{
  "date": "2026-09-27",
  "training": false,
  "models": {
    "balanced-skills-1920": {
      "bundle": "sft_skill_balance_v1/step_0001920_628a0a65099b",
      "sha256": "4cac7f733e6ef63230e26e2aa66418bf0d9661ef55f46fa1e01f2ac602b17407",
      "results": {
        "multiple-choice": {
          "completed": 15428,
          "expected": 15428,
          "status": "complete",
          "tasks": {
            "arc_challenge": {
              "acc": 0.25341296928327645,
              "acc_norm": 0.3003412969283277,
              "completed": 1172,
              "expected": 1172
            },
            "arc_easy": {
              "acc": 0.5904882154882155,
              "acc_norm": 0.5707070707070707,
              "completed": 2376,
              "expected": 2376
            },
            "hellaswag": {
              "acc": 0.30999800836486757,
              "acc_norm": 0.34534953196574386,
              "completed": 10042,
              "expected": 10042
            },
            "piqa": {
              "acc": 0.6474428726877041,
              "acc_norm": 0.6430903155603918,
              "completed": 1838,
              "expected": 1838
            }
          },
          "training": false
        },
        "ifeval": {
          "completed": 541,
          "expected": 541,
          "loose": {
            "instruction_accuracy": 0.2637889688249401,
            "instruction_correct": 220,
            "instruction_total": 834,
            "prompt_accuracy": 0.14417744916820702,
            "prompt_correct": 78,
            "prompt_total": 541
          },
          "status": "complete",
          "stop_counts": {
            "context_limit": 0,
            "eos": 539,
            "length": 2
          },
          "strict": {
            "instruction_accuracy": 0.2637889688249401,
            "instruction_correct": 220,
            "instruction_total": 834,
            "prompt_accuracy": 0.14417744916820702,
            "prompt_correct": 78,
            "prompt_total": 541
          },
          "training": false
        }
      }
    },
    "constraint-completion-2048": {
      "bundle": "sft_constraint_completion_1920_v1/step_0002048_1267d9f4cf7a",
      "sha256": "e4e8a0de28e2141f6a580b49cc950aa64492c9f94d0d3aa98e6a26a2f439764b",
      "results": {
        "multiple-choice": {
          "completed": 15428,
          "expected": 15428,
          "status": "complete",
          "tasks": {
            "arc_challenge": {
              "acc": 0.25426621160409557,
              "acc_norm": 0.31399317406143346,
              "completed": 1172,
              "expected": 1172
            },
            "arc_easy": {
              "acc": 0.5871212121212122,
              "acc_norm": 0.5677609427609428,
              "completed": 2376,
              "expected": 2376
            },
            "hellaswag": {
              "acc": 0.31029675363473413,
              "acc_norm": 0.3472415853415654,
              "completed": 10042,
              "expected": 10042
            },
            "piqa": {
              "acc": 0.64689880304679,
              "acc_norm": 0.6381936887921654,
              "completed": 1838,
              "expected": 1838
            }
          },
          "training": false
        },
        "ifeval": {
          "completed": 541,
          "expected": 541,
          "loose": {
            "instruction_accuracy": 0.29136690647482016,
            "instruction_correct": 243,
            "instruction_total": 834,
            "prompt_accuracy": 0.16820702402957485,
            "prompt_correct": 91,
            "prompt_total": 541
          },
          "status": "complete",
          "stop_counts": {
            "context_limit": 0,
            "eos": 532,
            "length": 9
          },
          "strict": {
            "instruction_accuracy": 0.28776978417266186,
            "instruction_correct": 240,
            "instruction_total": 834,
            "prompt_accuracy": 0.16635859519408502,
            "prompt_correct": 90,
            "prompt_total": 541
          },
          "training": false
        }
      }
    }
  },
  "protocol": {
    "harness_reference_revision": "b954108c9baaaa934b4ad842033b31a97ee30816",
    "datasets": {
      "arc_easy": {
        "repo": "allenai/ai2_arc",
        "revision": "210d026faf9955653af8916fad021475a3f00453",
        "file": "ARC-Easy/test-00000-of-00001.parquet",
        "split": "test",
        "rows": 2376,
        "sha256": "4160597d618ae851c7eb04e281574f3f654776216ac6b6641588d64527b47177"
      },
      "arc_challenge": {
        "repo": "allenai/ai2_arc",
        "revision": "210d026faf9955653af8916fad021475a3f00453",
        "file": "ARC-Challenge/test-00000-of-00001.parquet",
        "split": "test",
        "rows": 1172,
        "sha256": "62f03257e737aed263f55c6abf87c7bb0028a44a6bdd2a26eb1279eb42c1d1e9"
      },
      "piqa": {
        "repo": "baber/piqa",
        "revision": "142f6d7367fd9877f0fb3b5734ea6a545f54cdd1",
        "file": "piqa_validation.parquet",
        "split": "validation",
        "rows": 1838,
        "sha256": "2b17ee1215b1ceb2ddabf4c90a05c5a8b833bef9d207960f57d111c43a2080f3"
      },
      "hellaswag": {
        "repo": "Rowan/hellaswag",
        "revision": "218ec52e09a7e7462a5400043bb9a69a41d06b76",
        "file": "data/validation-00000-of-00001.parquet",
        "split": "validation",
        "rows": 10042,
        "sha256": "899813071e1e95efafec90f856e1987d2150fa4d020fc005df6962c259f660cd"
      },
      "ifeval": {
        "repo": "google/IFEval",
        "revision": "966cd89545d6b6acfd7638bc708b98261ca58e84",
        "file": "ifeval_input_data.jsonl",
        "split": "train",
        "rows": 541,
        "sha256": "6a85310ca8ce15eff755aa08a3a4ff931c7e273e7515ebb3c492ea85fd8288f2"
      }
    },
    "multiple_choice_normalization": "candidate log-likelihood divided by candidate character length",
    "few_shot_examples": 0,
    "ifeval": {
      "prompt_template": "User: {prompt}\nAssistant:",
      "max_context": 2048,
      "max_new_tokens": 1280,
      "greedy": true,
      "repetition_penalty": 1
    }
  },
  "limitations": [
    "Custom MLX evaluator, not an official leaderboard submission.",
    "Public suites repeatedly inspected during development, not an untouched final test.",
    "No full contamination audit or statistical-significance claim.",
    "Instruction compliance does not establish correct content or conversational quality."
  ]
}