Design an A/B testing analysis framework for our mobile fitness app

Our growth team at Repwise runs 4-6 experiments a month on our fitness app, but every analysis is a bespoke notebook and we have definitely shipped at least one winner that was actually noise. We need a statistician-engineer hybrid to build the framework that stops that: standardized metric definitions, correct handling of ratio metrics and pre-experiment bias, sequential testing or CUPED where it fits, and honest power calculations before we launch anything. Deliverables: a Python analysis library with tests, a written decision playbook (when to call it, when to extend, when to declare inconclusive), and templates that turn a raw experiment readout into a decision memo. You will validate the framework by re-analyzing our last six experiments - we genuinely want to know which of our past "wins" survive scrutiny. We are looking for someone who has done experimentation at scale and has opinions. A short paragraph in your proposal about a common A/B testing mistake you see teams make will get you to the top of our list.