Submission format.

One complete folder / 一份完整文件夹

Every package includes README.md, LICENSE and a type-specific JSON: benchmark.json, harness.json or case.json. Keep the original source and task layout. Upload files directly; archives are not extracted.

每份材料包含 README.md、LICENSE 和对应类型的 JSON。保留任务或源码原有的目录结构；直接上传文件夹，服务不解压压缩包。

The service generates submission.json from the verified account, version and file inventory. Each file is verified against its byte count and SHA256. Limit: 1 GiB and 2,000 files per version. JSON/TOML documents: 4 MiB; JSONL lines: 1 MiB.

服务根据当前账号、版本及文件清单生成 submission.json，并核验每个文件的字节数和 SHA256。每版最多 1 GiB / 2,000 个文件；JSON/TOML 文档不超过 4 MiB，JSONL 单行不超过 1 MiB。

References and receipts / 引用与回执

Failure cases reference benchmark and harness versions already accepted under your account. Copy the artifact object from each receipt into case.json and the run lock. The service checks ownership, type, version, hash and task/split membership.

Failure Case 引用你账号下已受理的 Benchmark 和 Harness。将回执的 artifact 对象填入 case.json 和运行锁中的对应引用字段，系统核对所有权、类型、版本、哈希和任务划分。

{
  "transport": "server",
  "submission_id": "<ID from receipt>",
  "version": 1,
  "submission_sha256": "<64-character SHA256 from receipt>"
}

A receipt confirms private intake and static format validation. It does not approve a benchmark, execute submitted programs or establish a reproduced result. Synthetic examples are explicitly marked and retain unknown runtime hashes and scores as null.

回执只确认材料已私下受理并通过静态格式校验，不代表任务已接纳、程序已执行或失败已复现。合成示例会明确标注，未知运行环境哈希与分数仍保留为空。