What problem does it solve? Running the ChicagoHAI HypoGeniC package for LLM-assisted hypothesis generation involves unbudgeted API calls, unpinned datasets, split leakage, and untrusted prompt text. This Skill adds a local, deterministic review layer that validates configs, audits datasets, and bounds cost before any model is called. ## Core Features & Use Cases - Config and policy validation: Validate official HypoGeniC task YAML and a local run policy covering provider, credentials by name, token/cost caps, and logging redaction. - Dataset auditing: Verify pinned HypoBench dataset checksums, schemas, and cross-split duplicate or identity leakage without printing raw text. - Cost planning and model-free evaluation: Compute a conservative token/cost upper bound and score saved predictions with accuracy, coverage, and macro-F1 without importing provider SDKs. - Use Case: Before running HypoGeniC on the deceptive-reviews HypoBench dataset, audit the pinned files, confirm the run policy, and get a bounded cost plan that flags the three known cross-split duplicate groups. ## Quick Start Ask the assistant to validate the example run policy and audit the pinned HypoBench dataset manifest before planning any HypoGeniC generation run.
