Changes in version 0.2.0                        

JSS revision release. Comprehensive S3 class system overhaul, expanded
methods coverage, and reproducible replication package.

Breaking changes

  - bdeb_diagnose() now returns a bdeb_diagnostics S3 object. All output
    goes through print(), summary(), and plot() (type = "rhat" or
    "ess"). Direct list access still works (n_divergent, summary, ...).
  - bdeb_summary() is deprecated; use summary(fit) on a bdeb_fit object
    instead. The wrapper still works but emits a deprecation warning and
    will be removed in a future release.

New methods

  - bdeb_data: print(), summary(), plot().
  - bdeb_model: print(), summary(), plot().
  - bdeb_prior: print(), summary(), plot().
  - bdeb_prediction: print(), summary() (the latter returns a tidy time
    / lower / median / upper data frame).
  - bdeb_diagnostics: print(), summary(), plot().
  - bdeb_fit (additional lm-style methods): summary() (now the primary
    posterior-summary API, accepts pars and prob), confint() (posterior
    credible intervals), fitted() (posterior median/mean of
    \eqn{\hat{L}_i}), residuals() (observed minus fitted), nobs()
    (observation count), vcov() (posterior covariance of model
    parameters), and logLik() (log-pointwise predictive density, lppd).
    fitted(), residuals(), and logLik() are available for "individual"
    and "growth_repro" models only.
  - bdeb_derived() is now an S3 generic with a bdeb_fit method
    (bdeb_derived.bdeb_fit). Existing calls are unchanged; the dispatch
    enables future support for derived quantities on prior-only or
    simulated objects.

Bug fixes

  - bdeb_diagnose() no longer fails on real-data fits whose generated
    quantities (log_lik, L_rep) contain NaN draws from sporadic
    ODE-solver failures. The summarise_draws() quantile lambdas now pass
    na.rm = TRUE.
  - plot() on a bdeb_fit (type = "trace", "posterior", "pairs") now
    subsets draws to the requested parameters before delegating to
    bayesplot. Without this, NaN draws in unrelated generated quantities
    caused bayesplot::prepare_mcmc_array() to abort with "NAs not
    allowed in 'x'".

Stan

  - Increased ode_bdf_tol max_num_steps from 1e4 to 1e5 in all four Stan
    models to reduce CVode mxstep messages during warmup.

Vignettes

  - Conditional execution via requireNamespace("cmdstanr").
  - Fixed plot(fit, type = "pairs") returning NULL.

Replication material

  - Reorganised JSS replication material into a single zip with README,
    data/, outputs/, and lite/full execution modes.
  - Bundled curves.txt locally to remove external dependency.

                 Changes in version 0.1.4 (2026-04-21)                  

CRAN-review compliance release. Addresses reviewer feedback on the use
of T as an identifier, which can shadow R's built-in T symbol (= TRUE).

Breaking changes

  - arrhenius(): first argument renamed from T to temp. Positional calls
    are unaffected (arrhenius(298.15)). Code that passed the argument by
    name (arrhenius(T = 298.15)) must be updated to arrhenius(temp
    = 298.15).
  - bdeb_model(temperature = ...): the list field T was renamed to T_obs
    to match the Stan data naming and to remove the T shadow. Replace
    list(T = 298.15, T_ref = ..., T_A = ...) with list(T_obs = 298.15,
    T_ref = ..., T_A = ...).

CRAN compliance

  - bdeb_diagnose() and bdeb_ec50() now expose a verbose = TRUE
    argument. All user-facing output (diagnostic alerts and summary
    tables) is routed through cli / [message()] rather than direct
    print() calls, so it can be silenced with [suppressMessages()] or by
    passing verbose = FALSE. Return values are unchanged.

Documentation

  - Updated R/utils.R, R/data_prep.R, R/model_spec.R and
    man/arrhenius.Rd, man/bdeb_model.Rd,
    man/temperature_to_stan_data.Rd, man/build_stan_data_individual.Rd
    to reflect the renaming.
  - Case-study vignette updated to use arrhenius(temp = ...).
  - Tests updated for the new temperature$T_obs field.
  - \dontrun{} replaced with runnable examples for prior_species(),
    bdeb_tox() and bdeb_prior_predictive() (all execute in < 0.2 s
    against bundled datasets). bdeb_fit() retains \dontrun{} because it
    requires the external CmdStan toolchain and a single Stan
    compilation + MCMC run takes > 30 seconds; the Rd comment explains
    the reason.

Citation

  - Switched README badge and CITATION.cff from the version-specific DOI
    (10.5281/zenodo.19500753, v0.1.3) to the concept DOI
    (10.5281/zenodo.19443804), which always resolves to the latest
    archived version. The v0.1.4 version DOI is 10.5281/zenodo.19642839
    for anyone needing to cite this specific release.

                        Changes in version 0.1.3                        

New features and data release.

New features

  - prior_species(): species-specific priors from the AmP collection for
    E. fetida, E. andrei, F. candida, D. magna, and L. rubellus.
  - plot(fit, type = "prior_posterior"): prior vs. posterior density
    comparison plot.

New datasets

  - eisenia_neuhauser: real E. fetida growth data from Neuhauser,
    Hartenstein & Kaplan (1980), 37 group-mean measurements over 250
    days.
  - eisenia_cd: real Eisenia andrei cadmium toxicity data from Van
    Gestel et al. (1991), 5 concentration groups over 85 days.

Documentation

  - Updated DOI to Zenodo 10.5281/zenodo.19500753.

                        Changes in version 0.1.2                        

ODE solver upgrade and bug fix release.

Breaking changes

  - R-side ODE integration now uses deSolve::lsoda() (adaptive
    BDF/Adams) instead of fixed-step Euler. This matches the BDF solver
    used in Stan, ensuring numerical consistency between R-side
    simulation and Stan-side inference. The dt parameter now controls
    output resolution, not integration accuracy.
  - deSolve (>= 1.40) added to Imports.

Bug fixes

  - Fixed bdeb_ec50()$NEC returning a data frame instead of numeric
    draws — median(ec50$NEC) now works correctly.
  - Fixed build_stan_data_hierarchical() crashing when growth data
    starts at time = 0 (same fix as individual model in 0.1.1).
  - Fixed build_stan_data_debtox() crashing when growth or reproduction
    data starts at time = 0.
  - Fixed build_stan_data_growth_repro() t=0 handling producing NA index
    matches — replaced convoluted fix with consistent epsilon shift
    across growth and reproduction times.

Internal

  - Renamed internal helpers sim_deb_euler() / sim_debtox_euler() to
    sim_deb_lsoda() / sim_debtox_lsoda() to reflect solver.
  - Updated package documentation to describe LSODA solver instead of
    Euler.

                        Changes in version 0.1.1                        

Bug fix and hardening release.

Bug fixes

  - Fixed ode_bdf_tol crash when growth data starts at time = 0:
    replaced with 1e-3 epsilon since Stan requires t_obs > t0.
  - Fixed %||% operator not imported: replaced with explicit if
    (is.null(...)) in bdeb_predict().
  - Fixed silent rnorm() fallback for unknown prior families in
    bdeb_prior_predictive(): now throws informative error.
  - Fixed bdeb_predict(newdata = NULL) using t_obs instead of t_L for
    "growth_repro" models.
  - Fixed bdeb_predict() silently returning raw draws when L_hat is
    missing: now throws informative error.
  - Fixed DEBtox reproduction using cumulative R instead of interval
    delta_R = k_R * (R(t_end) - R(t_start)): now consistent with
    growth_repro model.
  - Fixed phi_R prior hardcoded in DEBtox Stan model: now uses
    prior_phi_R_mu/sd from R prior system.
  - Fixed deb_fluxes() computing p_G with redundant / E_G * E_G.
  - Fixed growth_rate formula: was v/3 * p_M/(kappa*E_G), now correct
    Kooijman Eq. 3.23: k_M * g / (3*(f+g)).

Safety improvements

  - "individual" and "growth_repro" models now hard-error (not warn)
    when data contains multiple individuals.
  - DEBtox auto-aggregates multi-individual concentration groups to
    means with explicit warning.
  - Removed survival argument from bdeb_data() (was accepted but not
    implemented — "feature mirage").
  - All prior constructors validate hyperparameters (sigma > 0, a > 0,
    rate > 0, nu > 1).
  - bdeb_data() validates f_food in [0, 1].
  - bdeb_fit() validates all sampling parameters (chains, iterations,
    adapt_delta, max_treedepth, threads).
  - arrhenius() validates T > 0, T_ref > 0, T_A >= 0.
  - bdeb_model() validates prior objects, observation objects, and
    temperature fields (must be positive finite scalars).
  - bdeb_fit() wraps Stan compilation and sampling in tryCatch with
    contextualised error messages.
  - repro_to_intervals() warns when dropping individuals with < 2
    observations.
  - validate_growth() warns when max length > 10 cm (possibly physical
    rather than structural length).
  - DEBtox reproduction time matching now uses strict match() instead of
    nearest-neighbour which.min().

New features

  - bdeb_loo(): LOO cross-validation via loo::loo() with endpoint
    argument for "growth_repro" models.
  - bdeb_prior_predictive(): R-side prior predictive simulation with
    print() and plot() methods.
  - bdeb_session_info(): reproducibility report (R, Stan, package
    versions, fit configuration, Stan model hash).
  - coef.bdeb_fit(): S3 method returning posterior medians or means.
  - deb_simulate(), debtox_simulate(): standalone DEB/DEBtox simulators,
    independent of Stan.
  - L_m added to bdeb_derived(): maximum structural length at f=1.
  - plot.bdeb_prediction(): S3 plot method for prediction objects.
  - Observation model switching now fully implemented in all 4 Stan
    models via integer flags (no recompilation needed).
  - Arrhenius temperature correction implemented in all 4 Stan models.
  - Within-chain parallelism via reduce_sum for hierarchical and DEBtox
    models.
  - threads_per_chain argument in bdeb_fit().
  - seed argument in bdeb_predict(), plot_dose_response(), and plot(fit,
    type = "trajectory").
  - dt and t_end arguments in plot_dose_response().
  - Reproducibility metadata (seed, versions, timestamp) stored in
    bdeb_fit object.
  - inst/CITATION with package and Kooijman (2010) entries.
  - CITATION.cff with Zenodo DOI.

Documentation

  - Structural vs physical length documented throughout (delta_M).
  - All derived quantity formulas reference Kooijman (2010) equations.
  - Prior calibration rationale documented with AmP ranges.
  - Feature status table in README with stable/beta/planned.
  - "Current limitations" section in README.
  - "Numerical layers" section in package docs (Stan exact vs R-side
    approximate).
  - Lifecycle annotations on all functions and models.

Tests

  - 901 tests across 16 files (up from 58).
  - New test categories: API contracts, scientific consistency,
    snapshot/regression, deep validation, end-to-end integration.

                        Changes in version 0.1.0                        

Initial release.

Models

  - Individual-level growth model (2-state DEB: reserve, structure).
  - Growth + reproduction model (3-state DEB: reserve, structure,
    reproduction buffer) with negative binomial observation model.
  - Hierarchical multi-individual growth model with non-centred
    parameterisation and partial pooling of assimilation rates.
  - DEBtox (TKTD) model for ecotoxicology with scaled internal damage,
    stress on assimilation, and analytical EC50/NEC computation.

Features

  - Declarative model specification via bdeb_model().
  - Prior specification functions for all standard DEB parameters:
    prior_lognormal(), prior_beta(), prior_halfnormal(), and others.
  - Sensible weakly informative default priors via prior_default(),
    calibrated against the AmP collection (Marques et al., 2018).
  - Observation model selection: obs_normal(), obs_lognormal(),
    obs_student_t(), obs_poisson(), obs_negbinom().
  - MCMC fitting via cmdstanr with bdeb_fit().
  - Convergence diagnostics: bdeb_diagnose() reports R-hat, ESS,
    divergences, and E-BFMI (Vehtari et al., 2021).
  - Posterior predictive checks with bdeb_ppc().
  - Derived quantity computation: bdeb_derived() for ultimate length,
    von Bertalanffy growth rate, and somatic maintenance rate constant.
  - Publication-quality plots: trace, posterior density, pairs,
    trajectory, and PPC overlays.
  - DEBtox helpers: bdeb_tox(), bdeb_ec50(), plot_dose_response().
  - Utility functions: arrhenius() temperature correction, deb_fluxes()
    energy flux calculator, repro_to_intervals() data converter.

Data

  - eisenia_growth: simulated Eisenia fetida growth (21 individuals, 12
    weeks).
  - folsomia_repro: simulated Folsomia candida reproduction test (5 Cd
    concentrations, 6 replicates).
  - debtox_growth: simulated growth under toxicant exposure (4
    concentrations, 10 individuals each, 6 weeks).

Vignettes

  - getting_started: overview of the package workflow.
  - case_study_eisenia_folsomia: full workflow with Eisenia growth
    (individual + hierarchical) and Folsomia DEBtox analysis.