---
title: "Posterior Exploration and Publication Graphics"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Posterior Exploration and Publication Graphics}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

`gp3bayes` separates numerical posterior summaries from graphics. The same
posterior draw matrix can therefore be inspected, tabulated, and plotted
without changing the fitted model or its contract.

## Backend-independent posterior tables

```{r}
library(gp3bayes)

draws <- cbind(
  intercept = seq(-1, 1, length.out = 500),
  condition = seq(-0.5, 0.7, length.out = 500)
)

posterior_interval_table(draws)
posterior_probability_table(draws, rope = c(-0.1, 0.1))
posterior_correlation_table(draws)
```

## Publication graphics

```{r, eval=requireNamespace("ggplot2", quietly=TRUE) && requireNamespace("bayesplot", quietly=TRUE)}
plot_posterior_intervals(draws)
plot_posterior_areas(draws)
```

The plotting functions return ordinary plotting objects. They do not alter
posterior draws, set decision thresholds, or turn interval exclusion into an
automatic substantive conclusion.

## Fitted-model extraction

For an approved fitted model, the post-fit API standardises extraction through
the `posterior` package:

```{r, eval=FALSE}
draw_array <- extract_posterior_draws(fit, regex = "^b_", format = "array")
draw_df <- extract_posterior_draws(fit, regex = "^b_", format = "df")

mcmc_diagnostic_table(fit)
sampler_diagnostic_table(fit)
quality <- summarise_mcmc_quality(fit)

plot_rank_diagnostics(fit)
plot_autocorrelation(fit)
plot_mcmc_quality(quality)
plot_sampler_diagnostics(fit)
```

Diagnostic flags request inspection. Their absence is not encoded as proof of
model adequacy.