                        Changes in version 0.3.0                        

  - Added a self-contained workflow vignette and runnable examples for
    the result accessors and model-diagnostic methods.
  - Added model_criteria() to extract all model-selection criteria by
    cluster count, with selected = TRUE to return only the selected
    model.
  - Added C_criterion() and RDC() for the selected model's criteria.
  - Added a logLik() method returning the selected Poisson model's
    likelihood, effective parameter count, and number of observations
    without refitting.
  - AIC() now respects k; AIC() and BIC() support multiple fitted
    objects through the standard R methods. Previously additional
    arguments were ignored.
  - These extraction methods also work with complete saved 0.2.0
    objects. The cluster search, model selection, and Monte Carlo
    algorithms are unchanged.

                 Changes in version 0.2.0 (2026-09-14)                  

Improvements

  - Improved reproducibility of Monte Carlo results under parallel
    computation by using doRNG (%dorng%), which provides independent
    random streams across workers. A new seed argument seeds sequential
    runs via set.seed() and parallel runs via registerDoRNG(). Using
    set.seed() alone is not sufficient when cores > 1.
  - Added accessors nclusters(), pvalue(), clusters(), and
    get_setting().
  - Added S3 methods as.data.frame(), coef(), nobs(), AIC(), and BIC()
    for multiflexscan objects so that results can be extracted without
    relying on the internal list structure.

Bug fixes

  - When nb is supplied as a list, areas whose neighbor indices fall
    outside [1, nrow(coordinates)] are skipped when building the
    adjacency matrix, instead of stopping with an error. Such areas are
    treated as adjacent to no other areas.
  - Null Monte Carlo replications now use the same simulated counts for
    both candidate search (runFleXScan) and the subsequent GLM / RDC
    steps. Previously the GLM always used the observed counts.
  - Empty candidate sets no longer error in the cluster p-value loop
    (seq_along instead of 1:length). Null replications with no
    candidates contribute -Inf as the maximum scan statistic (instead of
    NA).
  - When several values of K attain the same maximum RDC, nclust is the
    smallest such K (which.max), yielding a single integer.

                 Changes in version 0.1.0 (2026-07-24)                  

Initial release

  - Added multiflexscan() for detecting multiple spatial disease
    clusters using the information criterion and scan statistic approach
    of Takahashi and Shimadzu (2020), built on rflexscan::runFleXScan().
  - Supports flexible and circular scan statistics, original and
    restricted likelihood ratio statistics, and hot-, cold-, and
    both-type cluster scanning.
  - Returns candidate clusters, relative difference criterion (RDC)
    values, selected number of clusters, and Monte Carlo p-values for
    the overall test and individual clusters.
  - Added S3 methods print(), summary(), and plot() for multiflexscan
    objects.
  - Added choropleth() to map selected clusters on an sf object.
  - Parallel Monte Carlo replications via foreach and doSNOW.
