Every observational study faces unmeasured confounding. Sensitivity analysis asks: how large would the bias need to be to change the sign or significance of the estimate?

Robustness means trying alternative specifications — different covariate sets, estimators, sample restrictions — and checking whether conclusions hold.

What to report

Primary estimate with confidence interval.
Balance or overlap diagnostics.
At least one alternative estimator.
Refutation or placebo results when available.
Plain-language limits for decision makers.