What problem does it solve? Turning a refined research proposal into a concrete, paper-defensible experiment plan is hard: researchers often produce bloated benchmark wishlists instead of a focused claim-to-evidence roadmap. This Skill converts a method idea into a structured plan that maps every experiment to a paper claim, orders runs by milestone, and separates must-run evidence from nice-to-have extras. ## Core Features & Use Cases - Claim Map Construction: Freezes primary and supporting claims, anti-claims to rule out, and the minimum convincing evidence a strong reviewer would need. - Experiment Block Specification: Defines each block with dataset, baselines, metrics, setup details, success criteria, failure interpretation, and target table or figure, covering novelty isolation, simplicity checks, and frontier-model necessity checks. - Execution Ordering and Tracking: Produces a milestone-based run order with compute cost, decision gates, and risks, writing EXPERIMENT_PLAN.md and EXPERIMENT_TRACKER.md under refine-logs/. - Use Case: After refining a VLM-based compression method, ask for an experiment plan and receive a compact roadmap with three must-run blocks, a 3-seed ablation matrix, GPU-hour budget, and the first three runs to launch. ## Quick Start Ask the AI to create a detailed experiment plan for your refined research proposal, including ablations, baselines, run order, and compute budget.
