Changes in version 0.3.1                        

  - All errors and warnings are now classed conditions: every condition
    carries a domain subclass (e.g. metasurvey_error_step,
    metasurvey_error_recipe, metasurvey_error_backend_unavailable) plus
    the base class metasurvey_error / metasurvey_warning, so they can be
    handled with tryCatch() without matching message text. See the new
    ?metasurvey_conditions topic for the full list. Messages are
    unchanged, but code that matched on condition classes such as
    rlang_error may need review.
  - Removed the last remote-credential residue from core: the unused
    internal get_user() helper and with it any reading of the
    metasurvey.api_key option. Credentials and API access belong to
    provider packages (e.g. metasurvey.explorer.backend); core reaches
    remote backends only through the injected
    options(metasurvey.backend_provider = ...) hook, and selecting the
    "api" backend without a provider now signals a
    metasurvey_error_backend_unavailable error that can be handled with
    tryCatch().
  - New ?metasurvey.core-options topic documenting every metasurvey.*
    option with its default and effect.
  - Examples: get_recipe(), load_panel_survey() and extract_surveys()
    now ship runnable offline examples; the remaining \dontrun{} blocks
    explain why they cannot run offline.

                        Changes in version 0.3.0                        

Initial release of the local processing engine of the metasurvey
ecosystem, extracted from the original monolith (archived as
metasurvey-legacy) following the modularization suggested during the
rOpenSci review. The version number is aligned with the metasurvey
meta-package.

  - Core engine: Survey, RotativePanelSurvey and PoolSurvey R6 classes,
    the lazy step pipeline (step_compute(), step_recode(),
    step_rename(), step_remove(), step_join(), step_filter(),
    bake_steps()), recipes with automatic documentation and validation,
    and estimation workflows built on the survey package (means, totals,
    ratios and domain estimation with CV-based quality labels).
  - Rotating panels with implantation and follow-up waves, pooled
    estimation, and bootstrap replicate weights via add_replicate().
  - Local (JSON-backed) recipe and workflow registries. The remote
    ("api") backend is provided through an injectable hook
    (options(metasurvey.backend_provider = ...)), registered by the
    companion package metasurvey.explorer.backend; core itself has no
    network, Shiny, or STATA dependencies.
  - 13 bilingual vignettes (English and Spanish), including an ECH case
    study that runs offline on a bundled microdata sample and
    reproducible examples for seven international household surveys.
  - rOpenSci-ready packaging: README, CONTRIBUTING, Code of Conduct,
    GPL-3 license files, codemeta.json, pkgdown site, and CI (R CMD
    check, pkgcheck, coverage, pre-commit).