What problem does it solve? Researchers with completed ML training pipelines often struggle to convert their engineering work into publishable science. This Skill audits a training pipeline project (datasets, pipeline steps, training logs) and produces a structured research output: identified research gaps, falsifiable scientific hypotheses, and an IMRaD-structured paper plan with literature scope and timeline. ## Core Features & Use Cases - Pipeline-to-Methodology Decomposition: Scans project files and decomposes pipeline steps (e.g., pupil localization, region growing, SAM segmentation) into a formal methodology description. - Research Gap & Hypothesis Generation: Identifies at least 3 research gaps located in literature contradictions or method gaps, and generates at least 3 testable hypotheses with test conditions and falsification paths. - Paper Planning with Data Honesty Gate: Outputs a paper_plan.json with IMRaD structure, 15-30 reference literature scope, target journal tier, and timeline, while rejecting any metric (e.g., Val Dice) that cannot be traced to training logs. - Use Case: Given a K230 eye-tracking project with 901 frames, a 7-step CV pipeline, and MobileNetV2 training logs, the Skill produces 5 research gaps, 5 falsifiable hypotheses, and 2 SCI paper plans (Q1 methodology + Q2 empirical analysis). ## Quick Start Audit my training pipeline project with its datasets, pipeline steps, and training logs, then generate research gaps, testable hypotheses, and an IMRaD paper plan.
