This door is for people who want to clone Zeta Lab, run their own agents against a bounded question and return the result as a pull request.
The repository welcomes that work. It does not accept a model's confidence as evidence. The authoring agent generates and attacks candidates; a non-model oracle or an independent checker decides what survived.
Read CONTRIBUTING.md, create a new hunts/<short-name>/ directory, and write MISSION.md before running the hunt. Record every run in RUNS.md, including the ones that found nothing, and put the bounded conclusion in RESULTS.md.
The first command after those files exist is:
.venv/bin/python scripts/71_contribution_check.py hunts/<short-name>That checks the contribution structure, the HuntSpec and run manifests, the case-log entry and the standing hunt discipline. It does not prove that the mathematics is right. The pull request must name the oracle, control or proof that carries that burden.