InterconectaR

• New sample_size_EGA_montecarlo(), with sample_size_EGA_control(), make_likert_thresholds(), plot_sample_size_EGA_montecarlo(), plot_power_curve_ci(), plot_failure_decomposition(), show_recommendation() and show_plots(): a priori sample size for EGA through Monte Carlo simulation.

Bug fixes

• NetPowerLab ran powerly with an increasing curve for specificity, which falls as N grows. It now reports a recommendation only when powerly converged and the value is not at the top of the range, and the script and paragraph follow the network and settings actually simulated.
• sample_size_EGA_control() merged nested lists: control = list(targets = list(ari = .80)) silently kept the default edge_correlation target. Entries now replace the defaults.
• sample_size_EGA_montecarlo(): items that EGA leaves without a community no longer inflate the ARI; bootstrap resamples that never reach the target are treated as censored instead of dropped, so the upper CI bound is no longer understated; the bootstrap CI is reproducible with parallel = TRUE; a non positive definite partial network stops with a message that gives the admissible edge weight; the sensitivity sweep skips the edge-threshold runs when no target depends on them and compares every row with the main run.
• optimize_leiden_resolution() treated untested resolutions (after an early stop) as having TEFI = 0, could choose one of them as the best, and misaligned the models with their rows in comparison when a resolution failed.
• plot_lvm() labelled residual correlations as residual partial correlations, and did not restore the graphical parameters.
• plot_latent_network_diagram() called getmatrix() without importing it, so psychonetrics models failed unless psychonetrics was attached.
• compute_netScores() wrote to the global environment when the stability step failed.
• convert_EGA_to_df() returned NA as the algorithm for EGAnet 2.x results.
• mgm_error_metrics() passed levels instead of level to mgm::mgm().

CRAN compliance

• No function assigns into the global environment: the print methods for silent_gg and silent_plot are now registered by the package.
• Nothing is written to disk by default: plot_latent_network_diagram() requires output_path.
• Functions report progress with message() instead of cat(), and compute_netScores() only asks for dimension names in interactive sessions.
• No fixed seeds inside functions: boot_and_evaluate(), refine_items_by_stability() and compute_netScores() default to seed = NULL; the user's future plan is restored on exit; parallel defaults use at most 2 cores.
• Every exported function documents its return value and has a runnable example.
• reshape2 and stringr are no longer needed; viridis moved to Suggests; psychonetrics added to Suggests. Requires R >= 4.1.0.