Ab Test Analyzer

Interpret the following A/B test results and return a JSON object with the exact shape specified.

## Test Details
- Metric: {{metric_name}}
- Test duration: {{test_duration_days}} days
- Desired confidence level: {{confidence_target}}%

## Control Group
- Visitors: {{control_visitors}}
- Conversions: {{control_conversions}}
- Conversion rate: {{control_rate}}%

## Variant Group
- Visitors: {{variant_visitors}}
- Conversions: {{variant_conversions}}
- Conversion rate: {{variant_rate}}%

## Pre-calculated Metrics
- Relative uplift: {{relative_uplift}}%

## Instructions
Use approximate z-test reasoning to estimate confidence. Based on the data, provide:
1. A recommendation: "ship" if the variant wins at the desired confidence level, "dont-ship" if it clearly loses, or "run-longer" if there's not enough data.
2. An estimated confidence percentage (0-100) using z-test approximation.
3. Plain-English reasoning (2-4 sentences).
4. 2-3 practical caveats.

Return ONLY valid JSON in this exact shape (no markdown, no code blocks):
{
  "result": {
    "recommendation": "ship" | "dont-ship" | "run-longer",
    "confidence": <number 0-100>,
    "controlRate": <decimal e.g. 0.032>,
    "variantRate": <decimal e.g. 0.041>,
    "relativeUplift": <percentage number e.g. 28.1>,
    "reasoning": "<plain English explanation>",
    "caveats": ["<caveat 1>", "<caveat 2>", "<caveat 3>"]
  }
}