These results can include a selected value alongside probability, confidence, or option/rubric probabilities. The release also introduces experimental ReAnchor, which fits decision thresholds, score cuts, and choice weights against a program-level metric.
The release is also a transition point for DSPy language-model integrations: legacy custom-LM integrations and OpenAI-style messages= calls remain available with deprecation warnings in 3.4, while DSPy identifies 3.5 as the migration deadline. Jev is explicitly a decision backend rather than a general-purpose text generator, with several unsupported experimental paths.