What problem does it solve? Empirical research sessions often lose track of estimands, identification assumptions, diagnostics, and negative results, making findings impossible to audit or hand off. This Skill enforces a bounded, documented workflow so every empirical task produces a spec, an evidence-cited result, and a clear stopping point. ## Core Features & Use Cases - Empirical Spec Authoring: Writes a ds-lite.empirical-spec.v1 defining the research question, estimand, population, sample rules, variables, identification strategy, assumptions, diagnostics, and robustness plan before any analysis runs. - Capability-Aware Execution: Inspects which runtimes are actually observed (Python as reproducibility reference, StatsPAI as comparison, R and Stata as not-observed until discovered) and never installs runtimes or downloads data without approval. - Auditable Result Handoff: Produces a ds-lite.empirical-result.v1 citing a Core Evidence Pack with project-relative commands and artifacts, preserving confounding warnings, failed pretrend tests, missingness, and null or negative results. - Use Case: A researcher running a difference-in-differences analysis uses this Skill to lock the estimand and pretrend diagnostics up front, execute one bounded work unit, and hand off a result envelope that honestly reports a failed parallel-trends test instead of hiding it. ## Quick Start Ask the assistant to use ds-lite-empirical to define and execute one bounded empirical work unit with explicit identification, diagnostics, robustness checks, and evidence references.
