Your Metric Is Not Your State

In a recent article, Ken Walger explores the crucial distinction between raw metrics and the actual state of a system, using the analogy of winemaking. Just as a refractometer provides misleading sugar readings once fermentation introduces alcohol, software metrics like CPU utilization or HTTP status codes can become unreliable proxies when system conditions change. Walger argues that observability often fails because we treat metrics as absolute truths rather than evidence. He emphasizes that meaningful interpretation requires context, such as historical data, calibration baselines, and the lineage of the measurement. By separating raw observations from derived conclusions and maintaining the provenance of data, developers can avoid the trap of misinterpreting steady metrics as signs of health. Ultimately, the author warns that a metric is merely a piece of evidence, and treating it as the system's state without understanding its history leads to flawed operational decisions.