AI Trust Experiment

A lightweight, minimal dependency Next.js implementation of the A/B AI Trust experiment.

Constraints Respected

• Next.js/React: Built securely using standard Next.js constraints.
• Short Codebase: codebase only within 4 files (layout.tsx, page.tsx, globals.css, package.json).
• Low Dependencies: No heavy frontend libraries like Tailwind or framer-motion. Just React and Next frameworks.

Features

• A/B Logic: Randomly assigns users(participants) with a recommendation as Condition A (Technical) or Condition B (Warm/Friendly).
• High-Resolution Latency Logging: Uses performance.now() to measure click latency to the nearest millisecond.
• Data Export: Logs all required parameters - participant_id, condition, decision, timestamp, latency_ms as structured JSON, downloaded immediately on decision, exactly as specified.
• AI Recommendation Ready: Includes a commented-out API block (e.g., Groq/OpenAI) for dynamically fetching generated recommendations based on either Condition A (Technical) or Condition B (Warm/Friendly); maintaining the prompt structure with a system prompt.

Condition Logic

• Condition A (Technical) - recommendation framed with precise, data-driven language
• Condition B (Warm/Friendly) - same recommendation framed in conversational, empathetic language

Assignment is done client-side on page load. Each participant sees only one condition.

Sample Output

{
  "participant_id": "f3a2c1d4-8e6b-4a2f-9c1e-7d5b3f2a1e8c",
  "condition": "B",
  "decision": "accepted",
  "timestamp": "2026-03-28T14:22:31.408Z",
  "latency_ms": 3421
}

Setup

1. Install dependencies:

   npm install

2. Start the development server:

   npm run dev

3. Open localhost

API Integration (Optional)

Check app/page.tsx for a commented-out standard fetch block utilizing Groq (https://api.groq.com/openai/v1/chat/completions) or any LLM compatible schema. You can use any APIKey endpoint to inject generative messaging by un-commenting the code.