Changes in version 0.4.0                        

Breaking changes

The kofn() constructor now takes a component argument (a dfr_dist
prototype from flexhaz) instead of a stringy family argument:

# Before (0.3.x):
kofn(k = 2, m = 3, family = "exponential")
kofn(k = 2, m = 3, family = "weibull")

# After (0.4.0):
kofn(k = 2, m = 3, component = dfr_exponential())
kofn(k = 2, m = 3, component = dfr_weibull())

Rationale: object-based component specification composes with the rest
of the ecosystem (flexhaz, serieshaz, maskedhaz) and makes the
supported-family table a property of class dispatch rather than a
parallel lookup in match.arg. Adding future supported component families
is now a one-line change.

The same rename applies to compare_fisher_info():

# Before:
compare_fisher_info(rates = c(1, 2), family = "exponential", ...)

# After:
compare_fisher_info(rates = c(1, 2), component = dfr_exponential(), ...)

Internal helpers parse_params(), kofn_dgp(), and kofn_components() take
component rather than family. These were @keywords internal and not part
of the public API, but downstream code that touched them via ::: will
need the same rename.

New internal helpers

  - n_par_kofn(m, component): expected parameter count for a homogeneous
    kofn system, replacing inline switch(family, ...) logic.
  - kofn_subclass(component): maps a dfr_dist prototype to the
    corresponding kofn subclass (exp_kofn or wei_kofn).

Dependencies

  - Added flexhaz (>= 0.5.2) to Imports. The dfr_dist prototype
    constructors provide family-specific S3 subclasses (dfr_exponential,
    dfr_weibull) that kofn_subclass() dispatches on.

Bug fixes

  - compare_fisher_info(): corrected the Weibull complete-data
    (Scheme 2) Fisher information formula. The shape-shape entry had a
    spurious leading 1 (about 55 percent too large) and the shape-scale
    cross term was divided by alpha * beta instead of beta. Both errors
    vanish at shape 1 (the exponential boundary), which is why
    exponential results were unaffected. The analytic block now lives in
    an internal helper verified against numerically integrated expected
    information in the test suite. Scheme 0/1 results (numerical
    Hessians) were never affected; for Weibull components, Scheme 2
    determinants and the efficiency ratios involving Scheme 2 change.
  - Interval-censored rows in a data frame lacking the upper-bound
    column (t_upper by default) now fail with an informative error in
    all likelihood paths. Previously the exponential parallel fast path
    crashed with "missing value where TRUE/FALSE needed" and the
    general-k path silently returned -Inf.

                        Changes in version 0.3.1                        

Documentation

  - Vignettes restructured around the post-refactor package scope.
    Deleted outdated vignettes (censoring-framework, kofn-systems,
    general-kofn, symmetry-breaking) that documented removed topology /
    general-system APIs; added:
      - getting-started: quick tour of the kofn model, fit, and
        observation schemes.
      - dist-structure-integration: how kofn delegates DGP and topology
        to dist.structure, convention conversion, migration notes for
        v0.2.0 users.
  - Added _pkgdown.yml configuring navbar, article menus, and a
    reference index grouped by theme.
  - Retained and verified: exponential-parallel, observation-schemes,
    weibull-em, periodic-inspection, ecosystem.

                        Changes in version 0.3.0                        

Major refactor: dist.structure adoption

kofn now delegates the data-generating process (DGP), topology, and
system-level distribution generics to dist.structure. The package
focuses exclusively on inference for k-out-of-n systems: log-likelihood,
score, Hessian, fit, observation schemes (right/left/interval/periodic),
masked cause-of-failure, periodic-inspection (Scheme 1), and Fisher
information comparison.

Removed

  - coherent_system(), parallel_system(), series_system(),
    kofn_system(), bridge_system(), consecutive_k_system(): use the
    corresponding constructors in dist.structure (coherent_dist,
    parallel_dist, series_dist, kofn_dist, bridge_dist,
    consecutive_k_dist).
  - system_signature(), system_lifetime(), system_censoring(),
    critical_states(), min_paths(), min_cuts(), phi(),
    min_cuts_from_paths(), minimize_sets(): use dist.structure
    equivalents (registered on every dist_structure subclass).
  - loglik_system(), fit_system(), rdata_system(): replaced by the
    loglik/fit/rdata S3 methods on kofn model objects (for k-of-n) plus
    the dist generics on dist.structure::exp_kofn, wei_kofn,
    coherent_dist (for general topologies).
  - f_sys_general(), S_sys_general(): use algebraic.dist::density() and
    surv() on a dist.structure object.
  - make_dists(), exp_dist(), weibull_dist() (the lightweight
    $pdf/$cdf/$surv list interface): replaced by
    algebraic.dist::exponential() and weibull_dist() plus the
    algebraic.dist generics.

Changed

  - kofn() constructor no longer accepts a system = ... argument. kofn
    is now exclusively for k-of-n estimation. Users with non-k-of-n
    topologies should use dist.structure::coherent_dist() directly.
  - loglik.exp_kofn and loglik.wei_kofn general-k path now delegates
    per-observation contributions to dist.structure::exp_kofn(k_dist,
    par) and wei_kofn(k_dist, shapes, scales) via the algebraic.dist
    density, surv, and cdf generics. Parallel-fast-path (k=m) IE
    expansion is preserved as the closed-form path.
  - loglik_scheme1 system density now uses
    density(dist.structure::exp_kofn(...)) / wei_kofn(...) instead of
    the local f_sys_general engine.
  - rdata.exp_kofn, rdata.wei_kofn, rdata_masked, rdata_scheme1 use a
    kofn-internal specialized kofn_censoring(k, times) helper (the k-th
    order statistic) instead of the general system_censoring algorithm.
  - assumptions.exp_kofn and assumptions.wei_kofn no longer mention
    "general coherent system" since kofn is k-of-n only.

Internal

  - New file R/internal_topology.R with private helpers
    kofn_censoring(), kofn_dgp(), kofn_components(). Uses the k_dist = m
    - k_kofn + 1 conversion between kofn's :F convention (k_kofn =
    number of failures triggering system failure) and dist.structure's
    :G convention (k_dist = number of functioning components required).
  - Deleted R/coherent_system.R (599 lines), R/system_density.R (336
    lines), and R/dist.R (118 lines).
  - Package shrank from 4312 lines to roughly 2750 lines (~36%
    reduction) in the R/ directory.

Added

  - Imports: dist.structure, algebraic.dist.
  - ncomponents re-exported from dist.structure (was a local generic).

                        Changes in version 0.2.0                        

  - Generalized masking: loglik_masked() and rdata_masked() extend
    masked cause-of-failure from series (k=1) to arbitrary k-out-of-m
    systems.

  - Periodic inspection (Scheme 1): loglik_scheme1() and fit_scheme1()
    support general k via the critical-state density engine.

  - Performance: eagerly precomputed critical/functioning states in
    coherent_system() and vectorized f_sys_general() give ~70x speedup
    for Scheme 1 likelihood evaluation.

  - Migrated optimization to the compositional.mle package. The internal
    helper multistart_mle() was removed and replaced by a private
    solve_mle() wrapper around compositional.mle::lbfgsb() with a
    Nelder-Mead fallback in log-parameterization.

  - New vignettes: symmetry-breaking comparison, periodic inspection,
    observation schemes, general k-out-of-m.

                        Changes in version 0.1.0                        

  - Initial release with exponential and Weibull parallel system MLE.
  - EM algorithm for Weibull parallel systems.
  - Coherent system infrastructure via minimal path/cut sets.
  - Fisher information comparison across observation schemes.