Changes in version 0.1.0 (2026-10-15)                  

First release. The software behind the companion paper on Lepanthes
eltoroensis (Tremblay, submitted to Oikos, 2026).

Behaviour settled in the pre-release review (2026-10-05)

  - expected_recruits() pairs each unit's recruits with its own
    reproduction by row name, whatever the row order of X.
  - With missing = "drop", lag_ceiling() leaves the scored periods whose
    lag window reaches before the record out of the observed series too,
    so that the observed and the expected concentration are measured on
    the same periods (n_dropped says how many).
  - A flat reproductive record gives an exceedance of NA (with flat =
    TRUE), not Inf; rain_gini() cannot return a negative value from
    rounding.
  - lag_ceiling_by() applies min_periods to scored periods, and reports
    the per-unit warnings once with the units they concern instead of
    silencing them.
  - convolve_lag() refuses a record too short for the window and
    validates t_R and lag0; lag weights are validated everywhere a
    profile is applied.
  - recruit_triage() refuses a long data frame with a misnamed column
    instead of reading it as a wide matrix, and refuses species =
    together with s/r.
  - The bootstrap and posterior probabilities use the finite replicates
    of their own statistic; the print method no longer fails when none
    is finite.
  - host_matrices() and series_from() warn when missing recruits are
    read as zero.
  - The Stan diagnostics cover ER, the expected series that is reported,
    and the period effects. CmdStan is looked for after the data are
    validated.
  - Every user-facing check lives in checks.R; the all-zero-reproduction
    diagnosis is a classed condition (recruitlag_zero_lagged).
  - The introductory vignette is recruitlag (formerly baby-steps). Lag
    bins are labelled from the first lagged census (0-6, 6-12,
    ..., 24-30 months) in all prose.

The check

  - lag_ceiling(): the concentration ceiling that any delay of a
    reproductive record implies, the searched ceiling over four families
    of lag weights (pure delays, windows, geometric decay, Dirichlet
    draws; lag_kernels()), the observed concentration of the recruit
    series, and their ratio, the exceedance.
  - lag_ceiling_by(): the same for every replicate unit; rain_indices():
    the ceiling under every candidate reproductive column, reported as
    one table and never as a menu.
  - ceiling_calibration(): what negative-binomial counts about a mean
    that obeys the ceiling can produce, which shows what the pooled
    comparison cannot refute.
  - host_lag_test() and host_matrices(): the test at the level at which
    the lag hypothesis is asserted, expected recruitment on each unit
    being that unit's own lagged reproduction scaled to its own total,
    with the negative-binomial size moment-matched about each kernel's
    own expectations. Both refuse non-count recruit series.

Uncertainty on the exceedance

  - Three routes returning one object class, lag_ceiling_draws:
    lag_ceiling_boot() (cluster bootstrap over units, BCa or percentile
    interval, one-sided bootstrap p), lag_ceiling_bayesboot() (Bayesian
    bootstrap, equal-tailed or HDI, P(exceedance > 1)), and
    lag_ceiling_stan() (negative-binomial model of the recruits with
    unit and period effects, the exceedance of the expected series;
    needs cmdstanr and CmdStan). Each print method says what it
    estimates and what its interval means.
  - ceiling_draws_table() and rain_forest(): one table and one forest
    plot for several records and routes.

Is a first sighting a recruit?

  - recruit_triage() and rain_sightings(): the state at first sighting,
    the runs of unseen periods between sightings, and the expected
    number of first sightings that are returns of plants present but
    unseen, calibrated on the published Caladenia survival and
    resighting rates (caladenia_dormancy).

Census conventions

  - rain_intervals(): the census schedule against its median interval,
    with the fraction of a structure that a visit can see for a given
    persistence.
  - period is an integer from 1 with no gaps; a missed census is a row
    of NA, not an absent row; NA reproduction is read as zero with a
    warning.

Quantities and figures

  - rain_gini(), rain_cv(), concentration(), phi_moment(),
    convolve_lag(), expected_recruits(), series_from().
  - rain_series(), rain_strip(), rain_strips(), rain_profiles(),
    rain_expected(), rain_units(), and plot() methods for
    ceiling_calibration and host_lag_test.

Data and vignettes

  - Bundled: lepanthes_census, lepanthes_hosts, lepanthes_monthly,
    lepanthes_profile, regimes, regimes_truth, caladenia_dormancy.
  - Vignettes: recruitlag (introduction), invented-regimes, lepanthes,
    mathematics, intervals.
  - Every user-facing message says what is wrong, in which column, and
    what to do.