NEWS

predmicror 1.3.2 (2026-06-19)

Improve predmicror_assistant() with a deterministic model registry, wrapper-first code generation, static code validation, and return_trace diagnostics.
Make the assistant fall back to registry-based output when Ollama is unavailable instead of failing immediately.
Bundle inst/shiny/predmicror-assistant/app.R and extend predmicror_assistant_app() with model, root, host, port, and browser arguments plus a fallback app.
Add assistant tests for registry metadata, deterministic output, and Shiny app bundling.
Add data-aware assistant support for data frames and uploaded .csv, .tsv, .xls, and .xlsx files, including automatic profiling, column detection, and data-specific wrapper code.
Improve the assistant Shiny app with a card-based layout, separate Answer/Code/Data/Trace views, local Ollama model selection, and manual task/column overrides.
Add initial dynamic modelling support with dynamic_profile(), RK4-based predict_dynamic_growth(), predict_dynamic_inactivation(), and finite-difference dynamic_sensitivity().
Add dynamic fitting wrappers fit_dynamic_growth() and fit_dynamic_inactivation() with predmicror_dynamic_fit methods and diagnostics.

Polish the GitHub README to reflect the current fitting, diagnostics, and model-comparison API.
Add explicit package website and issue tracker links to the README.
Make applied inactivation and cardinal-model vignettes more robust by using explicit fitted/residual column access.
Group applied workflow articles in the pkgdown articles index.

Added an applied vignette for microbial inactivation models using fit_inactivation(), predmicror_augment(), fit_metrics(), and compare_models().
Added an applied vignette for cardinal parameter models using fit_cardinal(), diagnostics helpers, and model comparison tools.
Expanded examples showing safer post-fitting workflows and prediction over new data grids.

Register default S3 methods for predmicror_augment() and fit_metrics() in roxygen documentation.
Add the pkgdown site URL to DESCRIPTION.
Add the new fitting and diagnostic topics to the pkgdown reference index.
Ignore temporary phase overlay folders created while applying local hotfixes.

Add predmicror_augment() to extract original data, fitted values, residuals, model name, and model type from predmicror_fit objects.
Add as.data.frame.predmicror_fit() as a lightweight base-R shortcut for predmicror_augment().
Add fit_metrics() to calculate residual diagnostics and information criteria for fitted models, including SSE, RMSE, MAE, bias, residual standard error, R2, adjusted R2, log-likelihood, AIC, BIC, and convergence status.
Add compare_models() to combine diagnostics across multiple fitted models and sort by AIC, BIC, RMSE, or MAE.
Add tests for diagnostic extraction, model metrics, and model comparison.
Add a model-comparison vignette showing how to compare alternative fitted predictive microbiology models.

Declare shiny as a suggested package for assistant functions.
Import utils::tail() for assistant history formatting.
Keep R CMD check free of errors and warnings, apart from environment-specific timestamp notes.

Add fit_growth(), fit_inactivation(), and fit_cardinal() wrappers around gslnls::gsl_nls().
Add predmicror_models() to list wrapper-supported models and required starting parameters.
Add the predmicror_fit class with print(), summary(), coef(), fitted(), residuals(), predict(), plot(), vcov(), logLik(), AIC(), and BIC() methods.
Add input validation for data columns and starting values in the fitting wrappers.
Update README content, package metadata, .Rbuildignore, and CI workflow templates.
Complete the WeibullMM() example and fix the HuangNLM() example label.
Configure testthat edition 3.

Fix Rosso full model parameter order and Baranyi reduced formulation.
Add numeric stability guards with log1p(), expm1(), and bounded square roots across growth and cardinal models.
Expose the Richards shape parameter and align examples with log10 inactivation scales.
Correct dataset documentation and row counts.
Add testthat coverage for growth, cardinal, and inactivation models.