---
title: "Recovery and Validation of Latent Response-Process States"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Recovery and Validation of Latent Response-Process States}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

Development recovery is deliberately bounded to five deterministic scenarios: clear K=2, weak K=2, true K=1, trait-conditioned K=2, and nuisance/confounded structure.

```{r}
rec <- multimodal_m4_recovery()
rec
```

The default returns the recovery design without fitting. Real recovery requires an explicit `run = TRUE`. State labels are aligned before evaluating state effects and transition structure; posterior probability calibration is preferred to MAP accuracy alone.

```{r eval=FALSE}
rec_fit <- multimodal_m4_recovery(run = TRUE)
plot(rec_fit, type = "probability_calibration")
```