---
title: "Response-time and process-aware IRT"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Response-time and process-aware IRT}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

Eye-tracking, pupil, response-time, and sequence channels are **measurement channels**. Their association with latent response models does not automatically identify cognitive strategy, effort, diagnosis, or mental state.

```{r}
spec <- eyeprocess_joint_process_irt_spec(
  response_family = "2pl", time_model = "lognormal",
  process_channels = c("dwell", "pupil", "transitions"),
  missingness = "ignorable"
)
spec
```

`eyeprocess_process_irt_data_bundle()` preserves sparse person-item data and declared process channels. Descriptive functions summarize response-time structure, speed-accuracy association, item/person process profiles, missingness patterns, and alignment without turning process measures into latent-state labels.

When exact joint response/response-time estimation is requested, `fit_eyeprocess_lnirt()` delegates to LNIRT if available and otherwise returns a gated result with `fit = NULL`.

Primary package source: <https://cran.r-project.org/package=LNIRT>.