VIDRAFT released Darwin-27B-ZTC-v2, an Apache-2.0 open-weights model that occupies the #1 position on the System One Mosaic Benchmark decision-engine leaderboard among 102 contenders, achieving a Borda score of 89.58 and a task average of 66.46. The architecture implements a typed-decision interface where routing nodes query Choice-type questions with a finite set of transition labels, policy nodes ask Noul-style binary statements, and scoring nodes evaluate ordinal scales; the model returns calibrated probabilities directly, eliminating the need for text parsing. This design, rooted in Zero-Token Calibration (ZTC), enables micro-decisions to be executed with minimal inference cost, making it economically viable for high-frequency agent-graph routing compared with prompting large language models. By confining non-decision logic to application code and delegating judgment to the model, developers can streamline agent pipelines while maintaining precise control over thresholds. The release provides a fully open alternative to proprietary typed-decision systems, targeting the same agent-graph use case without requiring proprietary APIs.