February 2022 Summaries

A new SDK for a feature flagging and experimentation platform has been released in the Go programming language, characterized by its lightweight nature, speed, and lack of external dependencies. It can be used independently or integrated with the platform, allowing for feature flagging, variation assignment, and user targeting. Installation is straightforward, using the command "go get github.com/growthbook/growthbook-golang." While the SDK can function on its own for feature flag rule evaluation and variation assignment, its full potential is realized when combined with the GrowthBook API, enabling users to adjust JSON payload values directly from the UI. Further details and use cases can be explored in the documentation and on GitHub.

The text explores the complexities and potential pitfalls associated with choosing metrics for experimentation, particularly highlighting issues related to Goodhart's Law, which warns against measures becoming ineffective when they turn into goals. It illustrates how the misuse of metrics can lead to unintended consequences, such as gaming metrics to achieve desired outcomes without improving overall objectives, as seen in historical examples like the Cobra Effect and modern instances such as school rankings manipulation. The text also discusses the importance of distinguishing correlation from causation in metric-driven decision-making, using a case study from a freemium content site that mistakenly equated increased content use with higher revenue, only to find the correlation was non-causal. It emphasizes the necessity of careful metric selection, the risks of proxy metrics, and the need for comprehensive analysis to avoid negative outcomes and improve the effectiveness of experiments.