Changes in version 1.0.0                        

MetaLab data now lives in versioned, citable releases on Redivis
(https://stanford.redivis.com/datasets/81tq-8dp5ge6b9), and metalabr
reads them by default. Every data-access call announces which release it
used.

Data access

  - get_metalab_data() now reads released data from Redivis by default,
    with a version argument ("current" by default, or a release name
    like "2023.1" to pin). The legacy live-Google-Sheets path is still
    available by passing dataset metadata (get_metalab_data(metadata));
    it is intended for curators building releases.
  - get_metalab_metadata() likewise reads the released dataset registry
    by default (with sheet_status/data_as_of provenance columns);
    passing a datasets.yaml path/URL retains the legacy behavior.
  - get_current_metalab_data() now downloads the current release served
    by the MetaLab site instead of a snapshot committed to GitHub
    in 2023, and returns the effect-size data frame — like every other
    data-access function — with the release announced in a message and
    recorded in the metalab_release attribute. (metalabr 0.x instead
    loaded objects into the global environment, which CRAN policy
    disallows; use get_metalab_metadata() for what was the dataset_info
    object.)
  - Correlation imputation no longer sets a hardcoded RNG seed: the seed
    is a documented argument (get_metalab_data(imputation_seed = 111),
    the historical default, reproducing released data exactly), and the
    session's RNG state is saved and restored around the imputation.
  - New get_metalab_versions() lists releases and their Redivis
    versions.
  - New introductory vignette (vignette("metalabr")) covering
    released-data access and version pinning.
  - All network access fails gracefully with a message (returning NULL)
    rather than an error.

Visualizations

  - metalab_scatter_plot() was unusable outside the retired Shiny app
    (it referenced reactive objects); it now has a working
    data-in/ggplot-out interface consistent with the other plot
    functions.
  - metalab_violin_plot() supports every effect-size type (previously it
    errored for anything but Hedges' g) and handles multiple moderators.
  - metalab_forest_plot() fits its model once instead of four times, and
    its moderator handling works outside the app.
  - metalab_funnel_plot() works with moderators outside the app (it
    previously referenced free variables that only existed in the app
    environment).
  - New metalab_funnel_test(): an Egger-style asymmetry test adapted to
    the multilevel model (√vᵢ as a fixed-effect moderator). The legacy
    app's regtest() call always errored (invisibly) on multilevel
    models.
  - Plot colors no longer require the GitHub-only langcog package.

Removed / deprecated

  - viz_app() is deprecated: the interactive applications are superseded
    by the in-browser visualizations at https://metalab.stanford.edu.
    The internal Shiny app code (visualization, power analysis,
    validation) and the abandoned Google-Drive revision machinery were
    removed.

Fixed

  - The package now declares and imports its dependencies correctly —
    library(metalabr) works without attaching dplyr first (fixes #6).
  - dplyr::add_rownames() (defunct) and other retired tidyverse idioms
    were replaced; the pipeline runs on current tidyverse releases.
  - Invalid License field replaced with MIT + file LICENSE.

Compatibility notes

  - Effect-size computation is characterization-tested against the
    released database: the 1.0.0 pipeline reproduces the legacy
    pipeline's output exactly (all effect-size branches, the hardcoded
    special-case studies, seed-111 correlation imputation, and derived
    columns).
  - In released data on Redivis, the Abstract rule learning moderator
    column rule.type is named rule_type (Redivis does not allow dots in
    column names). Data frames loaded via get_current_metalab_data()
    keep the legacy name.