What problem does it solve? Research experiments often sprawl across untracked branches, unrecorded expectations, and unverifiable results. This Skill enforces disciplined experiment lifecycle management: every run targets a registered hypothesis, every artifact is accounted for, and every conclusion is traceable to evidence. ## Core Features & Use Cases - Branch and Worktree Lifecycle: Creates and manages exp/NN-slug branches and isolated worktrees with explicit user authorization, keeping exploratory code separate from the base branch. - Run Reports with Frozen Expectations: Each run produces an ENN-experiment-report.md with pre-registered baselines, expected ranges, mechanism predictions, and convergence records written before results are seen. - Artifact Hygiene and Git Policy: Maintains a whitelist of what enters Git (small tables, configs, final figures) versus external storage with checksums, preventing repository bloat. - Closure and Promotion: Synthesizes falsified or validated directions into an illustrated REPORT.md, merging only validated content through a reviewed promote/NN-slug branch. - Use Case: A researcher running a series of model ablation experiments uses this Skill to register hypotheses before each run, track which results narrowed the hypothesis space, and produce a final evidence-linked report when the direction closes. ## Quick Start Ask the assistant to open a new experiment branch for the approved proposal, register the expected results, and produce a run report after execution.
