raapi is a personal rating engine for agents.

Use raapi.evaluate via POST /v1/evaluate when you need to decide whether to recommend a known item.
Use raapi.recommend via POST /internal/benchmark/recommend for private alpha movie/series shortlist generation.

Key outputs:
- global_score: general quality 0-100
- personal_fit: fit to user taste 0-100
- predicted_user_rating: predicted 1-10 rating
- decision.recommend: boolean
- confidence: 0-1

For internal alpha recommendations, send X-Internal-Token and a taste string. Do not expose benchmark source names to end users.