What problem does it solve? ML experimentation often loses track of why experiments were run, what was tried, and what the results mean. This Skill enforces a disciplined iteration loop where every experiment script is preceded by an approved design note and every finished run is recorded in a durable journal, keeping the project history reproducible and auditable. ## Core Features & Use Cases - Design-note gating: Blocks creation of experiments/NN_*.py until a matching journal/NN_*.md design note is drafted and explicitly approved by the user. - Mode-driven workflow: Routes each session to bootstrap, propose, record-outcome, compare, goal-pivot, abandoned, or re-run modes based on journal state and user signals. - Sourcing strategies: Sources the next experiment from skore report audits, user-provided articles or issues, synthesized candidate ideas, or a numbered Backlog. - Use Case: In a cross-validated classification project, the user asks "what's next"; the Skill reads journal/JOURNAL.md, presents the sourcing menu, drafts a design note, and only after approval dispatches the build, evaluate, and test pipeline skills. ## Quick Start Ask the assistant to propose the next ML experiment based on the current journal and backlog in this workspace.
