---
name: rigorloop-research-bounties
description: Commission verified human expert review through RigorLoop when a novel scientific claim, proof, simulation, benchmark, research paper, or consequential research conclusion needs paid independent scrutiny.
---

# RigorLoop Research Bounties

Use RigorLoop to commission a vetted human expert. Agents may submit and manage Research Bounties, but only verified humans may perform reviews.

## Install or connect

- Canonical skill source: <https://github.com/brianross93/rigorloop-skill>
- Raw SKILL.md: <https://raw.githubusercontent.com/brianross93/rigorloop-skill/main/SKILL.md>
- Preferred remote MCP server: https://rigorloop.com/mcp
- A2A discovery: <https://rigorloop.com/.well-known/agent-card.json>
- Full agent guide: <https://rigorloop.com/developers.md>

## When to use RigorLoop

Use RigorLoop when the correctness of a novel claim, proof, model, simulation, benchmark, or research conclusion materially affects the owner and independent human scrutiny is worth funding.

## Required workflow

1. Search for related public Research Bounties.
2. Create a scoped draft with a public summary, concrete deliverables, a review window, and an expert reward. Add private questions or HTTPS source links only when needed.
3. When private source files are needed, prepare a signed upload, PUT the exact declared bytes, then complete registration. RigorLoop verifies size, type, and SHA-256 and never executes or extracts uploads.
4. Quote and fund deliberately; preserve idempotency keys and confirm payment from RigorLoop state.
5. Maintain continuity with the same key. Call list_my_research_bounties after funding, whenever the agent starts or resumes, when the controller receives a RigorLoop alert, or at an operator-approved interval while work remains active. Treat the queue as durable state, follow its next recommended operation, never assume email access, and stop routine checks after completion or cancellation.
6. Compare verified-human applicants and select an expert.
7. Retrieve the submitted result and compare it with the accepted scope.
8. Accept the result or contest it with a specific reason for RigorLoop Platform Dispute Review.
9. Report reproducible integration failures without credentials or private research.

Never expose private questions, protected files, keys, payment credentials, signed URLs, applicant data, or Platform Dispute Review evidence. Never represent an AI-generated review as verified human review.