{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "https://agenteval.dev/schemas/suites/1/definition.schema.json",
  "title": "AgentEval suite definition manifest 1.0 (external to AEF)",
  "type": "object",
  "additionalProperties": false,
  "required": ["schemaVersion", "kind", "suiteRef", "definitionVersion", "frozen", "inputs", "definition", "caseIds"],
  "properties": {
    "schemaVersion": { "const": "1.0" },
    "kind": { "enum": ["dataset", "probe", "composite"] },
    "suiteRef": { "type": "string", "minLength": 1, "maxLength": 1024 },
    "definitionVersion": { "type": "string", "minLength": 1, "maxLength": 128 },
    "frozen": { "type": "boolean" },
    "inputs": {
      "description": "At most 10,000 named exact input content digests (datasets, prompt material, code/rubric definitions), enforced by the reader. No raw prompt or dataset text is required.",
      "type": "object",
      "additionalProperties": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" }
    },
    "definition": {
      "description": "Producer-owned declared evaluation configuration, including thresholds, weighting, aggregation and trial/generation policies. No observed scores, durations or run IDs.",
      "type": "object", "minProperties": 1
    },
    "caseIds": {
      "description": "Complete planned case roster, not observed results; order has no meaning here. Ordered case content belongs in the hashed inputs.",
      "type": "array", "maxItems": 100000, "uniqueItems": true,
      "items": { "type": "string", "minLength": 1, "maxLength": 256 }
    }
  }
}