---
title: "Predictive Distribution and Calibration Uncertainty"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Predictive Distribution and Calibration Uncertainty}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

Posterior predictive diagnostics can retain uncertainty in entire outcome
distributions as well as in predictive scores.

```{r, eval=FALSE}
atlas <- create_predictive_distribution_atlas(
  fit,
  ndraws = 500,
  include_group_effects = TRUE,
  seed = 2026
)

plot_predictive_atlas_statistics(atlas)

quantiles <- predictive_quantile_envelope(atlas)
plot_predictive_quantile_envelope(quantiles)

scores <- prediction_score_uncertainty(
  fit,
  ndraws = 1000
)

prediction_score_uncertainty_table(scores)
plot_prediction_score_uncertainty(scores)
```

For binary models:

```{r, eval=FALSE}
calibration <- binary_calibration_uncertainty(
  fit,
  bins = 10,
  ndraws = 1000
)

binary_calibration_uncertainty_table(calibration)
plot_binary_calibration_uncertainty(calibration)
```

These summaries do not automatically establish calibration, predictive
adequacy, or model superiority.