---
title: "Reproducible statistical reports with DCC"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Reproducible statistical reports with DCC}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

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

`dcc_report_model()` is the single reporting source for staff, statistical,
and machine outputs. A statistical bundle does not recalculate cleaning
decisions: it writes the model's complete findings, audit, reconciliation,
before/after missingness and distributions, types, scoring, mapping,
parameters, hashes, and provenance.

```{r workflow, eval=FALSE}
out_dir <- file.path(tempdir(), "dcc-run")
run <- dcc_run(data_path, plan = plan_path, output_dir = out_dir,
               mode = "execute")
model <- dcc_report_model(run$result, run)
dcc_report_statistical(model, file.path(out_dir, "statistical"),
                       table_format = "parquet")
```

Use CSV when a plain-text exchange format is required. CSV files are UTF-8
and RFC 4180-compatible. Parquet retains canonical column types. In both
cases `artifact-manifest.json` records a SHA-256 digest for every artifact
except the manifest itself.

The distribution exports are complete rather than sampled and can therefore
contain sensitive values. Store the statistical directory under the same
access controls as the cleaned data and audit log. Staff reports remain
redacted by default, while PDF is optional and is not part of the base report
contract.