# Eval Bench

Run declared evaluation cases and retain their results.

Type: Evaluation toolkit

Availability: Public source package

Evidence: Source code · Sample files

## The work

Acceptance criteria need inspectable execution records before they become completion claims.

A deterministic evaluation runner producing result JSON, evidence files, a report, and proof-case definitions.

## Where it stands

Passing declared cases supports those checks, not broad model quality or production readiness.

## Take a look

Browse the public repository, documentation and examples.

- [Explore the repository](https://github.com/unit27research/unit27-eval-bench)
- [Inspect the public examples](https://github.com/unit27research/unit27-eval-bench/tree/main/examples)



[Portfolio project page](https://joshuabloodworth.com/projects/unit27-eval-bench.html)