{
  "name": "coding",
  "description": "Tool-free software-engineering tasks focused on algorithms, debugging, language semantics, API design, testing judgment, and instruction following.",
  "defaults": { "runs": 3, "temperature": 0, "maxTokens": 512, "reasoning": "low" },
  "tasks": [
    {
      "id": "complexity",
      "tags": ["algorithms", "exactness"],
      "prompt": "What is the time complexity of binary search over a sorted array of n items? Return only the Big-O expression.",
      "grader": { "type": "normalized-exact", "expected": "O(log n)" }
    },
    {
      "id": "bug-diagnosis",
      "tags": ["debugging", "instruction-following"],
      "prompt": "A loop runs `for (let i = 0; i <= items.length; i++) { console.log(items[i].name); }`. Return exactly two lines: first line is the bug, second line is the fix. Do not include a code fence.",
      "grader": { "type": "regex", "pattern": "(?:<=|one iteration|out of bounds|undefined).*(?:<\\s*items\\.length|less than|replace)", "flags": "is" }
    },
    {
      "id": "api-contract",
      "tags": ["structured-output", "design"],
      "prompt": "Return JSON only, with exactly these string fields: `method` must be `GET`, `path` must be `/users/{id}`, and `auth` must be `bearer`.",
      "grader": { "type": "json-exact", "expected": { "method": "GET", "path": "/users/{id}", "auth": "bearer" } }
    },
    {
      "id": "scope-control",
      "tags": ["instruction-following"],
      "prompt": "Explain in one sentence why a unit test should not depend on the system clock. Your answer must contain the words `deterministic` and `reliable`.",
      "grader": { "type": "contains", "values": ["deterministic", "reliable"] }
    },
    {
      "id": "loop-trace",
      "tags": ["algorithms", "state-tracing", "exactness"],
      "prompt": "Return only the integer value. Start with `value = 7`. For i from 1 through 5, if i is odd add i*3 to value; otherwise divide value by 2. What is value after the loop?",
      "grader": { "type": "number", "expected": 22, "tolerance": 0 }
    },
    {
      "id": "js-coercion",
      "tags": ["language-semantics", "structured-output", "exactness"],
      "prompt": "Return JSON only. Evaluate these JavaScript expressions in order and return their string results: `\"\" || \"fallback\"`, `0 || \"fallback\"`, `false || \"fallback\"`, `null || \"fallback\"`, and `\"ok\" || \"fallback\"`.",
      "grader": { "type": "json-exact", "expected": ["fallback", "fallback", "fallback", "fallback", "ok"] }
    },
    {
      "id": "test-boundaries",
      "tags": ["testing", "boundaries", "structured-output"],
      "prompt": "An API accepts an inclusive integer range from 0 through 10. Return JSON only: the four smallest useful boundary test inputs, in ascending order, covering just below, the lower bound, the upper bound, and just above.",
      "grader": { "type": "json-exact", "expected": [-1, 0, 10, 11] }
    },
    {
      "id": "api-evolution",
      "tags": ["api", "backward-compatibility", "structured-output"],
      "prompt": "Return JSON only with exactly two string fields: `breaking` and `reason`. A server adds an optional response field that old clients ignore. Set `breaking` to `false` and explain why this change is backward compatible in one short phrase.",
      "grader": { "type": "json-exact", "expected": { "breaking": "false", "reason": "optional response fields are backward compatible" } }
    }
  ]
}