Evals · Schema

EvalCase

A single test case in an eval suite: the input given to the system-under-test plus optional expected output, retrieval context, and metadata. The smallest reusable unit of evaluation — equivalent to a row in a Hugging Face dataset, a 'golden' in DeepEval, or an example in LangSmith.

EvalsLLM EvaluationAI QualityBenchmarksLLM as a JudgeObservabilityAgent EvaluationRAG EvaluationTest-Driven AI

Properties

Name Type Description
id string
input string The prompt or input string given to the system-under-test.
input_structured object Structured input when the case is not a single string (multi-turn conversation, agent goal, tool-calling task).
expected string Ground-truth expected output for reference-based scoring. Optional.
expected_structured object Structured ground-truth (expected tool calls, expected agent trajectory, JSON answer).
context array Optional retrieval context provided to the model for RAG-style cases.
metadata object Arbitrary metadata: difficulty, domain, persona, source URL, license.
tags array
View JSON Schema on GitHub

JSON Schema

evals-eval-case-schema.json Raw ↑ 
{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "https://raw.githubusercontent.com/api-evangelist/evals/refs/heads/main/json-schema/evals-eval-case-schema.json",
  "title": "EvalCase",
  "description": "A single test case in an eval suite: the input given to the system-under-test plus optional expected output, retrieval context, and metadata. The smallest reusable unit of evaluation — equivalent to a row in a Hugging Face dataset, a 'golden' in DeepEval, or an example in LangSmith.",
  "type": "object",
  "properties": {
    "id": {
      "type": "string",
      "example": "case_0042"
    },
    "input": {
      "type": "string",
      "description": "The prompt or input string given to the system-under-test.",
      "example": "What is the refund window for a damaged item?"
    },
    "input_structured": {
      "type": "object",
      "description": "Structured input when the case is not a single string (multi-turn conversation, agent goal, tool-calling task)."
    },
    "expected": {
      "type": "string",
      "description": "Ground-truth expected output for reference-based scoring. Optional.",
      "example": "30 days from delivery for damaged items."
    },
    "expected_structured": {
      "type": "object",
      "description": "Structured ground-truth (expected tool calls, expected agent trajectory, JSON answer)."
    },
    "context": {
      "type": "array",
      "items": { "type": "string" },
      "description": "Optional retrieval context provided to the model for RAG-style cases.",
      "example": ["Refund policy section 4.2: Damaged items may be returned within 30 days of delivery for a full refund."]
    },
    "metadata": {
      "type": "object",
      "description": "Arbitrary metadata: difficulty, domain, persona, source URL, license.",
      "properties": {
        "domain": { "type": "string", "example": "customer-support" },
        "difficulty": { "type": "string", "enum": ["easy", "medium", "hard"], "example": "easy" },
        "persona": { "type": "string", "example": "first-time-buyer" },
        "source": { "type": "string", "format": "uri" }
      }
    },
    "tags": {
      "type": "array",
      "items": { "type": "string" },
      "example": ["policy", "refunds"]
    }
  },
  "required": ["id", "input"]
}