What problem does it solve? Research experiments often proceed without a trustworthy reference point, making it impossible to tell whether a new method actually improves on the status quo. This Skill freezes the current baseline, evaluation protocol, primary metric, and failure profile before any method changes begin. ## Core Features & Use Cases - Baseline Inspection: Reads the active implementation, evaluation code, configs, data splits, and raw results to identify the actual method under test rather than relying on stale summaries. - Baseline Artifact Generation: Creates research/BASELINE.md capturing run id, commit, config, seeds, budget, aggregate metrics, comparators, target delta, and localized failure profiles. - Research Contract Setup: Generates research/RESEARCH_CONTRACT.md from a preset template, persisting the entrance mode, deliverable, and fixed evaluation protocol. - Use Case: Before modifying a model training pipeline, run this Skill to record the official baseline's measured performance and failure modes, so every subsequent experiment is compared against a fixed, reproducible reference. ## Quick Start Assess the current implementation, evaluation protocol, and baseline metrics, then write the baseline artifact before any research changes.
