sci story skill

Scientific papers in ML and CV often fail not because of the English, but because of weak logic, misaligned evidence, and indefensible claims. This skill transforms your research from a collection of experiments into a rigorous, reviewer-defensible narrative by applying a three-layer engineering system. It focuses on the underlying scientific argument rather than mere grammar correction.

What it does

• Corpus mining extracts underlying scientific argument patterns from best-paper award winners rather than just summarizing them.
• Research architecture builds a coherent chain from research gap to central insight, hypothesis, and evidence map.
• Claim-evidence auditing flags overclaiming by checking if claim strength exceeds evidence strength across direct and ablation results.
• Reviewer simulation generates critical attacks from three distinct personas to identify rejection risks before submission.
• Technical English editing refines stable manuscripts for precision, concision, and logical flow after the scientific logic is locked.

How it works

1. Analyze your idea or draft using the "Story Lock" mechanism to ensure the narrative remains consistent throughout the paper.
2. Execute specific modes like Experiment Planning to design RQs that satisfy skeptical reviewers or Reviewer Attack to find vulnerabilities.
3. Refine the manuscript based on a prioritized revision plan that fixes scientific validity before addressing language polish.

Frameworks & tools

Optimized for top-tier venues including NeurIPS, ICLR, ICML, CVPR, and ICCV. Expertise covers ML, Computer Vision, 3D Vision, and Generative AI domains.

Why this beats prompting it yourself

Generic prompts often produce "academic-sounding" fluff that obscures the actual contribution. This skill enforces a "Claim Strength ≤ Evidence Strength" rule and uses a formal Story Lock to prevent post-hoc storytelling, ensuring your paper survives rigorous peer review.

Use cases

• Drafting a high-impact introduction that clearly defines a research gap and central insight.
• Auditing a finished manuscript for unfair baseline comparisons or missing ablation experiments.
• Simulating a Rebuttal phase by generating and answering potential reviewer attacks.
• Restructuring a "result dump" into a logical scientific argument.

Known limitations

Does not fabricate data, citations, or results. Requires a stable research idea or existing results to provide architectural value. English polishing is only performed on scientifically stable drafts.