OpenJevPROJECT STATUS / SEPTEMBER 23, 2026
Built today.
Explored next.
An explicit distinction between implemented capabilities and future directions.
0.1 The decision layer
Noul, Choice and Score. Strict validation, probability normalization, an OpenAI-compatible provider, offline mocks and Goal Loop integration.
0.2 The adaptive runtime
Optional Laya / Laya-MLX backends, System One-compatible HTTP adapter, deterministic pre-model gates, confidence-aware routing, calibration helpers and benchmark sweeps.
0.3–0.4 Evidence before expansion
First-party, consented Goal Loop evaluation cases and wider multi-provider comparisons. Measure false completion, missed triggers and unnecessary escalation on held-out workloads.
0.5+ A trained model, if justified
A first-party open decision model depends on workload evidence, appropriate data rights and reproducible evaluation. It is not available in this release.
1.0 A portable service
Stable HTTP schema, Python and TypeScript SDKs, trace/replay and calibration dashboards are roadmap goals, not current shipping features.