---
title: "Guided verified weighting in WFC 2.0"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Guided verified weighting in WFC 2.0}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

`wf_autoweigh()` gives survey researchers a narrated path over the verified
engines. It no longer builds a target from raw runtime data. Prepare `design`,
import `target`, and declare `dims` first.

```{r guided, eval=FALSE}
guided <- wf_autoweigh(
  design,
  target,
  dims,
  method = "raking",
  trim = NULL,
  interactive = FALSE,
  lang = "en"
)

guided$weights
guided$diagnostics
guided$ledger
```

The ledger uses stable English programmatic keys while its `detail` text may be
localized. Artifacts align with ledger rows so a statistician or agent can
inspect each precheck, recommendation, and report.

Automatic mode chooses only from declared safe settings. It does not inspect
outcomes, change the target, widen bounds, or manufacture approval.

```{r views, eval=FALSE}
wf_report(guided$weights, audience = "decision", lang = "zh_CN")
wf_report(guided$weights, audience = "statistician", lang = "en")
```

For production, prefer `wf_plan_weights()` plus separate
`wf_approve_plan()` and `wf_execute_plan()` so human approval is bound to one
immutable plan.