tda-statistical-modeling-toolkit

Classify the model as robustness, state the statistical question, reference the fitted sample's provenance block, and specify family, link, covariates, and clustering. Report effect sizes with uncertainty intervals and name exactly which paper claim the result supports.

It covers GLMs, OLS, logistic and probit regression, survival models, clustered models, bootstrap summaries, Bayesian robustness checks, simulation-based sensitivity checks, and power analysis. It explicitly excludes persistent homology computation and design-validity auditing.

No. A p-value alone is treated as an incomplete result. Every reported model must include effect sizes and uncertainty intervals, plus diagnostics and a provenance source before the result is usable.

Escalate when the estimand or eligibility rule is unclear, routing to panel-estimand-audit first. Also escalate when a result would become a headline paper claim, which requires full lane review before any prose is written.

Only under strict conditions. Causal language is forbidden unless a causal design with an explicit identification strategy and named assumptions has been specified and reviewed. Otherwise results must use associational language bounded by the model classification.