Your question is Conflicting Primary and Guardrail Results. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Explain how you would interpret results when an A/B test shows conflicting directional movement between primary and guardrail metrics.
Discuss how you would distinguish statistical significance from practical significance, assess whether the guardrail movement is an acceptable trade-off, and decide whether to ship, iterate, or stop. State the assumptions and analysis checks you would require before making the decision, including the pre-registered MDE, confidence intervals, multiple comparisons, peeking, and possible sample ratio mismatch.