What problem does it solve? After experiments finish, it is hard to judge objectively whether the numbers actually support the intended research claim. This Skill removes post-hoc rationalization by delegating the verdict to an external LLM reviewer and routing the project to the right next step. ## Core Features & Use Cases - Result Collection: Gathers metrics from W&B runs, EXPERIMENT_LOG.md, EXPERIMENT_TRACKER.md, training logs, and the research contract. - External LLM Judgment: Sends results to a reviewer via the llm-chat MCP and parses a strict pass/fail verdict with evidence gaps and suggested claim revisions. - Integrity-Aware Routing: Checks EXPERIMENT_AUDIT.json for integrity warnings, then routes to claim narrowing, supplementary experiments, pivoting, ablation planning, or paper writing. - Use Case: After a mechanistic interpretability experiment completes, run this gate before writing the paper to confirm the claim is supported or to identify which supplementary experiments are still needed. ## Quick Start Run the result-to-claim gate on my latest W&B run to judge whether the results support the intended claim.
