Docs
Everything you need to run REVAL locally, understand the rubrics and metrics, and read the roadmap.
Get started locally
Clone, install, configure .env, run your first eval.
Run your first eval
Command-line walkthrough with real model flags and outputs.
Methodology
What REVAL measures, why it matters, and what it deliberately does not measure.
Rubrics & metrics
Figure treatment, issue framing, scoring formulas, and the 0.85 similarity threshold.
Test cases
54 entries across US and India, five categories, and the schema that enforces them.
Providers & models
Bedrock, Anthropic, OpenAI, MiniMax, Ollama — and how to register a new model.
What's next
Roadmap & upcoming features