matrixCorr 0.13.0

New features

robust_ccc() supports the package’s standard matrix, sparse, and edge-list outputs; error, complete, and pairwise missing-data modes; and the shared print(), summary(), plot(), estimate(), coef(), tidy(), ci(), and confint() interfaces.
Optional paired percentile-bootstrap confidence intervals are available via ci = TRUE, with bootstrap success counts retained in result diagnostics.

Native pair coordinates are authoritative. Duplicate-named edges carry a row-keyed corr_pair_pos map.
Summary metadata now keeps the output filter threshold separate from latent correlation cutpoints.
Seed offsets use overflow-safe arithmetic that preserves the previously valid mapping and RNG streams.
Only a Pearson summary without stored inference adds the normal approximation, and only for pairs with n_complete > 3. Rank, robust, and other estimators do not gain a Fisher column from that fallback.
Summary and tidy append stored adjusted p-values, rejection decisions, and skipped-correlation counts and proportions when those matrices are present. Unadjusted p-values stay as stored.
Unnamed matrix edge accessors use source coordinates rather than the edge frame’s column names. Stored edge values and the corr_dimnames attribute are unchanged.
Empty edge and dense accessor tables no longer fail when a scalar bootstrap count is attached. The count is repeated once per row, including zero rows. confint() on an empty edge result reports that no confidence limits are stored. Thin edge lists avoid allocating the estimate matrix. Denser edge lists reuse the existing dense path when that cost is preferable. Stored inference payloads retain their representation.