Deriving evaluation signals from available ground truth

Welcome to lesson 3.3. The regression suite is built, the system is instrumented, and traces are available. Now that data needs to be turned into signals: measurable indicators that inform decisions. This is where evaluation stops being data collection and starts being decision evidence. Here's what happens next. All the instrumentation from Week 2 is in place. Every stage emits data. Retrieval scores. SQL strings. Execution logs. Narrative text. Token counts. Latencies. The dashboard opens and there are hundreds of numbers. The PM walks over and asks: "Is v1 ready to ship?" Freeze. Not because the data is missing, but because nobody has decided what to measure. That's the gap this lesson fills. It defines evaluation signals by type and role, so when someone asks if it's ready, there's an answer.