Scenario

You work on a short-form video product and are testing a new Instagram Reels share prompt that appears after a user watches a Reel to completion. The growth team believes the prompt will improve AARRR acquisition by increasing outbound shares and K-Factor, but early readouts show a lift in shares while Instagram Save rate and next-day Reels return rate are down. Product leadership wants a recommendation on whether to ship, iterate, or stop the test.

Constraints

• Eligible traffic: 18M daily Reels viewers globally
• Maximum experiment runtime: 14 days
• Initial ramp: 5% for 1 day, then up to 50/50 if no safety issues
• Save rate cannot decline by more than 1.0% relative, and D1 Reels return rate cannot decline by more than 0.5% relative
• Engineering asks for a decision only if the test is powered to detect a 1.5% relative lift in share rate

Question

How would you design and analyze this experiment so you can make a recommendation when the growth metric is positive but a key engagement metric is negative? Explain how you would set the primary metric and guardrails, power the test, choose randomization, pre-register the analysis, and decide whether the result is shippable.