Strengthen IEEE Experiment Claims with a Routed Evidence Audit

Where this fits

You are finalizing an IEEE paper. The reviewers ask: "Where is the comparison against the standard baseline?" or "Your ablation only removes one component—what about the others?" You need a systematic way to check that every claim in your paper is backed by the right experiment, not just a generic checklist. The ieee-experiment skill turns that into a routed, evidence-focused audit that you can run before submission or during revision.

Why agents benefit

• Routes to the right evidence type: The skill reads a manifest and loads only the relevant fragments based on your task type (classification, control, hardware, etc.) and evidence axis (baseline, ablation, robustness, complexity). This avoids generic advice and produces targeted checks.
• Detects failure modes early: It explicitly flags missing traditional baselines, unfair comparisons, weak ablations, and overclaims—so you can fix gaps before reviewers do.
• Builds a claim-to-evidence matrix: The skill helps you map each claim in your paper to the exact experiment that supports it, making the audit output directly usable in your response letter.
• Handles reviewer concerns: When you receive feedback about insufficient experiments, the skill routes you to the appropriate robustness, statistical, or real-world test fragments to strengthen your submission.
• Works as an agentic tool: You can integrate it into a larger agent workflow that reads your manuscript, extracts claims, and runs the audit automatically, saving hours of manual cross-checking.

Practical scenarios

Pre-submission audit for a computer vision paper

You have a new detection model. The skill detects that your task_type is detection and evidence_type is baseline. It loads the baseline comparison fragment and asks: "Have you compared against at least three standard detectors from the last two years?" You realize you only used one—so you add the missing runs before submitting.

Responding to reviewer concerns about ablations

A reviewer says your ablation is incomplete. The skill's failure_mode routing points you to the weak-ablation fragment, which suggests a systematic component removal study. You generate a table showing each module's contribution, and your rebuttal now has concrete evidence.

Checking fairness and robustness in a communications system

Your paper proposes a new signal-processing algorithm. The skill's evidence_type includes fairness and robustness. It loads the relevant checks, such as testing under varying channel conditions and ensuring your metric is not biased toward a specific SNR range. You add those experiments, making your claims more defensible.

Add it to your agent workflow

Install the skill via npx, then let your agent read your manuscript and run the routing protocol:

npx skills add CloudWave818/ieee-skills --skill ieee-experiment

Example usage in an agent prompt: "Run the ieee-experiment skill on my draft. Identify missing baselines and weak ablations. Output a claim-evidence matrix." The skill will read manifest.yaml, load the always-load files, and route to the relevant fragments based on your paper's axes.

Compare before adopting

Before committing, compare this skill with other validation tools: check the GitHub repo's freshness (last commit date), star count (currently 259), and whether the author maintains it actively. Look for alternatives that might be more specific to your field (e.g., ML-specific validation checklists). Also verify that the skill's routing protocol matches your workflow—if you need a simpler checklist, this might be overkill. Finally, test it on a small section of your paper to see if the output is actionable.

Why it is worth tracking

The skill addresses a real pain point: turning vague reviewer demands into concrete experimental tasks. Its community momentum is visible in the star count and the author's responsiveness. Track it if you regularly write IEEE submissions or build agents that assist with research validation. Re-evaluate after each major update to see if new axes (like reproducibility or statistical tests) have been added. For more context, see the skill page at OpenAgentSkill.