Open research and tooling around JEV: structured decisions instead of free text.
JEV is a model that answers small structured questions (for example, which of five known actions fits a situation) instead of generating free text.
We build open tooling around that idea: finding where a decision point in a codebase might benefit, compiling typed decision questions, and researching parallel typed decisions and calibration.
| Repository | Summary | License |
|---|---|---|
jev-integration-evaluator |
Scan a codebase for decision points where JEV might help, offline and read-only, and test them with evidence | MIT |
TypeWright |
Declare a decision, compile typed Jev questions, measure them, and ship a JSON program | MIT |
layev |
Open research implementation of parallel typed decisions and calibration inspired by Kev and Laya | Apache-2.0 |
jraphyte |
TRACE-GC evidence-bound graph synthesis, GraphRAG, and first-page research validation | MIT |
layevis a research candidate, not a validated Jev replacement. Its CPU fixture tests pass; native Qwen/CUDA checks and Jev-relative quality are still open.jraphyteis a reference implementation with explicit deployment gates, not a qualified production service. Its demo uses fabricated observations and no provider.jev-integration-evaluatoris read-only and offline by default.TypeWrightis alpha: its no-key demo and core tests pass, but no live provider study has been completed and no measured accuracy gain is claimed.- Live provider results are not claimed here; each README states what was and was not exercised.
Contributing · Support · Security · Steward: Complete Tech LLC