Changes in version 0.1.6                        

  - Mixed-space hybrid auto-dispatch widened to 17 presets:
    ParticleSwarm and MothFlameOptimization join the list (since 0.1.6),
    and DifferentialEvolution(variant = "jde") is now allowed on mixed
    spaces (float blocks keep the preset's own generator, GA variation
    on discrete blocks; a space with no float block is all GA).
  - Engine: gen/pso, gen/de-jde and gen/mfo now own their state (PSO
    velocity/pbest, jDE per-individual F/CR, MFO flame memory) and are
    on the gen/compound self-contained allow-list. The companion
    components (replace/pso-commit, adapter/jde-commit,
    adapter/mfo-flame-update) are kept for spec compatibility; their
    state duties moved into the generators. Single-block preset runs are
    bit-identical to 0.1.5.
  - Still excluded from hybrid dispatch (structural): shade/lshade,
    cmaes (+ ipop), abc, sa, harmony_search, nelder_mead, hho.

                        Changes in version 0.1.5                        

  - Mixed-space hybrid auto-dispatch widened to 15 presets:
    GravitationalSearch and BatAlgorithm join the list (float blocks
    keep the preset's own generator, GA variation on discrete blocks; a
    space with no float block is all GA).
  - Engine (gen/compound): every block now owns a nested blackboard
    stored in the parent blackboard under cmp{i}/bb, so self-contained
    stateful generators keep per-block state. An allow-list
    (SELF_CONTAINED_COMPOUND_KINDS: gen/gsa, gen/ba) admits them; other
    stateful generators (PSO, CMA-ES, SHADE, ...) stay rejected.
    Existing stateless compound specs are bit-identical.

                        Changes in version 0.1.4                        

  - Mixed-space hybrid auto-dispatch widened to 13 presets: AntLion and
    EvolutionStrategy join the 0.1.3 list (float blocks keep the
    preset's own generator, GA variation on discrete blocks; a space
    with no float block is all GA). SimulatedAnnealing stays ineligible:
    its fixed population of 1 cannot host the GA discrete
    sub-generators.
  - Engine: the post-adapter genotype shape sweep now runs only for
    stages that have an adapter.
  - New R guard tests for .sz_hybrid_wrap_stages (malformed preset
    specs).
  - The ineligible-preset error now lists the real preset names
    (cuckoo_search, alo, es_mu_plus_lambda) instead of "cuckoo-search".

                        Changes in version 0.1.3                        

  - gen/compound now accepts any registered sub-generator that is
    eligible (stateless, pop-to-pop, no internal evaluation, no nesting)
    instead of a fixed GA-only table; ineligible kinds are rejected with
    the reason.
  - Mixed-space auto-dispatch for 11 presets: DifferentialEvolution
    (rand_1/best_1), GreyWolfOptimizer, WhaleOptimization,
    SineCosineAlgorithm, JAYA, GrasshopperOptimization, SalpSwarm,
    FireflyAlgorithm, FlowerPollination, TLBO, CuckooSearch. The result
    is a hybrid: the preset's own generator on float blocks, GA
    variation on binary/int/categorical/permutation blocks. A space with
    no float block gets zero preset-specific variation (all variation is
    GA).
  - New shape guards: the evaluator panics with "genotype shape
    mismatch:" on a mis-shaped genotype, and the engine re-checks
    genotype shape after the adapter step.
  - The multi-block engine error hint now mentions eligible non-GA
    generators inside gen/compound.
  - Presets that cannot ride inside gen/compound (stateful,
    one-offspring, internal-evaluation) raise an error naming why on a
    mixed space.

                        Changes in version 0.1.2                        

  - GeneticAlgorithm now auto-dispatches multi-block single-kind spaces
    (e.g. several Float blocks) through gen/compound; mixed-kind spaces
    still raise. Other flat presets run on a multi-block space now fail
    fast with an engine genotype-shape error instead of silently
    corrupting.

                 Changes in version 0.1.1 (2026-09-15)                  

  - Provenance and licensing documentation: corrected TSPLIB status
    note, added inst/COPYRIGHTS and Copyright field, rescoped
    LICENSE.note, added THIRD-PARTY-NOTICES; no functional changes.

                        Changes in version 0.1.0                        

  - Initial release. sezgi ships 29 built-in algorithm classes over one
    deterministic Rust engine, with Python and R frontends that are
    bit-exact against each other for shared algorithms.
  - BBOB and CEC 2014/2017/2022 benchmark suites.
  - NSGA-II multi-objective optimization over ZDT/DTLZ/WFG.
  - IOH-format logging with ECDF/COCO export.
  - A statistical-comparison toolkit.
  - A structural-bias scanner.