What problem does it solve?
Experiments often produce numbers without a clear, evidence-grounded decision about which claims those numbers actually support, which undermines paper writing, reviews, and follow-on work.
Core Features & Use Cases
- Claim-support judgment via Codex: Routes experiment results through a structured evaluation that determines whether the intended claim is supported, partially supported, or not supported.
- Verdict-based research workflow: Automatically drives next actions like pivoting, designing supplementary experiments, or marking confirmation-ready outcomes.
- Optional research-wiki updates: When the wiki exists, records verdicts and updates claim/idea status and edges to keep the knowledge graph consistent.
Quick Start
Use the result-to-claim gate after experiments finish by providing the experiment description or the relevant W&B run identifier so the skill can compile results, ask Codex for a verdict, and route to the next research step.