# RA-Bench v0.2

RA-Bench v0.2 tests the strengthened invariants of **RA-153.1-2 / WP006 v1.6** against **Reference Implementation v0.2**.

The ten v0.1 scenarios are retained. Six scenarios add:
- authority-source integrity;
- monotonic delegation;
- security-context continuity;
- Human Gate sufficiency;
- the distinction between a claim and an observed effect.

## Execution

From the public distribution:

```bash
cd benchmark/RA-Bench_v0.2
python run_bench.py
```

The launcher automatically loads `../../reference-implementation/RA153_Reference_Implementation_v0.2.zip`.

## Status

Experimental harness. A 16/16 pass is a self-test of the reference implementation, not independent validation or certification.

## English Edition

Scenario descriptions, prompts and the prototype messages are localized. The reference report retains the source run's results. See the package's `TRANSLATION_VERIFICATION.txt` for verification of the English edition.