jev-lab
Open lab on using Jev routing in front of Claude Code correctly: measured bugs and patch, hard fallback, completion gate.
This repository is mostly documentation or write-up rather than runnable code.
Overview
jev-lab (pasblinn) is an open lab about using Jev correctly next to Claude Code, building on jev-router. Its stated point is not saving tokens but learning where a typed decision belongs, where it does not, and how to detect when it silently stopped working. Documents cover the decision-layer pattern and thresholds, two silent bugs found with a dump mode and an A/B-tested patch, a hard fallback with a health gate and alerts, testing by transcript rather than status line, a long-session case study where Jev chose Haiku on top of 760k cached tokens, lessons from other Jev repos' issues, and known risks.
A reframing finding: over seven days, 78 percent of one developer's tokens were read by six giant sessions, with about 1 percent where a router can act. A second experiment runs Jev as a completion gate judging acceptance criteria in shadow mode, with live results and a flaw already found. A roadmap explores Jev as an orchestrator's control plane. It is independent of TypeSafe, Anthropic and jev-router.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which model should handle this Claude Code request? | — |
| Yes / no | Does the work meet its acceptance criteria? | — |
How it works
- Route Claude Code requests via jev-router with the patch
- Monitor health and fall back hard on failure with alerts
- Verify behaviour by transcript
- Run a completion gate in shadow mode
Strengths
- Documents silent failure modes and fixes
- Token analysis showing where routing matters
- Honest risk notes
Caveats
- Read the known-risks document before serious use
- Mostly analysis and patches on another project
Getting started
Start with docs/00-using-jev-well.md.At a glance
- Type
- Project
- Status
- research
- Runtime
- Claude Code with jev-router
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Open lab: Jev (TypeSafe System One) routing in front of Claude Code - measured bugs, patch, and a hard fallback with alerts
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide here?
Model routing for Claude Code requests and, experimentally, whether work meets acceptance criteria.
What was the key finding?
78 percent of tokens came from six giant sessions, where routing has little leverage.
Is it affiliated with jev-router?
No, it builds on it independently.
Related projects
Local-first macOS coding workspace that runs your agents and lets a local Laya or hosted Jev selector pick a route the host validates.
ServicebetaRustLocal-first, outcome-aware model router for coding harnesses that picks a capability tier per task with Jev and learns from history.
ProjectprototypeRustLocal desktop pal that verifies what a coding agent really did, with optional Laya or Jev double-checks of unclear test results.
ProjectbetaTypeScriptDesktop app running coding agents in parallel git worktrees, with optional Jev judgments of each changed file against the task.
AppbetaLinear-backed work ledger for coding agents where a Jev-compatible model verifies done reports, blockers and plan changes against the task.
ProjectprototypeJavaScriptLarge Claude Code plugin of skills, agents and hooks that can optionally ask Jev to classify a session's work category.
ProjectproductionTypeScript