ml-science-discipline

What problem does it solve?

This Skill helps you prevent invalid machine learning results by enforcing rigorous experimental design, leakage controls, and publication-standard reporting practices.

Core Features & Use Cases

• Reproducible, hypothesis-first experimentation: define falsifiable hypotheses, pre-register the primary metric and baselines, and report multi-seed variance instead of single lucky runs.
• Leakage-proof data splitting and evaluation: enforce sealed test-set usage, choose the correct split strategy (stratified, group, temporal), and audit preprocessing to ensure fit happens only on training data.
• Q1-ready evaluation and reporting artifacts: select appropriate metrics, run statistical comparisons with effect sizes, include calibration/uncertainty, and plan external validation against distribution shift.

Quick Start

Use ml-science-discipline when you are about to design or review an ML experiment intended for a paper, such as dataset splitting, evaluation planning, and reporting results you want to defend in peer review.