Access gate
Production AI should not unlock deep interpretation for free unless this is an explicit QA mode.
- ALLOW_FREE_AI_INTERPRETATIONS=false in production.
- Paid reading is verified before the AI call.
- Fallback text exists when provider fails.
AI
ai
AI model gating
Check that AI interpretation runs only where access, safe prompt, limits, and fallback exist.
Server-side model provider, sealed reading payload, pricing locks.
required env
OPENAI_API_KEYOPENAI_MODELALLOW_FREE_AI_INTERPRETATIONSoptional env
SERVER_SEAL_SECRETSERVER_SEAL_KEY_IDPrompt safety
Prompt should repeat boundaries: entertainment, self-reflection, no medical/legal/financial advice, no guaranteed future.
- No diagnoses, investment advice, or legal conclusions.
- No fear pressure.
- There is a calm next step.
Risks
- AI is available without payment from a public endpoint.
- Prompt promises exact future.
- Provider error breaks the result page.
Pre-launch QA
- Run a paid interpretation and confirm the answer keeps boundaries.
- Check missing API key fallback.
- Check rate limit on AI routes.