{
  "schema_version": "1.0",
  "seed": "yao-output-eval-blind-v1",
  "summary": {
    "pair_count": 5,
    "answer_key_separate": true,
    "with_skill_hidden_count": 5
  },
  "pairs": [
    {
      "case_id": "skill-package-contract",
      "prompt": "Turn this repeated workflow into a reusable team skill package.",
      "input_files": [],
      "metadata": {},
      "review_instruction": "Pick A or B based only on the rubric. Do not infer which output came from the skill.",
      "rubric": [
        {
          "id": "has-entrypoint",
          "description": "Output names the routeable Skill entrypoint.",
          "weight": 1.0
        },
        {
          "id": "has-interface",
          "description": "Output includes neutral interface metadata.",
          "weight": 1.0
        },
        {
          "id": "has-report",
          "description": "Output points reviewer to generated overview report.",
          "weight": 1.0
        },
        {
          "id": "has-resource-boundary",
          "description": "Output preserves entrypoint/resource separation.",
          "weight": 1.0
        }
      ],
      "variant_a": {
        "blind_id": "skill-package-contract:A",
        "output": "Create a routeable SKILL.md, agents/interface.yaml, reports/skill-overview.html, and a focused eval plan. Keep the root entrypoint lean, move durable guidance into references/, scripts into scripts/, and evidence into reports/."
      },
      "variant_b": {
        "blind_id": "skill-package-contract:B",
        "output": "I can write a prompt for that workflow and include a short checklist."
      }
    },
    {
      "case_id": "output-eval-expectation",
      "prompt": "Upgrade this production skill so we know whether the generated output is better than baseline.",
      "input_files": [],
      "metadata": {},
      "review_instruction": "Pick A or B based only on the rubric. Do not infer which output came from the skill.",
      "rubric": [
        {
          "id": "has-baseline-comparison",
          "description": "Output explicitly compares with-skill and baseline outputs.",
          "weight": 1.0
        },
        {
          "id": "has-assertion-grading",
          "description": "Output includes objective assertion grading.",
          "weight": 1.0
        },
        {
          "id": "has-scorecard",
          "description": "Output produces a scorecard report path.",
          "weight": 1.0
        },
        {
          "id": "has-failure-taxonomy",
          "description": "Output records failure taxonomy.",
          "weight": 1.0
        }
      ],
      "variant_a": {
        "blind_id": "output-eval-expectation:A",
        "output": "Add Output Eval Lab cases with baseline_output and with_skill_output, run assertion grading, report with-skill vs baseline pass-rate delta, and record failure taxonomy plus next fixes in reports/output_quality_scorecard.md."
      },
      "variant_b": {
        "blind_id": "output-eval-expectation:B",
        "output": "Add more examples and run the trigger eval again."
      }
    },
    {
      "case_id": "ir-before-packaging",
      "prompt": "Package this skill for OpenAI, Claude, Agent Skills, and generic targets.",
      "input_files": [],
      "metadata": {},
      "review_instruction": "Pick A or B based only on the rubric. Do not infer which output came from the skill.",
      "rubric": [
        {
          "id": "has-ir-first",
          "description": "Output requires Skill IR before packaging.",
          "weight": 1.0
        },
        {
          "id": "has-core-ir-fields",
          "description": "Output lists core IR contract fields.",
          "weight": 1.0
        },
        {
          "id": "has-targets",
          "description": "Output names requested runtime targets.",
          "weight": 1.0
        },
        {
          "id": "has-semantic-preservation",
          "description": "Output says targets must preserve the capability contract.",
          "weight": 1.0
        }
      ],
      "variant_a": {
        "blind_id": "ir-before-packaging:A",
        "output": "Export folders for each platform and adjust files as needed."
      },
      "variant_b": {
        "blind_id": "ir-before-packaging:B",
        "output": "Export Skill IR first with job_to_be_done, trigger_surface, workflow, resources, eval_plan, risk, and governance. Then compile or package targets from the IR so OpenAI, Claude, Agent Skills, and generic packages preserve the same capability contract."
      }
    },
    {
      "case_id": "near-neighbor-boundary",
      "prompt": "I only need a one-off summary of these notes, not a reusable process.",
      "input_files": [],
      "metadata": {
        "case_type": "near_neighbor",
        "tier": "production"
      },
      "review_instruction": "Pick A or B based only on the rubric. Do not infer which output came from the skill.",
      "rubric": [
        {
          "id": "declines-skill",
          "description": "Output refuses unnecessary skill creation for one-off work.",
          "weight": 1.0
        },
        {
          "id": "names-near-neighbor",
          "description": "Output labels the request as near-neighbor instead of owned work.",
          "weight": 1.0
        },
        {
          "id": "requires-reuse-signal",
          "description": "Output asks for repeat-use evidence before packaging.",
          "weight": 1.0
        }
      ],
      "variant_a": {
        "blind_id": "near-neighbor-boundary:A",
        "output": "Create a SKILL.md and a reusable workflow anyway so future notes can use it."
      },
      "variant_b": {
        "blind_id": "near-neighbor-boundary:B",
        "output": "Do not create a skill for this one-off request. Treat it as a near-neighbor: answer the summary directly unless the user confirms repeated use, shared ownership, or a reusable output contract."
      }
    },
    {
      "case_id": "file-backed-governed-package",
      "prompt": "Turn the attached release brief source into a governed skill package.",
      "input_files": [
        "fixtures/release-brief-source.md"
      ],
      "metadata": {
        "case_type": "boundary",
        "tier": "governed"
      },
      "review_instruction": "Pick A or B based only on the rubric. Do not infer which output came from the skill.",
      "rubric": [
        {
          "id": "uses-file-backed-evidence",
          "description": "Output names file-backed source evidence.",
          "weight": 1.0
        },
        {
          "id": "has-governance",
          "description": "Output preserves governed ownership metadata.",
          "weight": 1.0
        },
        {
          "id": "has-output-contract",
          "description": "Output preserves output and rollback boundaries.",
          "weight": 1.0
        },
        {
          "id": "has-trust-and-scorecard",
          "description": "Output requires trust and output scorecard artifacts.",
          "weight": 1.0
        },
        {
          "id": "does-not-invent-evidence",
          "description": "Output forbids invented launch evidence.",
          "weight": 1.0
        }
      ],
      "variant_a": {
        "blind_id": "file-backed-governed-package:A",
        "output": "Draft a release announcement with the changelog, support notes, and owner name."
      },
      "variant_b": {
        "blind_id": "file-backed-governed-package:B",
        "output": "Use the file-backed fixture as source evidence, then create a governed skill package with SKILL.md, agents/interface.yaml, owner, review cadence, input_files, output contract, rollback boundary, trust report, and reports/output_quality_scorecard.md. Mark missing launch metrics as missing evidence instead of inventing them."
      }
    }
  ]
}