Apple's Evaluations framework (new in Xcode 27) provides APIs for building robust evaluations of AI-powered Swift app features. The session covers two advanced topics: generating and validating synthetic evaluation data using the makeSamples and SampleGenerator APIs, and evaluating agentic tool-calling workflows using TrajectoryExpectations. Synthetic data generation supports custom session providers, sampling strategies (random or sliding window), and per-sample validation closures. Tool evaluations verify not just model output but the sequence and arguments of tool calls, using matchers like naturalLanguage, contains, and oneOf. Both synthetic data generation and tool call evaluations can be combined into a single evaluation suite, and results are visualized in Xcode 27's new Evaluations Report.