Metadata-Version: 2.4
Name: failure-gallery
Version: 0.2.0
Summary: Validate and build a synthetic agent and robotics failure gallery with reproducible review records.
Author-email: AuraOne <opensource@auraone.ai>
License-Expression: MIT
Project-URL: Homepage, https://auraoneai.github.io/failure-gallery/
Project-URL: Documentation, https://github.com/auraoneai/failure-gallery#readme
Project-URL: Repository, https://github.com/auraoneai/failure-gallery.git
Project-URL: Issues, https://github.com/auraoneai/failure-gallery/issues
Project-URL: Changelog, https://github.com/auraoneai/failure-gallery/blob/main/CHANGELOG.md
Keywords: agent-evaluation,robotics-evaluation,failure-analysis,review-fixtures,evals,synthetic-data
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Software Development :: Testing
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Dynamic: license-file

# failure-gallery

Validate and build a searchable gallery of reproducible synthetic agent and robotics failure records.

`failure-gallery` is for evaluation educators, review-tool maintainers, and robotics or agent teams that need inspectable fixtures for failure-review workflows. Its differentiator is evidence-first synthetic case design: every record names the failure, domain, expected review label, expected finding, related tool, exact reproduction command, source JSON path, and synthetic-data boundary.

## Inspectable Output

- `validate`: checks case fields, requires at least 12 records, and requires both agent and robotics domains.
- `render`: writes one requested HTML file from the case records.
- `build`: deterministically writes both `site/index.html` and `docs/index.html` by default.
- `check`: fails when either generated deploy target differs from the canonical renderer output.

The case JSON under `cases/` is the review evidence. `src/failure_gallery/render.py` and packaged local CSS/JavaScript assets are the canonical generator inputs; generated HTML should not be edited by hand.

## Runtime Boundary

Validation and rendering read local JSON, CSS, and JavaScript files and make no network requests. The generated page is standalone, although its navigation links point to public AuraOne and GitHub pages when a browser follows them. No customer, private-lab, or production incident data is bundled.

## Install

The current `0.2.0` source workflow is installed from a clone:

```bash
python -m pip install -e .
```

PyPI currently provides the older `0.1.0` release. Use the source install above for the `build` and `check` workflow documented here.

## Quickstart

```bash
failure-gallery validate cases/
failure-gallery build cases/
failure-gallery check cases/
```

## Release Status

Registry status verified July 13, 2026:

- The checked-in package metadata is `0.2.0`.
- PyPI and the latest public repository tag remain at `0.1.0`.
- The `0.2.0` build/check and packaged-static-asset changes are source-only until a new release is published.

The project is alpha software. No incident-volume, customer, benchmark, or adoption claim is made.

## Limits

The records are synthetic tutorials with expected review outcomes. They are not real incidents, benchmark results, model comparisons, or proof that a related tool will detect every production failure.

## Next Action

Install the current source, run `validate`, `build`, and `check`, then choose one record and execute its `reproduce_command`; investigate any mismatch with the documented `expected_finding`.
