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 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.
predmicror 1.1.3
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.
