Jev
Projects and libraries using Jev for routing, verification and typed decisions.
Repository owner imageTypeSafe AI Benchmark Theater
Compare native Jev decisions with alternative models on shared synthetic application contracts.
CategoryEvaluation & benchmarks | UseSynthetic application-contract benchmarks | SourceGitHub |
Owneriammrduncan | Original projecthttps://github.com/iammrduncan/typesafe-ai-benchmark | |
Research evidence
packages/demos/lib/jev.ts compiles contract fields into native Choice/Noul questions, sends one System One request and validates the returned answer set and distributions. runtime.ts applies decoded decisions to the demonstration contract and distinguishes live requests from explicit fixture mode.
One benchmark application with seven workload scenes, not seven independent production applications. The generative compatibility API elsewhere in the repository is not counted as Jev execution.
The September 17-18 report uses synthetic repeated fixtures and different local/cloud measurement paths. Source and methodology were inspected, but raw measurements were not recomputed and no accuracy, latency, cost or superiority claim was independently verified.
documentation source 1 (retrieved 2026-09-18)code source 2 (retrieved 2026-09-18)code source 3 (retrieved 2026-09-18)documentation source 4 (retrieved 2026-09-18)