---
title: "Item seeding, accessibility review, and presentation fairness"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Item seeding, accessibility review, and presentation fairness}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

## Experimental item-parameter seeding

```{r}
seed <- fit_item_parameter_seed_model(
  calibrated_items,
  predictors = c("visual_density", "text_complexity", "word_count", "screen_luminance")
)

candidate_predictions <- predict_item_parameter_priors(seed, candidate_items)
audit_candidate_item_bank(seed, candidate_items)
plot(seed, candidate_data = candidate_items)
```

These predictions are screening priors/cold-start estimates only. They do not replace content review, accessibility/bias review, pilot testing, or IRT calibration.

## Presentation/accessibility sensitivity

```{r}
a <- audit_presentation_accessibility(person_process_data)
sim <- simulate_presentation_variants(a)
plot(a)
```

This audit must not be used to infer dyslexia, ADHD, neurodivergence, visual impairment, or another diagnosis. It identifies presentation patterns worth evaluating with calibrated alternative versions.

```{r}
compare_presentation_fairness(
  experiment_data,
  variant = "presentation_version",
  outcome = "accuracy"
)
```