{
  "id": "curation-default",
  "version": "0.1.0",
  "description": "Data curation profile. Schema-compliance and consistency are primary quality gates. Suited for content classification, dataset cleaning, taxonomy building, and structured output preparation.",
  "domain": "curation",

  "assumedModel": {
    "minTier": "standard",
    "capabilities": ["long-context", "tool-use"]
  },

  "strategy": {
    "kind": "single-worker",
    "evaluatorEnabled": true
  },

  "bootstrap": [
    {
      "id": "read-task",
      "description": "Read the saga task.json so the worker knows the curation goal and current stage.",
      "kind": "read-file",
      "spec": { "path": "task.json" },
      "failurePolicy": "fail-fast"
    },
    {
      "id": "read-progress",
      "description": "Read progress.json to see which items have already been processed and validated.",
      "kind": "read-file",
      "spec": { "path": "progress.json" },
      "failurePolicy": "fail-fast"
    },
    {
      "id": "read-latest-checkpoint",
      "description": "Read the most recent checkpoint to resume batch processing context.",
      "kind": "read-file",
      "spec": { "path": "checkpoints/latest.json", "optional": true },
      "failurePolicy": "report-and-continue"
    }
  ],

  "progressSchema": {
    "kind": "checklist",
    "itemFields": {
      "itemId": "string",
      "category": "string",
      "schemaValid": "boolean",
      "notes": "string"
    },
    "appendOnly": true
  },

  "contractTemplate": {
    "requiredSections": ["scope", "doneCriteria", "verificationPlan", "outOfScope"],
    "minMachineChecks": 2,
    "maxNegotiationRounds": 3
  },

  "evaluator": {
    "intensity": "standard",
    "requireExternalCheck": false,
    "fewShotCalibrationPath": "data/few-shot-rubrics/curation.md",
    "checklist": {
      "hard": [
        {
          "id": "H1",
          "title": "Schema compliance",
          "passDescription": "every output record validates against the declared schema",
          "failReworkDescription": "some records have schema violations that can be fixed",
          "failEscalateDescription": "the declared schema is contradictory or the source data cannot fit the schema"
        },
        {
          "id": "H2",
          "title": "Coverage",
          "passDescription": "outputCount / inputCount ≥ declaredThreshold (default 0.95 unless filtering was explicitly requested)",
          "failReworkDescription": "coverage is below threshold but more items can be processed",
          "failEscalateDescription": "source data is too incomplete or corrupt to reach the coverage threshold"
        },
        {
          "id": "H3",
          "title": "Subjective scores calibrated",
          "passDescription": "for any 1–5 score field, spot-checks show scores are not all clumped (e.g., not 90% scored 3)",
          "failReworkDescription": "score distribution is skewed; recalibration in rework can fix it",
          "failEscalateDescription": "the scoring criteria are inherently subjective and cannot be made consistent"
        }
      ],
      "soft": [
        { "id": "S1", "title": "Schema correctness", "weight": 0.3, "scoringGuide": "5 = zero schema violations; 1 = widespread violations" },
        { "id": "S2", "title": "Subjective-score consistency", "weight": 0.3, "scoringGuide": "5 = scores follow rubric consistently; 1 = random-seeming scores" },
        { "id": "S3", "title": "Coverage", "weight": 0.2, "scoringGuide": "5 = all scoped items processed; 1 = major gaps" },
        { "id": "S4", "title": "Output organization", "weight": 0.2, "scoringGuide": "5 = well-structured, easy to navigate; 1 = disorganized dump" }
      ]
    },
    "rubric": {
      "criteria": [
        {
          "id": "completeness",
          "weight": 0.25,
          "description": "All items in scope have been processed and appear in the output.",
          "scoringGuide": "5 = 100% of scoped items processed; 1 = major gaps in coverage."
        },
        {
          "id": "consistency",
          "weight": 0.3,
          "description": "Similar items are categorised/labelled consistently throughout the dataset.",
          "scoringGuide": "5 = no inconsistencies found; 1 = same item type labelled differently in multiple places."
        },
        {
          "id": "schema-compliance",
          "weight": 0.3,
          "description": "All output records conform to the declared output schema.",
          "scoringGuide": "5 = zero schema violations; 1 = widespread schema violations."
        },
        {
          "id": "coverage",
          "weight": 0.1,
          "description": "All declared categories/tags have been applied where appropriate; no under-coverage.",
          "scoringGuide": "5 = all categories well-represented; 1 = major categories missing."
        },
        {
          "id": "clarity",
          "weight": 0.05,
          "description": "The curation report is clear and a successor worker can understand decisions made.",
          "scoringGuide": "5 = decision rationale documented throughout; 1 = opaque categorisation."
        }
      ]
    },
    "hardThresholds": [
      { "criterionId": "consistency", "min": 3 },
      { "criterionId": "schema-compliance", "min": 4 }
    ]
  },

  "invariants": [
    {
      "kind": "file-exists",
      "id": "progress-file",
      "path": "progress.json"
    }
  ],

  "recovery": {
    "resumeStrategy": "reset",
    "maxReworkPerStage": 3,
    "contextResetThresholdRatio": 0.7,
    "stuckDetection": {
      "noProgressReworkLimit": 2
    }
  },

  "budget": {
    "maxTokens": 400000,
    "maxWallClockMinutes": 120,
    "perStageTokenCap": 100000
  },

  "benchGoal": {
    "goal": "Classify 5 product descriptions into categories (electronics, clothing, food, books) and write results to categories.json.",
    "expectedArtifacts": ["categories.json"],
    "maxTokens": 50000,
    "maxWallClockMs": 120000
  }
}