Evaluation cadence and governance

Six months after shipping v2, a production incident forces emergency remediation. The SQL correctness judge has drifted from Kappa 0.72 to 0.48. The VP asks: "How did we not catch this?" The answer: nobody remembered who owns the judge calibration work. The monthly deep dive meeting got cancelled twice. Never rescheduled. Nobody checked. The tooling was there. The dashboards were there. The judge was there. Nobody looked. This is Lesson 6.5, covering evaluation cadence. The previous lessons built instrumentation, metrics, judges, experiments, and monitoring. A complete system. But it does not run itself. This pattern repeats at every organization. Something great gets built, shipped, and celebrated. Then three months later nobody remembers who is supposed to check the judge. The system decays quietly. Without structured ownership and recurring cadences, evaluation becomes a side project that gets deprioritized when deadlines loom. This lesson defines who owns evaluation activities, how often they occur, and what happens when evaluation work gets deferred. The result is an Evaluation Cadence Package: a three-tier framework with ownership assignments and a debt register that keeps evaluation from becoming invisible work.