# CRE source-to-output acceptance kit

**Everything in this example is synthetic.** The properties, entities, people, records and figures are fictional training inputs authored by NextAutomation. No client, prospect, county or registry data is included. No model was run. The seeded mistakes illustrate review, not measured AI accuracy.

Start with [the expected answer](EXPECTED-ANSWER.md), then inspect the [inputs](inputs/properties.csv), [source register](sources/source-register.json), [evidence ledger](expected/evidence-ledger.csv), [corrections](expected/review-events.jsonl) and [acceptance checklist](ACCEPTANCE.md).

The exercise follows three properties through mandate fit, documented entity roles and a limited underwriting screen. It deliberately keeps unknown ownership/contact facts and missing financial statements unresolved. Passing this fixture is not a title opinion, investment approval, compliance certification or a performance guarantee.

## Reproduce offline

Python 3.10+ with the standard library only. No account, API, model, network connection or installation required. Extract the complete ZIP into a new directory, open a terminal there, and run:

```sh
python3 verify.py
python3 -m unittest discover -s tests -v
```

CSV amounts are USD. Periods and role definitions matter. The JSON expected output preserves null for unknowns. A shared registered agent never becomes a common-owner join. The contact channel stays unknown; this kit sends nothing.

All links and scripts resolve relative to this directory. The verifier reads supplied inputs without modifying them. JSON/CSV are open for inspection in any editor or spreadsheet. No external fonts, logos, client screenshots or dependencies are bundled.