csdm

• Evaluate formulas, transformed cross-sectional-average variables, subsets, and missing-value policies consistently.
• Correct missing-value and leave-one-out cross-sectional averages.
• Use one eligible unit sample for mean-group coefficients and covariance, and require at least two eligible units.
• Correct fixed-weight mean-group covariance scaling and HC0--HC3 sandwich covariance calculations.
• Check structural identification after cross-sectional-average projection and report excluded units.
• Align CS-ARDL parameter components and covariance, and expose unit-level AR stability and long-run-ratio eligibility.
• Validate panel keys, retain original observation identity when sorting, and preserve named cross-sectional-average lag specifications.

Dependence diagnostics

• Remove time periods containing no estimated residuals before assessing CD sample balance, while retaining the selected policy for partially observed periods.
• Implement the paper-defined pooled-variance CDw statistic and the correlation-scale screening term used by CDw+ for balanced samples.
• Correct CD-star unit-specific residual scales and validate PCA rank.
• Make randomized diagnostics opt-in, preserve seeded RNG state, and use pairwise samples for the classical CD statistic.
• Validate clustered covariance inputs and make residual transformations explicit.

R interfaces

• Add standard extraction and update methods, original-row fitted and residual outputs, and stored-data model updates.
• Add tidy(), glance(), and augment() methods with inference and row-alignment checks.
• Document cross_sectional_avg() as a supported standalone data utility and distinguish it from averages configured by csdm_csa().
• Retain evaluated subset, time-spacing, missing-value, and pdata.frame time information when models are updated.
• Reject unsupported model, trend, cross-sectional-average, long-run, and covariance specifications instead of silently storing or ignoring them.

Deprecations

• Deprecate csdm_pooled() because pooled restrictions are not implemented.
• Deprecate get_residuals() in favor of the standard residuals() method.
• Deprecate prepare_cd_input() because its transformations are not used by cd_test() and can change the hypothesis represented by the residuals.
• Deprecate the exported cluster_vcov(), sandwich_vcov(), and pooled_vcov() utilities. Existing calls continue to work with a migration warning.

Documentation and maintenance

• Rewrite the README, introductory vignette, and pkgdown navigation around the implemented API. Correct the description of log_ngd, update the CDw/CDw+ explanations, and make the introductory examples runnable.
• Add reference fixtures, statistical validation scripts, platform checks, and Codecov reporting.
• Share unit-regression and sample-bookkeeping code across the MG, CCE, and DCCE engines, and remove obsolete internal helpers and imports.