Changes in version 3.0.0 (2026-11-21) First CRAN release. - agreement() estimates Conger's (Cohen's for two raters), Fleiss', and Brennan-Prediger's agreement coefficients from a subjects-by-raters matrix with missing ratings. Estimators: inverse-probability weighting ("ipw", the default, valid under MCAR), categorical FIML ("cat_fiml", MAR), and for quadratic weights a pairwise moment estimator ("pairwise") and robust normal-theory FIML ("nt_fiml"), including a compound-symmetry variant for planned-incomplete designs ("nt_fiml_cs"). - "nt_fiml" falls back to "pairwise" by default (fallback) when its fit fails, for example when the fitted covariance is singular or nearly so because two raters agree perfectly; the fit records the reason. On complete data the two estimators coincide. - agreement_counts() handles counts-format data (subjects by categories), with the Fleiss-Cuzick moment estimator ("pairwise"), categorical FIML, and compound-symmetry normal FIML; ratings_to_counts() converts rater-identified data to that format. - Ratings may be numeric, text, or factors; categories declares the full rating scale, including categories nobody used. - Every fit carries influence-function standard errors. confint() gives Wald intervals (optionally Fisher-transformed) and exact jackknife intervals on complete data; agreement_test() tests equality of a coefficient across independent or paired fits. Categorical FIML fits also carry two diagnostics, null_frac and se_cutoff_ratio. - Multirater g-wise kernels (g > 2) and vector-valued ratings are supported. Versions 1.0.0 to 2.0.0 were released on GitHub only; their history is in the repository.