causal-heterogeneity-policy-learning

name	causal-heterogeneity-policy-learning
description	Plan CATE, causal forests, meta-learners, policy learning, conformal causal inference, fairness audits, subgroup stability, and heterogeneous-effect reporting. Use when causal effects vary across people, places, time, or policy rules.
user-invocable	true
license	Apache-2.0

Causal Heterogeneity and Policy Learning

Source adaptation: BioRouter-original synthesis inspired by Auto-Empirical Research Skills capability clusters, rewritten under Apache-2.0. This is native BioRouter guidance and does not copy upstream text.

When to Use

Use this skill when the user asks about CATE, causal forests, meta-learners, policy learning, heterogeneous treatment effects, fairness audits, or conformal causal inference. If the task is biomedical, clinical, or omics-specific, combine this with the appropriate BioSkills category and extension instead of treating the methods in isolation.

Intake

Collect:

• Research question, estimand, and audience.
• Dataset/source, file formats, licenses, access constraints, and identifiers.
• Unit of analysis, time period, sample design, and outcome/treatment definitions.
• Software stack and expected artifact: code, table, figure, memo, manuscript, appendix, or review response.
• Whether the work is exploratory, confirmatory, replication, submission, or referee-response.

Method Gates

Before writing conclusions:

1. State the design and estimand before the model.
2. Preserve provenance for every dataset, variable, citation, and transformation.
3. Match inference to the sampling/assignment structure.
4. Identify weights, clusters, panels, repeated measures, or dependence before estimating.
5. Check robustness against the actual design threat, not decorative alternative models.
6. Keep code/table/figure outputs reproducible from a clean entry point.

BioRouter Routing

• Use empirical-research-router for full workflow staging.
• Use causal-identification-gates before causal language.
• Use econometrics-toolkit, statistical-modeling-bayes, or financial-accounting-econometrics for adjacent method detail.
• Use citation-temporal-integrity, claim-evidence-integrity, and replication-package-audit for publication-facing outputs.

Output Pattern

Return a compact plan with:

• Primary route and why.
• Required data/design evidence.
• Minimal next command or artifact.
• Failure modes.
• Verification checklist before final claims.