{
  "id": "research-default",
  "version": "0.1.0",
  "description": "Deep research profile. Uses citation footnotes (e.g., [1], [2]) with sources listed at document end. Users can assess reliability themselves. No hard threshold on groundedness. Suited for literature reviews, comprehensive surveys, and multi-source report generation.",
  "domain": "research",

  "assumedModel": {
    "minTier": "strong",
    "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 research goal and current stage.",
      "kind": "read-file",
      "spec": { "path": "task.json" },
      "failurePolicy": "fail-fast"
    },
    {
      "id": "read-progress",
      "description": "Read progress.json to see which sources and claims are already verified.",
      "kind": "read-file",
      "spec": { "path": "progress.json" },
      "failurePolicy": "fail-fast"
    },
    {
      "id": "read-latest-checkpoint",
      "description": "Read the most recent checkpoint to resume context and source list.",
      "kind": "read-file",
      "spec": { "path": "checkpoints/latest.json", "optional": true },
      "failurePolicy": "report-and-continue"
    }
  ],

  "progressSchema": {
    "kind": "checklist",
    "itemFields": {
      "source": "string",
      "claim": "string",
      "verified": "boolean",
      "notes": "string"
    },
    "appendOnly": true
  },

  "contractTemplate": {
    "requiredSections": ["scope", "doneCriteria", "verificationPlan", "outOfScope"],
    "minMachineChecks": 1,
    "maxNegotiationRounds": 3
  },

  "evaluator": {
    "intensity": "deep",
    "requireExternalCheck": false,
    "fewShotCalibrationPath": "data/few-shot-rubrics/research.md",
    "checklist": {
      "hard": [
        {
          "id": "H1",
          "title": "Citations present",
          "passDescription": "citations exist and link to identifiable sources",
          "failReworkDescription": "citations are missing but sources clearly exist to support the claims",
          "failEscalateDescription": "the claims are about proprietary/internal information that cannot be publicly cited"
        },
        {
          "id": "H2",
          "title": "Completeness",
          "passDescription": "goal is addressed with specifics",
          "failReworkDescription": "goal is partially addressed; more work would help",
          "failEscalateDescription": "the goal asks for information not publicly available"
        },
        {
          "id": "H3",
          "title": "Factual specificity",
          "passDescription": "≥5 specific verifiable facts (names, dates, figures)",
          "failReworkDescription": "mostly generic statements; more research can fix",
          "failEscalateDescription": "specifics genuinely do not exist in public domain"
        }
      ],
      "soft": [
        { "id": "S1", "title": "Source diversity", "weight": 0.3, "scoringGuide": "5 = diverse primary + secondary; 1 = single source repeated" },
        { "id": "S2", "title": "Analytical depth", "weight": 0.3, "scoringGuide": "5 = synthesis with identified tensions; 1 = list of facts" },
        { "id": "S3", "title": "Actionable conclusions", "weight": 0.2, "scoringGuide": "5 = specific action items with rationale; 1 = description only" },
        { "id": "S4", "title": "Clarity and structure", "weight": 0.2, "scoringGuide": "5 = clear headings, logical flow; 1 = disorganized" }
      ]
    },
    "rubric": {
      "criteria": [
        {
          "id": "completeness",
          "weight": 0.25,
          "description": "All declared research questions and done-criteria are addressed with evidence.",
          "scoringGuide": "5 = every question answered with direct evidence; 1 = most questions unanswered."
        },
        {
          "id": "groundedness",
          "weight": 0.35,
          "description": "Factual claims use citation footnotes (e.g., [1], [2]) with sources listed in 'References' section.",
          "scoringGuide": "5 = all key claims have footnote citations or [citation needed] markers; 1 = significant claims without any citation mechanism."
        },
        {
          "id": "source-diversity",
          "weight": 0.2,
          "description": "Multiple independent sources consulted; not over-reliant on a single source.",
          "scoringGuide": "5 = diverse primary and secondary sources; 1 = single source used throughout."
        },
        {
          "id": "analytical-depth",
          "weight": 0.1,
          "description": "Analysis goes beyond summary; synthesises across sources and identifies tensions.",
          "scoringGuide": "5 = genuine synthesis with identified contradictions; 1 = raw summaries only."
        },
        {
          "id": "clarity",
          "weight": 0.1,
          "description": "Report is structured, readable, and a successor worker could resume from it.",
          "scoringGuide": "5 = self-contained, well-structured; 1 = requires original chat to interpret."
        }
      ]
    },
    "hardThresholds": [
      { "criterionId": "completeness", "min": 3 }
    ]
  },

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

  "recovery": {
    "resumeStrategy": "reset",
    "maxReworkPerStage": 3,
    "contextResetThresholdRatio": 0.65,
    "stuckDetection": {
      "noProgressReworkLimit": 2
    }
  },

  "budget": {
    "maxTokens": 800000,
    "maxWallClockMinutes": 240,
    "perStageTokenCap": 200000
  },

  "benchGoal": {
    "goal": "Summarise the key arguments for and against static typing in programming languages and write a brief report to report.md.",
    "expectedArtifacts": ["report.md"],
    "maxTokens": 50000,
    "maxWallClockMs": 120000
  }
}