Validation and method support

This Book documents package 32.2.3. The release decision remains NOT_RELEASE_READY. A readable Book, a successful installation and a passed synthetic example are separate from statistical calibration or release validation.

Runtime support

The executable method matrix depends on exposure type, outcome, analysis mode and target estimand. A stable runtime label means an implemented path in eligible settings; it does not certify all assumptions or validate every combination.

Download the complete support snapshot. It preserves support, assumptions, skip reasons and historical evidence fields. The snapshot was generated on 2026-10-29 without fitting an estimator. Its documentation date is not a new benchmark date.

Method family	Important limit
GLM	Regression summaries and survival targets need their own scales and assumptions
GComp	Eligible survival contrasts are RMST or survival-probability differences
AIPW / TMLE	Risk/mean differences; policy and survival paths have separate limits
IPW / PSM	Binary-exposure survival HR is limited; primary continuous-exposure survival is unsupported
Causal Forest	Exploratory heterogeneity; no individually calibrated treatment rule
GPS	Continuous exposure only; survival GPS is not implemented

Convenience wrappers do not expose every advanced setting. Some dedicated adapters, including longitudinal and frozen OS mediation paths, are outside the formal matrix and have separate contracts.

What documentation checks establish

Book checks cover rendering, local links and anchors, resource integrity, English content, migration coverage and selected synthetic examples. They do not update the row-wise Codex acceptance fields in the method snapshot, rerun patient analyses or establish a biological mechanism.

For any scientific result, retain its exact code version, inputs, requested family, effective settings, complete numeric output and diagnostic findings. Historical evidence stays labeled with its original scope and version.