33ee004eaddc34_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0,
0.67,
0.21,
0.12
] | An automation plan assumes an API quota that expired last month; additionally includes a rollback step for each mutation. Context: for a compliance-critical workflow, with staffing at reduced levels. Reported by the supervisor. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_dff841a5fca40b79_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.09,
0.04,
0.12,
0.3,
0.4,
0.05
] | A task decomposition includes a rollback step for each mutation; additionally assumes an API quota that expired last month. Context: under a read-only budget, during peak load. Reported by the supervisor. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_54130c1ed1ed902b_n | noul | [
"false",
"true"
] | [
0.56,
0.44
] | A task decomposition uses batch endpoints where streaming is required; additionally parallelizes steps that share a lock. Context: for a compliance-critical workflow, during the morning shift. Reported by the supervisor. | Is this scenario one where: the plan can complete within budget? | plan_quality | agent | yuri_v3 |
v3_713fdde29303229a_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0,
0.29,
0.65,
0.06
] | An automation plan estimates 2 hours for a task with a known 2-day dependency; additionally calls the refund API before checking dispute status. Context: under a read-only budget, as part of a compliance sweep. Reported by the operator. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_bcbfc548f1f66d1e_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.06,
0.01,
0.03,
0.25,
0.6,
0.05
] | An agent plan skips the sandbox validation step; additionally assumes an API quota that expired last month. Context: with a hard spend cap, as part of a compliance sweep. Reported by the automated check. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_f9dda360e0fd8f05_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0,
0.74,
0.24,
0.02
] | An agent plan assumes an API quota that expired last month; additionally sequences approvals before the technical check. Context: with a hard spend cap, with staffing at reduced levels. Reported by the automated check. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_c6e43fe666e20628_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0.1,
0.43,
0.09,
0.38
] | An automation plan parallelizes steps that share a lock; additionally uses batch endpoints where streaming is required. Context: under a read-only budget, mid-incident. Reported by the monitoring system. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_f49866d5c9129a42_n | noul | [
"false",
"true"
] | [
0.41,
0.59
] | A proposed workflow fetches customer data before verifying the account ID exists; additionally assumes an API quota that expired last month. Context: with a 40-minute deadline, following a recent audit. Reported by the on-call engineer. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_1561a885ffa801fd_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.06,
0.02,
0.02,
0.11,
0.61,
0.18
] | A task decomposition sequences approvals before the technical check; additionally requests write access it never uses. Context: with a 40-minute deadline, immediately after go-live. Reported by the review queue. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_b7484b95f2950c9f_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.05,
0.02,
0.04,
0.28,
0.58,
0.03
] | A task decomposition uses batch endpoints where streaming is required; additionally requests write access it never uses. Context: for a compliance-critical workflow, with a new vendor onboarded. Reported by the supervisor. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_f77a9a22fb2c3429_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.0404,
0.0202,
0.0505,
0.2121,
0.5657,
0.1111
] | An agent plan estimates 2 hours for a task with a known 2-day dependency; additionally handles the failure mode explicitly. Context: with a 40-minute deadline, shortly after a software update. Reported by the monitoring system. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_2199ea34c82ddbfc_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.5,
0.13,
0.27,
0.1
] | A task decomposition ends with a human confirmation gate; additionally fetches customer data before verifying the account ID exists. Context: with a 40-minute deadline, following a recent audit. Reported by the operator. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_39919b1ff3e12e7e_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.13,
0.31,
0.18,
0.31,
0.06,
0.01
] | A proposed workflow handles the failure mode explicitly; additionally sequences approvals before the technical check. Context: for a compliance-critical workflow, during normal operations. Reported by the on-call engineer. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_9c9e90d4ff10c448_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.11,
0.29,
0.12,
0.33,
0.12,
0.03
] | An automation plan handles the failure mode explicitly; additionally parallelizes steps that share a lock. Context: with a hard spend cap, as part of a compliance sweep. Reported by the monitoring system. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_cf55c73888d068ba_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.04,
0.02,
0.07,
0.53,
0.32,
0.02
] | An automation plan skips the sandbox validation step; additionally uses batch endpoints where streaming is required. Context: under a read-only budget, during normal operations. Reported by the monitoring system. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_ab40858c20e66da9_n | noul | [
"false",
"true"
] | [
0.75,
0.25
] | A proposed workflow parallelizes steps that share a lock; additionally sequences approvals before the technical check. Context: with a 40-minute deadline, at the end of the fiscal period. Reported by the on-call engineer. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_834929b62c6a1bdf_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.17,
0.22,
0.16,
0.34,
0.09,
0.02
] | An agent plan uses batch endpoints where streaming is required; additionally handles the failure mode explicitly. Context: with a 40-minute deadline, as part of a compliance sweep. Reported by the operator. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_f4f82fbac181400a_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.01,
0.79,
0.19,
0.01
] | A proposed workflow sequences approvals before the technical check; additionally assumes an API quota that expired last month. Context: with a hard spend cap, during peak load. Reported by the supervisor. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_c757b0e72c2fa035_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.06,
0.03,
0.08,
0.46,
0.34,
0.03
] | A task decomposition uses batch endpoints where streaming is required; additionally includes a rollback step for each mutation. Context: under a read-only budget, with a new vendor onboarded. Reported by the monitoring system. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_b9dd70c40a315caa_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.11,
0.03,
0.08,
0.32,
0.43,
0.03
] | A task decomposition uses batch endpoints where streaming is required; additionally ends with a human confirmation gate. Context: for a compliance-critical workflow, under an active deadline. Reported by the monitoring system. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_3c2fbe5cd2f83545_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.14,
0.17,
0.17,
0.36,
0.13,
0.03
] | An automation plan parallelizes steps that share a lock; additionally uses batch endpoints where streaming is required. Context: with a 40-minute deadline, during an external partner integration. Reported by the supervisor. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_4ba7ad1b503e9233_n | noul | [
"false",
"true"
] | [
0.96,
0.04
] | A task decomposition requests write access it never uses; additionally fetches customer data before verifying the account ID exists. Context: for a compliance-critical workflow, as part of a compliance sweep. Reported by the monitoring system. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_dbfb49fc3e225d22_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0,
0.77,
0.14,
0.09
] | A task decomposition sequences approvals before the technical check; additionally estimates 2 hours for a task with a known 2-day dependency. Context: with a hard spend cap, during an external partner integration. Reported by the monitoring system. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_f5ca8198862eba52_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.03,
0.01,
0.04,
0.2,
0.6,
0.12
] | An automation plan estimates 2 hours for a task with a known 2-day dependency; additionally parallelizes steps that share a lock. Context: with a 40-minute deadline, under reduced staffing. Reported by the monitoring system. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_63e1d13d0afdf691_n | noul | [
"false",
"true"
] | [
0.95,
0.05
] | An automation plan calls the refund API before checking dispute status; additionally skips the sandbox validation step. Context: under a read-only budget, while legacy systems are still active. Reported by the operator. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_7cdb58f1055fa3c8_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0.23,
0.61,
0.11,
0.05
] | An automation plan includes a rollback step for each mutation; additionally fetches customer data before verifying the account ID exists. Context: under a read-only budget, under an active deadline. Reported by the operator. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_5b6af9e9709bc33d_n | noul | [
"false",
"true"
] | [
0.76,
0.24
] | A task decomposition handles the failure mode explicitly; additionally estimates 2 hours for a task with a known 2-day dependency. Context: with a 40-minute deadline, while on-call rotation changed. Reported by the supervisor. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_7cb8dccb3461dc0f_n | noul | [
"false",
"true"
] | [
0.67,
0.33
] | An agent plan sequences approvals before the technical check; additionally calls the refund API before checking dispute status. Context: with a hard spend cap, during an external partner integration. Reported by the automated check. | Is this scenario one where: the plan can complete within budget? | plan_quality | agent | yuri_v3 |
v3_618dbd980e02fab5_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0,
0.01,
0.95,
0.04
] | A proposed workflow handles the failure mode explicitly; additionally sequences approvals before the technical check. Context: with a 40-minute deadline, immediately after go-live. Reported by the automated check. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_a8cefde14f961d9e_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.03,
0.04,
0.04,
0.89
] | A task decomposition skips the sandbox validation step; additionally requests write access it never uses. Context: with a 40-minute deadline, during an external partner integration. Reported by the supervisor. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_3fa17a4512fca752_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.1,
0.31,
0.13,
0.33,
0.1,
0.03
] | An automation plan handles the failure mode explicitly; additionally ends with a human confirmation gate. Context: with a 40-minute deadline, during the quarterly review. Reported by the automated check. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_f29b86096a51f068_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.04,
0,
0.03,
0.93
] | A task decomposition requests write access it never uses; additionally sequences approvals before the technical check. Context: under a read-only budget, during the system migration. Reported by the monitoring system. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_700ecd50bb38daa8_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0.01,
0.05,
0.34,
0.53,
0.06
] | An agent plan fetches customer data before verifying the account ID exists; additionally sequences approvals before the technical check. Context: with a 40-minute deadline, mid-incident. Reported by the on-call engineer. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_2ac6e3cf729a318c_c | choice | [
"data_access",
"external_call",
"timing",
"permissions"
] | [
0.01,
0.64,
0.32,
0.03
] | A task decomposition assumes an API quota that expired last month; additionally ends with a human confirmation gate. Context: with a hard spend cap, following a recent audit. Reported by the automated check. | Weakest step for this scenario. | plan_quality | agent | yuri_v3 |
v3_6c20b6926910828f_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0,
0.74,
0.13,
0.13
] | An automation plan assumes an API quota that expired last month; additionally handles the failure mode explicitly. Context: for a compliance-critical workflow, during a scheduled maintenance window. Reported by the review queue. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_a98431bcea23e4dd_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.05,
0.01,
0.01,
0.06,
0.54,
0.33
] | A proposed workflow estimates 2 hours for a task with a known 2-day dependency; additionally calls the refund API before checking dispute status. Context: with a 40-minute deadline, mid-incident. Reported by the automated check. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_abb7db2cc34ec3b9_n | noul | [
"false",
"true"
] | [
0.76,
0.24
] | A task decomposition skips the sandbox validation step; additionally uses batch endpoints where streaming is required. Context: with a 40-minute deadline, mid-incident. Reported by the on-call engineer. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_b6bca7419a25f7c2_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.02,
0.03,
0.14,
0.49,
0.3,
0.02
] | An automation plan calls the refund API before checking dispute status; additionally ends with a human confirmation gate. Context: with a hard spend cap, during the morning shift. Reported by the monitoring system. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_d03f29a753f8f9ea_n | noul | [
"false",
"true"
] | [
0.71,
0.29
] | A task decomposition sequences approvals before the technical check; additionally ends with a human confirmation gate. Context: with a 40-minute deadline, during the holiday freeze. Reported by the automated check. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
v3_2ffd8aef1ff27eb6_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.08,
0.04,
0.12,
0.47,
0.27,
0.02
] | An agent plan uses batch endpoints where streaming is required; additionally includes a rollback step for each mutation. Context: with a hard spend cap, with a new vendor onboarded. Reported by the automated check. | Rate completion likelihood for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_83ba125d5543f087_n | noul | [
"false",
"true"
] | [
0.84,
0.16
] | A proposed workflow requests write access it never uses; additionally parallelizes steps that share a lock. Context: with a hard spend cap, at the end of the fiscal period. Reported by the monitoring system. | Is this scenario one where: the ordering is correct? | plan_quality | agent | yuri_v3 |
v3_ca8dd4cc0c1619dc_s | score | [
"0",
"1",
"2",
"3",
"4",
"5"
] | [
0.01,
0.01,
0.03,
0.32,
0.6,
0.03
] | A task decomposition estimates 2 hours for a task with a known 2-day dependency; additionally requests write access it never uses. Context: with a hard spend cap, mid-incident. Reported by the operator. | Rate plan quality for this scenario on a 0-5 scale. | plan_quality | agent | yuri_v3 |
v3_16fd45c9d5e925a6_n | noul | [
"false",
"true"
] | [
0.5,
0.5
] | A proposed workflow includes a rollback step for each mutation; additionally skips the sandbox validation step. Context: with a 40-minute deadline, during the morning shift. Reported by the supervisor. | Is this scenario one where: the plan can complete within budget? | plan_quality | agent | yuri_v3 |
v3_02ef1aee0da9f8fa_c | choice | [
"approve_plan",
"revise",
"reject",
"defer_to_human"
] | [
0.06,
0.44,
0.29,
0.21
] | A task decomposition sequences approvals before the technical check; additionally uses batch endpoints where streaming is required. Context: under a read-only budget, at the end of the fiscal period. Reported by the supervisor. | Verdict for this scenario. | plan_quality | agent | yuri_v3 |
v3_8074deb77843f9db_n | noul | [
"false",
"true"
] | [
0.36,
0.64
] | A task decomposition fetches customer data before verifying the account ID exists; additionally sequences approvals before the technical check. Context: with a 40-minute deadline, following a recent audit. Reported by the supervisor. | Is this scenario one where: a step violates least privilege? | plan_quality | agent | yuri_v3 |
End of preview. Expand in Data Studio
JEV corpus - plan_quality
9,714 typed calibrated decision rows from the jev-distill-corpus-v3 corpus (domain: plan_quality).
Each row is one typed decision: kind (noul / choice / score), a state, a question, options, and a target probability distribution over the options.
Part of the per-domain split of SargeDev/jev-distill-corpus-v3 - the full corpus in one repo.
Domains:
- Downloads last month
- 30