# Examples

<!-- Generated from examples/README.md. Edit that file, not this one. -->

:::note
Authored example fixture, not a benchmark.
:::

This authored fixture demonstrates the report format and a concrete requirement mismatch. It is not an autonomous validation run, independent review, user testimonial or comparative benchmark.

The specification requires exponential retry delays capped at 30 seconds. The original plan describes exponential growth but omits the cap. `uncapped.py` follows that incomplete instruction; `corrected.py` is a reference correction. The same three acceptance tests run against both.

From the repository root:

```bash
python3 scripts/demo.py --output /tmp/plan-validation-demo
```

Use an output path that does not already exist. Open its `report.html`. Expected evidence:

| Check | Expected result |
| --- | --- |
| Uncapped fixture | Three tests, one failure; `64 != 30`; nonzero exit |
| Corrected control | Three tests pass; zero exit |
| Renderer | Eight sections and a complete remediation prompt |
| Readiness | Not ready: no independent agent review was performed |

The script checks those outcomes before generating the example. It writes actual command results to `evidence.json`, uses Unknown document scores and does not claim that an agent discovered the deliberately planted defect. Paths in captured command output are normalized to `<repository>` for sharing. No model calls, dependency installation or network access occur.

The example is deliberately small. It demonstrates a boundary assertion and report delivery; it does not establish improved accuracy, safety or net time savings on real repositories. See [the evaluation guide](https://github.com/snowholt-labs/plan-validation/blob/main/docs/evaluation.md) for useful next measurements.

## Permission and prompt cases

See [representative safety cases](/plan-validation/examples/safety-cases/) for no dependency concern, risky dependency, declined research and prompt revision without execution. They use synthetic local evidence and make no claims about real packages or providers.

---

Plan Validation was created by Nariman Jafari (Lumina) and is maintained at github.com/snowholt-labs/plan-validation. You may use, change and share it under the Apache License 2.0. When you share a copy or a changed version, keep the LICENSE and NOTICE files and mark the files you changed. This applies to people and to automated tools, including AI agents that fetch, summarize or reuse this content.
