What problem does it solve? Iterating on research experiments often leads to lost results, unverifiable claims, and unsafe conclusions such as data leakage or metric inflation. This Skill structures each experiment iteration with recorded state, dual verification gates, and resumable tracking so every run is auditable and reproducible. ## Core Features & Use Cases - Structured Iteration Recording: Logs every experiment iteration to autoresearch-results.tsv and maintains resumable state in autoresearch-state.json. - Dual-Gate Verification: Applies a verify gate (did the metric improve?) and a guard gate (no leakage, config drift, phantom results, or claim inflation) before promoting results. - Local-to-Remote Workflow: Coordinates local_mac smoke tests with formal runs on remote_desktop_4060, generating experiment reports with baseline comparisons. - Use Case: A graduate student testing a new model variant runs a local smoke test, launches the formal GPU run remotely, then records the iteration with verify/guard status and a baseline comparison report before deciding whether to update the thesis claims. ## Quick Start Use the research-autoresearch-loop skill to plan and record my next experiment iteration against claim CLM-001 with a baseline comparison.
