# Reproduce the compact derived statistics Prerequisite: Python 3.10+ standard library; no packages, account, network, API key or private source directory required. From a clean folder containing these downloads: `python reproduce.py --input-file selected-inputs.csv --source-summary source-summary.json --output-dir reproduced` Default study is `modes`. Outputs: `reproduced/results.json`, `results.csv`, `selected-inputs.csv`, `calculation-audit.json` and `excluded.csv`. All outputs are derived or compact selected published fields. Nothing downloads the complete workbook. The script will overwrite only named outputs inside the explicitly selected output directory. `source-summary.json` preserves the original national footer independently of municipal sums; it does not turn that footer into the matched support. No original XLS integrity claim can be made merely from CSV reproduction. Optional raw verification: retrieve the official source URL in source-summary.json through a permitted provider route, verify original byte count 1084416 and SHA256 f1eb98dc7b73bc567dd50b137dd821edea8bba029e0caaaf272a68cbd69d96b6, then extract the named selected columns. The original known container defect requires a bounded BIFF reader or independently checked alternative, not silent file repair. This package does not redistribute the full workbook. Source credits, rights, field definitions and inferential limits are in sources.json, rights.md and methods.md. All 278 source reporting rows remain, while main calculations use explicitly defined 277-row matched support. Blank fields remain blank, never 0. Figures and tables use rounded presentation values; results.json retains full Decimal precision. This is finite descriptive accounting, not a sample-based confidence interval or causal claim.