Test fixture
Implementation-focused coding tasks with structured correctness checks.
The model receives the prompt (and optional system message). The run uses scorer javascript_function_tests with the JSON configuration below. Pass/fail and partial credit are determined entirely by that scorer against the model output; no human grading.
Return JSON only with a string field named code. The code must be dependency-free JavaScript, define the requested function in the top level or module.exports, and include no markdown, imports, require, timers, network, filesystem, eval, or placeholders.
Implement function applyJsonPatch(doc, operations). Apply a small JSON Patch subset without mutating the original document. Support operations with op add, replace, and remove, using slash-separated paths for object properties and array indexes. Return the patched document.
Your JSON must look like {"code":"function applyJsonPatch(...) { ... }"}.{
"function_name": "applyJsonPatch",
"timeout_ms": 220,
"test_pass_threshold": 1,
"partial_credit_threshold": 0.5,
"tests": "[hidden executable tests]"
}temperature
0
max_tokens
1400
timeout (s)
120
type
scored
file
coding-hard-json-patch.json