{
  "cases": [
    {
      "name": "on_policy_unpacked",
      "loss_tokens": 14,
      "grpo_loss": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.00714289303869009,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "grpo_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0452161505818367,
        "relative_l2_error": 0.0,
        "cosine_similarity": 0.9999999999999998
      },
      "sdpo_zero_loss": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0,
        "relative_l2_error": null,
        "cosine_similarity": null
      },
      "sdpo_zero_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0,
        "relative_l2_error": null,
        "cosine_similarity": null
      },
      "sdpo_loss": {
        "max_abs_error": 0.13540398702025414,
        "reference_max_abs": 0.09341718256473541,
        "relative_l2_error": 1.4494548358533836,
        "cosine_similarity": -1.0
      },
      "sdpo_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.04461999982595444,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "sdpo_is_clip_2_loss": {
        "max_abs_error": 0.13540398702025414,
        "reference_max_abs": 0.09341718256473541,
        "relative_l2_error": 1.4494548358533836,
        "cosine_similarity": -1.0
      },
      "sdpo_is_clip_2_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.04461999982595444,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "values": {
        "configured_is_loss": 0.04198680445551872,
        "authors_sampled_rkl_loss": -0.09341718256473541,
        "authors_sampled_rkl_is_clip_2_loss": -0.09341718256473541
      }
    },
    {
      "name": "on_policy_packed",
      "loss_tokens": 14,
      "grpo_loss": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.00714289303869009,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "grpo_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0452161505818367,
        "relative_l2_error": 0.0,
        "cosine_similarity": 0.9999999999999998
      },
      "sdpo_zero_loss": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0,
        "relative_l2_error": null,
        "cosine_similarity": null
      },
      "sdpo_zero_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0,
        "relative_l2_error": null,
        "cosine_similarity": null
      },
      "sdpo_loss": {
        "max_abs_error": 0.13540398702025414,
        "reference_max_abs": 0.09341718256473541,
        "relative_l2_error": 1.4494548358533836,
        "cosine_similarity": -1.0
      },
      "sdpo_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.04461999982595444,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "sdpo_is_clip_2_loss": {
        "max_abs_error": 0.13540398702025414,
        "reference_max_abs": 0.09341718256473541,
        "relative_l2_error": 1.4494548358533836,
        "cosine_similarity": -1.0
      },
      "sdpo_is_clip_2_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.04461999982595444,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "values": {
        "configured_is_loss": 0.04198680445551872,
        "authors_sampled_rkl_loss": -0.09341718256473541,
        "authors_sampled_rkl_is_clip_2_loss": -0.09341718256473541
      }
    },
    {
      "name": "stale_unpacked",
      "loss_tokens": 14,
      "grpo_loss": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.01445979718118906,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "grpo_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.05277106165885925,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "sdpo_zero_loss": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0,
        "relative_l2_error": null,
        "cosine_similarity": null
      },
      "sdpo_zero_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0,
        "relative_l2_error": null,
        "cosine_similarity": null
      },
      "sdpo_loss": {
        "max_abs_error": 0.07669473811984062,
        "reference_max_abs": 0.11364800482988358,
        "relative_l2_error": 0.6748445626884763,
        "cosine_similarity": 1.0
      },
      "sdpo_gradient": {
        "max_abs_error": 0.05502688023261726,
        "reference_max_abs": 0.07033934444189072,
        "relative_l2_error": 0.8069652139237306,
        "cosine_similarity": 0.5951560917671562
      },
      "sdpo_is_clip_2_loss": {
        "max_abs_error": 0.01658177375793457,
        "reference_max_abs": 0.053535040467977524,
        "relative_l2_error": 0.3097368305503217,
        "cosine_similarity": 1.0
      },
      "sdpo_is_clip_2_gradient": {
        "max_abs_error": 0.07663571718148887,
        "reference_max_abs": 0.0785936489701271,
        "relative_l2_error": 0.8620550410069173,
        "cosine_similarity": 0.5171523120298515
      },
      "values": {
        "configured_is_loss": 0.036953266710042953,
        "authors_sampled_rkl_loss": 0.11364800482988358,
        "authors_sampled_rkl_is_clip_2_loss": 0.053535040467977524
      }
    },
    {
      "name": "stale_packed",
      "loss_tokens": 14,
      "grpo_loss": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.01445979718118906,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "grpo_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.05277106165885925,
        "relative_l2_error": 0.0,
        "cosine_similarity": 1.0
      },
      "sdpo_zero_loss": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0,
        "relative_l2_error": null,
        "cosine_similarity": null
      },
      "sdpo_zero_gradient": {
        "max_abs_error": 0.0,
        "reference_max_abs": 0.0,
        "relative_l2_error": null,
        "cosine_similarity": null
      },
      "sdpo_loss": {
        "max_abs_error": 0.07669473811984062,
        "reference_max_abs": 0.11364800482988358,
        "relative_l2_error": 0.6748445626884763,
        "cosine_similarity": 1.0
      },
      "sdpo_gradient": {
        "max_abs_error": 0.05502688023261726,
        "reference_max_abs": 0.07033934444189072,
        "relative_l2_error": 0.8069652139237306,
        "cosine_similarity": 0.5951560917671562
      },
      "sdpo_is_clip_2_loss": {
        "max_abs_error": 0.01658177375793457,
        "reference_max_abs": 0.053535040467977524,
        "relative_l2_error": 0.3097368305503217,
        "cosine_similarity": 1.0
      },
      "sdpo_is_clip_2_gradient": {
        "max_abs_error": 0.07663571718148887,
        "reference_max_abs": 0.0785936489701271,
        "relative_l2_error": 0.8620550410069173,
        "cosine_similarity": 0.5171523120298515
      },
      "values": {
        "configured_is_loss": 0.036953266710042953,
        "authors_sampled_rkl_loss": 0.11364800482988358,
        "authors_sampled_rkl_is_clip_2_loss": 0.053535040467977524
      }
    }
  ],
  "effective_validation": {
    "status": "failed",
    "required": true,
    "checks": [
      {
        "path": "grpo_judge_advantage_via_custom_path",
        "check": "baseline_loss",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.001
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.00714289303869009,
          "relative_l2_error": 0.0,
          "cosine_similarity": 1.0
        }
      },
      {
        "path": "grpo_judge_advantage_via_custom_path",
        "check": "baseline_gradient",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.01,
          "min_cosine": 0.999
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.0452161505818367,
          "relative_l2_error": 0.0,
          "cosine_similarity": 0.9999999999999998
        }
      },
      {
        "path": "grpo_judge_advantage_via_custom_path",
        "check": "reference_loss",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.001
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.00714289303869009,
          "relative_l2_error": 0.0,
          "cosine_similarity": 1.0
        }
      },
      {
        "path": "grpo_judge_advantage_via_custom_path",
        "check": "reference_gradient",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.01,
          "min_cosine": 0.999
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.0452161505818367,
          "relative_l2_error": 0.0,
          "cosine_similarity": 0.9999999999999998
        }
      },
      {
        "path": "sdpo_self_teacher_advantage",
        "check": "baseline_loss",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.001
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.0,
          "relative_l2_error": null,
          "cosine_similarity": null
        }
      },
      {
        "path": "sdpo_self_teacher_advantage",
        "check": "baseline_gradient",
        "status": "unverified",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.01,
          "min_cosine": 0.999
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.0,
          "relative_l2_error": null,
          "cosine_similarity": null
        }
      },
      {
        "path": "sdpo_self_teacher_advantage",
        "check": "reference_loss",
        "status": "failed",
        "tolerance": {
          "atol": 0.001,
          "rtol": 0.01
        },
        "measured": {
          "max_abs_error": 0.13540398702025414,
          "reference_max_abs": 0.09341718256473541,
          "relative_l2_error": 1.4494548358533836,
          "cosine_similarity": -1.0
        }
      },
      {
        "path": "sdpo_self_teacher_advantage",
        "check": "reference_gradient",
        "status": "passed",
        "tolerance": {
          "atol": 0.001,
          "rtol": 0.02,
          "min_cosine": 0.995
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.04461999982595444,
          "relative_l2_error": 0.0,
          "cosine_similarity": 1.0
        }
      }
    ],
    "reason": "sdpo_self_teacher_advantage: reference_loss; sdpo_self_teacher_advantage: baseline_gradient cosine/tolerance",
    "evidence_source": "executor report; remote execution is not attested",
    "applicability": "required"
  },
  "immutable_spec_validation": {
    "status": "failed",
    "required": true,
    "checks": [
      {
        "path": "grpo_judge_advantage_via_custom_path",
        "check": "baseline_loss",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.001
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.00714289303869009,
          "relative_l2_error": 0.0,
          "cosine_similarity": 1.0
        }
      },
      {
        "path": "grpo_judge_advantage_via_custom_path",
        "check": "baseline_gradient",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.01,
          "min_cosine": 0.999
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.0452161505818367,
          "relative_l2_error": 0.0,
          "cosine_similarity": 0.9999999999999998
        }
      },
      {
        "path": "grpo_judge_advantage_via_custom_path",
        "check": "reference_loss",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.001
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.00714289303869009,
          "relative_l2_error": 0.0,
          "cosine_similarity": 1.0
        }
      },
      {
        "path": "grpo_judge_advantage_via_custom_path",
        "check": "reference_gradient",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.01,
          "min_cosine": 0.999
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.0452161505818367,
          "relative_l2_error": 0.0,
          "cosine_similarity": 0.9999999999999998
        }
      },
      {
        "path": "sdpo_self_teacher_advantage",
        "check": "baseline_loss",
        "status": "passed",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.001
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.0,
          "relative_l2_error": null,
          "cosine_similarity": null
        }
      },
      {
        "path": "sdpo_self_teacher_advantage",
        "check": "baseline_gradient",
        "status": "unverified",
        "tolerance": {
          "atol": 0.0001,
          "rtol": 0.01,
          "min_cosine": 0.999
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.0,
          "relative_l2_error": null,
          "cosine_similarity": null
        }
      },
      {
        "path": "sdpo_self_teacher_advantage",
        "check": "reference_loss",
        "status": "failed",
        "tolerance": {
          "atol": 0.001,
          "rtol": 0.01
        },
        "measured": {
          "max_abs_error": 0.13540398702025414,
          "reference_max_abs": 0.09341718256473541,
          "relative_l2_error": 1.4494548358533836,
          "cosine_similarity": -1.0
        }
      },
      {
        "path": "sdpo_self_teacher_advantage",
        "check": "reference_gradient",
        "status": "passed",
        "tolerance": {
          "atol": 0.001,
          "rtol": 0.02,
          "min_cosine": 0.995
        },
        "measured": {
          "max_abs_error": 0.0,
          "reference_max_abs": 0.04461999982595444,
          "relative_l2_error": 0.0,
          "cosine_similarity": 1.0
        }
      }
    ],
    "reason": "sdpo_self_teacher_advantage: reference_loss; grpo_judge_advantage_via_custom_path: matching reference_revision; grpo_judge_advantage_via_custom_path: matching baseline_revision; sdpo_self_teacher_advantage: matching reference_revision; sdpo_self_teacher_advantage: matching baseline_revision; sdpo_self_teacher_advantage: baseline_gradient cosine/tolerance",
    "evidence_source": "executor report; remote execution is not attested",
    "applicability": "required"
  },
  "torch_version": "2.14.0+cu130",
  "seconds": 0.035233414000003904,
  "optimizer_updates": 0,
  "training_clients_created": 0,
  "source_hashes": {
    "sdpo_core_algos": "4f949c323f005b4945f1983bdf89a9ddbd53865d0d7511993c37fbe7c47d6903",
    "reference_extracts": "a85127a983359636e38df5036635b9bd7c0f399d9e3ef1a2fcfbc463c131ab00"
  },
  "reference_config": {
    "full_logit_distillation": false,
    "alpha": 1.0,
    "is_clip": null
  },
  "additional_reference_config": {
    "full_logit_distillation": false,
    "alpha": 1.0,
    "is_clip": 2.0
  }
}