5-minute harness upload

1. Validate the tree

Your archive must contain harness.json and an entrypoint that accepts --task and --workdir. Run the local validator before creating the archive:

python3 tools/validate_harness.py /path/to/harness --smoke

The smoke check must write an answer artifact. Keep the archive below 20 MiB compressed and 200 MiB after extraction.

2. Start the reviewed console

Use the endpoint supplied by the project maintainer. For a local mock:

python3 console/uat/mock_submissions.py --port 4471

Open /submit.html?api=http://127.0.0.1:4471, choose an available scenario, and select your archive. The page validates the extension and compressed size before it sends anything.

3. Confirm the measured state

The service returns a submission_id. The page follows validating, queued, and running until it reaches scored, failed, or rejected. A scored response shows each returned pool on its own row, counts n/measured/invalid/harness_failed, costs when available, and evidence references.

For the mock, the normal flow returns an example identifier such as sub-20261101T000000Z-1234abcd. The identifier is a receipt handle for that mock service and is not a benchmark result.

4. Read the evidence

Keep the submission_id, the uploaded archive digest, the tree digest, the scenario name and the permanent receipt together. Do not compare a pool score with another pool’s score or treat an absent value as zero.

The service is currently internal and invite-only. Do not publish a submission URL or expose the loopback service publicly.