{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "https://github.com/stadeummwt/dsh-supreme/schemas/benchmark-record.schema.json",
  "title": "DSH Supreme Benchmark Record",
  "description": "One append-only JSONL line of the supreme-benchmark routing-evidence store. kind=task|run|score. v1.2 adds commitHash/irVersion provenance binding on run records. v1.3 adds the optional anti-sandbagging flag evidenceBacked (run+score), written only when requireEvidenceForScores is enabled.",
  "oneOf": [
    {
      "type": "object",
      "required": ["schemaVersion", "kind", "taskId", "category", "createdAt"],
      "properties": {
        "schemaVersion": { "const": 1 },
        "kind": { "const": "task" },
        "taskId": { "type": "string", "minLength": 1 },
        "category": { "type": "string" },
        "description": { "type": "string" },
        "createdAt": { "type": "number" }
      },
      "additionalProperties": false
    },
    {
      "type": "object",
      "required": ["schemaVersion", "kind", "runId", "taskId", "provider", "model", "profile", "startedAt"],
      "properties": {
        "schemaVersion": { "const": 1 },
        "kind": { "const": "run" },
        "runId": { "type": "string", "minLength": 1 },
        "taskId": { "type": "string" },
        "taskCategory": { "type": "string" },
        "sessionId": { "type": "string" },
        "provider": { "type": "string" },
        "model": { "type": "string" },
        "profile": { "type": "string" },
        "commitHash": {
          "type": "string",
          "description": "40-hex upstream commit sha, or UNAVAILABLE when unresolvable.",
          "oneOf": [{ "const": "UNAVAILABLE" }, { "pattern": "^[a-f0-9]{40}$" }]
        },
        "irVersion": { "type": "string", "pattern": "^[A-Za-z0-9._-]{1,32}$" },
        "startedAt": { "type": "number" },
        "finishedAt": { "type": "number" },
        "latencyMs": { "type": "number" },
        "ttftMs": { "type": "number" },
        "usageIn": { "type": "number" },
        "usageOut": { "type": "number" },
        "toolCount": { "type": "number" },
        "subagentCount": { "type": "number" },
        "workflowCount": { "type": "number" },
        "success": { "type": "boolean" },
        "qualityScore": { "type": "number", "minimum": 0, "maximum": 1 },
        "failureClass": {
          "enum": [
            "AUTH", "RATE_LIMIT", "QUOTA", "TIMEOUT", "NETWORK", "SERVER",
            "INVALID_MODEL", "INVALID_SCHEMA", "WRONG_TOOL", "TOOL_EXECUTION",
            "WRONG_ANSWER", "FORMAT", "CONTEXT", "COST_POLICY", "VERIFICATION", "UNKNOWN"
          ]
        },
        "verification": {
          "type": "object",
          "required": ["validatorId", "status"],
          "properties": {
            "validatorId": { "type": "string" },
            "status": { "enum": ["PASS", "FAIL", "ERROR", "UNAVAILABLE"] },
            "reasonCode": { "type": "string" }
          }
        },
        "evidenceBacked": {
          "type": "boolean",
          "description": "v1.3 anti-sandbagging: quality claim on this run is backed by verifier-PASS evidence. Present only when requireEvidenceForScores is enabled."
        }
      },
      "additionalProperties": false
    },
    {
      "type": "object",
      "required": ["schemaVersion", "kind", "runId", "qualityScore", "scoredAt"],
      "properties": {
        "schemaVersion": { "const": 1 },
        "kind": { "const": "score" },
        "runId": { "type": "string" },
        "qualityScore": { "type": "number", "minimum": 0, "maximum": 1 },
        "validatorId": { "type": "string" },
        "scoredAt": { "type": "number" },
        "evidenceBacked": {
          "type": "boolean",
          "description": "v1.3 anti-sandbagging: score claim is backed by verifier-PASS evidence. Present only when requireEvidenceForScores is enabled."
        }
      },
      "additionalProperties": false
    }
  ]
}