Changes in version 0.1.0 (2026-08-28)                  

  - Removed the package-wide two-worker ceiling. cores still defaults to
    one, while explicit requests are now reduced only to the number of
    subjects.
  - Relayed warning conditions raised by PSOCK subject-fit workers
    through the main R process during parallel execution, without
    changing fitting or selection behavior.

                 Changes in version 0.0.12 (2026-08-24)                 

  - Renamed the package and its complete public identity from ctsgimme
    to ctgimme. The primary functions are now ctgimme() and
    ctgimme_demo(); internal helper prefixes, returned-object
    attributes, help aliases, tests, citation metadata, and source
    archives use the same name. The old function names are not exported.
  - Added the public verbose argument. verbose = FALSE suppresses
    package progress messages and captures OpenMx optimizer/progress
    output, while warnings, errors, returned results, and saved
    artifacts remain available. Removed the unconditional console print
    of subgroup membership.
  - Replaced the non-executable wrapped example with a self-contained,
    timed toy fit that writes below the session temporary directory and
    always cleans its output.
  - Capped package-managed parallel execution at two workers in all
    contexts, strengthened documented input and control validation, and
    guaranteed an extractable membership object when subgroup detection
    falls back to one group.
  - Added a complete, ID-aligned ctgimme.membership attribute to every
    successful PAM or legacy clustering return; PAM results also retain
    their direct membership element.
  - Propagated the master session's library paths to PSOCK workers and
    exported the refactored initial-covariance helpers, preventing
    two-worker fits from failing after an R/library upgrade or during
    subject-model construction.
  - Hardened subject-artifact handling for portable filesystems:
    identifiers must be unique without regard to case, generated
    prefixes/suffixes are decoded literally, wildcard characters are
    never expanded during cleanup, and failed RDS fallback copies now
    raise an error.
  - Retained the 0.0.11 multisubject subgroup-model implementation: one
    shared parameter vector per subgroup, independently initialized
    subject likelihood blocks, requested exp(A * delta) plots, and one
    Subgroup_<g>Model.RDS artifact.

                       Changes in version 0.0.11                        

  - Converted the historical single-file repository into an installable
    R package with generated help, tests, citation metadata, and
    standard GitHub installation. The 0.0.6 monolithic script and
    generated example outputs remain recoverable from the
    pre-package-0.0.6 Git tag.
  - Replaced the previous concatenated optional subgroup fit with one
    shared-parameter multisubject likelihood. Each subject receives an
    independent continuous-time state-space filter while the subgroup
    has one shared A/Q/R specification whose free entries are jointly
    optimized. Subject times are rebased locally to preserve elapsed
    intervals without making the likelihood depend on arbitrary start
    offsets.
  - Removed the chained estimator and its public selector and
    concatenation controls: subgroup.model.method, subgroup.time.mode,
    measurement.schedule, cycle.interval, insert.na.rows, and
    subject.gap. This intentionally changes calls that supplied
    those 0.0.10 arguments, including positional calls after
    time.intervals.
  - Retained time.intervals for producing exp(A * delta) discrete-time
    transition plots. Every successfully fitted subgroup also writes its
    parameter plot and Subgroup_<g>Model.RDS joint fitted model.
  - Expanded the margins on subgroup parameter and transition graphs so
    outer nodes, self-loops, titles, and edge labels remain visible in
    saved PNGs.
  - Added an optional internal P0.values override and factored the
    existing empirical initial-covariance calculation into reusable
    helpers.

                       Changes in version 0.0.10                        

  - Reused one initialized PSOCK worker pool across group, subgroup, and
    individual fitting batches, avoiding repeated R-process and package
    startup costs during a single analysis.
  - Ensured package-owned worker pools are stopped after successful
    completion and during error unwinding.

                        Changes in version 0.0.9                        

  - Added ME.free and PE.free options for estimating selected diagonal
    measurement-error and process-noise variances. Both default to
    FALSE, and off-diagonal noise covariances remain fixed to zero.
  - Prevented OpenMx's interactive imxReportProgress callback from being
    used by internal mxTryHard() fits. This applies to group, subgroup,
    and individual estimation paths.
  - Fixed subgroup parameter plotting by passing the drift edge labels
    to qgraph as a square matrix.
  - Expanded documentation and tests for the diagonal-noise options and
    the RStudio/OpenMx compatibility fixes.