NEWS.md
In rblimp: Integration of 'Blimp' Software into R

rblimp 1.2.0

New install_blimp() and uninstall_blimp() functions let rblimp download and manage the Blimp engine directly, without going through the system installer. Installs to a user-writable directory (~/.blimp/ on macOS and Linux, %LOCALAPPDATA%/Blimp/ on Windows), overridable via the R_BLIMP_HOME environment variable.
When rblimp() / rblimp_source() is run in an interactive session and Blimp is not installed, the user is now offered the option to install it via install_blimp() on the spot.
detect_blimp() gained a prompt = argument; defaults to TRUE for the new interactive install offer.
update_blimp() now routes based on install type: for managed installs it checks the manifest version first and re-downloads only when a newer version is available; system installs still launch the Blimp Updater application.
Update version checks now use a JSON manifest endpoint (https://updates.blimpstats.com/v1/blimp_base.json) instead of parsing the plain-text changelog. Manifest URL is overridable via the R_BLIMP_MANIFEST_URL environment variable.
New R_BLIMP_NO_UPDATE_CHECK environment variable disables automatic update checks (alternative to options(check_blimp_update = FALSE)).
New R_BLIMP_PREVENT_INSTALL environment variable blocks install_blimp() for locked-down environments.

Added an as.array() method for blimp_obj that returns an [iterations, chains, parameters] 3-D array compatible with bayesplot. Supports a warmup = c("exclude", "include", "only") argument for including burn-in draws; when warmup = "include" the returned array carries an "n_warmup" attribute so it pipes directly into bayesplot::mcmc_trace(arr, n_warmup = attr(arr, "n_warmup")).
Added S3 methods registering blimp_obj with the posterior package's as_draws(), as_draws_array(), and as_draws_df() generics, unlocking the posterior / bayesplot summary and plotting ecosystems. posterior is a soft dependency (in Suggests).
Trace plots now display rank-normalized split-Rhat (Vehtari et al., 2021) instead of the basic PSRF, and fixed facet labels when plotting a subset of parameters.