Changes in version 1.0                         

Breaking changes

  - New primary API: bca_nonpar() replaces bcajack(), bcajack2(), and
    bcanon(). bca_par() replaces bcapar().

  - All computation functions now return a canonical "bcaboot" object
    with consistent structure: $limits (9x4 matrix), $stats (2x5
    matrix), $B_mean, $ustats. The old field names ($lims, $B.mean) are
    aliased via $.bcaboot for backward compatibility.

  - User-facing parameter conf.level (e.g., c(0.95, 0.90)) replaces
    alpha (e.g., c(0.025, 0.05)). Internally, alpha expansion is
    unchanged.

  - Descriptive parameter names: n_groups (was m), group_reps (was mr),
    kl_fraction (was pct), n_jack (was K), jack_groups (was J),
    truncation (was trun), conf_density (was cd), boot_data (was
    overloaded B).

  - bcanon() removed (was never on CRAN).

Deprecations

  - bcajack(), bcajack2(), bcapar(), and bcaplot() are deprecated with
    lifecycle warnings (once per session). They will continue to work
    through thin wrappers that translate to the new API.

Bug fixes

  - Issue #2: Fixed grouped jackknife (n_groups < n) using
    sapply(seq_len_m, sample.int, ...) which accidentally passed
    iteration values as the first positional arg, overriding the n=
    keyword and enabling replacement. Replaced with matrix(sample.int(n,
    n-r), nrow=m) for correct without-replacement partition. (Reported
    by admash)

  - Issue #7: Fixed dimension drop when x is a single-column matrix
    (vector input). x[-i, ] on an n x 1 matrix returned a vector instead
    of a matrix, causing type inconsistency between jackknife and
    bootstrap code paths. Added drop = FALSE to all matrix subsetting
    passed to func. (Reported by R180)

  - Issue #4 / #1: regression_accel now checks for underdetermined
    regression (nearby samples < ncol(Y)) and errors with an actionable
    message instead of silently returning NAs. Suggests increasing B,
    increasing kl_fraction, or switching to accel = "jackknife".
    (Reported by Tim Pollington, Thomas Covert)

  - PR #8: Fixed missing return() in K = 0 early-exit path of bcajack(),
    bcajack2(), and bcapar(), which caused execution to fall through to
    the K > 0 code. In the rewrite, all early exits use explicit
    return(new_bcaboot(...)). (Reported by Bettina Gruen)

New features

  - tidy() method: returns a tibble with one row per (confidence level,
    method) pair, following broom conventions (conf.level, method,
    estimate, conf.low, conf.high).

  - glance() method: returns a one-row tibble summarizing the bootstrap
    run (method, accel, theta, sdboot, z0, a, sdjack, B, boot_mean).

  - autoplot() method: ggplot2 visualization of BCa vs standard
    confidence intervals. Requires ggplot2 (in Suggests).

  - Improved print(): formatted summary with confidence limits table
    instead of raw list dump.

  - bca_nonpar() always computes the gbca diagnostic when using accel =
    "regression" (with tryCatch for robustness).

  - All user-facing messages use cli for consistent formatting.

Dependencies

  - Added cli, tibble, generics, lifecycle to Imports.
  - Added ggplot2 to Suggests (for autoplot()).

Internal

  - Shared helpers extracted to R/bca_core.R: expand_alpha(),
    bootstrap_resample(), jackknife_accel(), regression_accel(),
    compute_ustats().
  - new_bcaboot() canonical constructor guarantees consistent return
    structure across all computation functions.
  - Test suite expanded from 14 to 34 tests, including adapter tests
    proving numerical equivalence against pre-1.0 fixtures.