---
title: "Optional Engines and Cross-Platform Portability"
description: "Reviewable optional-engine availability without making optional engines mandatory."
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Optional Engines and Cross-Platform Portability}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
options(gp3ml.reproducible_examples = TRUE)
library(gp3ml)
```

## Capability table

Optional engines remain optional. Missing packages are reported explicitly
rather than changing the scientific task or silently selecting another model.

```{r capabilities}
capabilities <- gp3ml_engine_capabilities(
  check_keras_backend = FALSE
)
capabilities
plot(capabilities)
```

`glm` and `lm` are always available. `ranger`, `xgboost`, `nnet`, and `keras3`
are exercised by a dedicated GitHub Actions matrix. Keras backend readiness is
queried only when explicitly requested.

```{r base-assertions}
assert_gp3ml_engine_available("glm")
assert_gp3ml_engine_available("lm")
```

Engine availability is not a model-selection rule. Candidate selection remains
explicit, metric-declared, direction-declared, and reviewable.